a16z Crypto Fund V — $2.2B — May 5

a16z crypto closed its fifth fund at $2.2 billion. Total raised across five funds: $9.8 billion. Fund 4 was $4.5B in 2022. Fund 5 is half that — sized intentionally for a quieter market. Eddy Lazzarin promoted to general partner (CTO → GP).

Named target sectors: stablecoins, onchain lending, perpetual futures, prediction markets, tokenized assets, creator platforms, AI-agent infrastructure, decentralized infrastructure.

Their thesis in one line: "turning new infrastructure into products people use every day."

Same day: The largest crypto VC fund launch and the largest crypto exchange layoff happened on the same morning.

Halborn April Hack Report — published May 5

$577M lost in two attacks. We wrote about both.

Drift — $285M

KelpDAO — $292M

Halborn conclusion: the attack surface moved from code to people and configuration. Code audits would not have caught either attack.

Other New:

• Aztec launched alpha privacy stack on Ethereum — first feature-complete privacy from execution to settlement

• Animoca Brands tokenized its own equity on Solana via Republic with BitGo custody. also launched $10M program for developers building on its AI agent platform Minds

• Angular Ventures published on BYOT (bring your own tokens) and AI-native open source as the emerging enterprise architecture. enterprises rejecting thin wrappers and proprietary model lock-in

• Salesforce Ventures AI pricing report: 87% of sellers plan to change pricing in 12-18 months. only 16% think AI capabilities are why customers choose them

  

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

For Founders

if you build in any of a16z’s named sectors — they just told you where $2.2B is going. stablecoins, onchain lending, perps, prediction markets, tokenized assets, creators, AI-agent infra, decentralized infra. read their 8 publications from last week. those are the thesis documents for the fund.

if you have governance keys in production — review your multisig thresholds, timelock durations, verifier configurations, and signing policies. if a signer cannot explain what a transaction does, they should not be signing it.

if you are pricing an AI product — salesforce ventures surveyed 300+ executives. 87% plan to change their model. hybrid pricing (flat base + consumption) is the current standard. per-seat is dying. the market has not converged yet, which means you have room to experiment but also no safe default to copy.

Today Signal Question: if you had to pick one signal to plan your next 6 months around, which one would it be?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the data. Now go build something.

— Byblos.digital

Subscribe now

• Newsletters scanned - 60+

• Top themes - AI stack value migration, stablecoin B2B integration, BTC perp/spot divergence

• Notable raises - MARA / Long Ridge $1.5B, MoonPay / Sodot $100M, Belo $14M, Fence $20M, Blockworks $192M val

• Macro flags - Fed chair succession in play, KelpDAO hack still unwinding

Signal of the day

The unusual move today was the timing of the publishing. Chamath, Madrona, Nascent, and a16z's Charts of the Week each published a version of the "AI value is moving up the stack, away from chips and infra, toward services" thesis inside a 90-minute window this morning.

The crypto signal is sharper than the AI one though. MARA Holdings, one of the biggest publicly-traded Bitcoin miners, announced on April 30 that it's buying Long Ridge Energy for $1.5B specifically to expand into AI data centers. That single deal IS the merged trade nobody was pitching: crypto infrastructure pivoting hard into AI compute. The thesis convergence on the AI side has already bled into actual M&A on the crypto side.

Stories

01 — Four funds, same thesis, same morning

Tag [AI / VENTURE]

Source: Chamath, Madrona, Nascent, a16z Charts of the Week all today

What: Four independent VC publications today argued AI value is migrating from chips and infrastructure up to the application and services layer. a16z framed it as “first semis, then software.” Chamath ran the AI stack as six layers and argued the value slides up. Madrona tied the thesis to public earnings. Nascent narrowed it to Latin American professional services.

Numbers: Four publications, ~90-minute window, zero coordination.

Byblos signal: Same-day same-thesis convergence is rare and it’s almost always a leading indicator that allocations have already shifted internally at these funds. “Infrastructure for AI” is now in a queue of forty decks.

Why it matters: The bid for tools is dead. The bid for outcome-priced services is alive.

02 — A Bitcoin miner just bought $1.5B of AI infrastructure

Tag [CRYPTO / AI / M&A]

Source: CoinDesk (April 30), aggregated in Castle Island Round-up today

What: MARA Holdings announced the $1.5B acquisition of Long Ridge Energy & Power to fuel its expansion into AI data centers. A pure-play Bitcoin miner is rebranding itself as compute infrastructure.

Numbers: $1.5B deal, announced April 30. Long Ridge is a power-and-energy asset, the unsexy bottleneck for AI compute build-out.

Byblos signal: The pivot is the story. Bitcoin miners already have power contracts, cooling infrastructure, real estate, and grid relationships - exactly what AI hyperscalers need. The miners that survived the last cycle are sitting on the most valuable infrastructure in tech. MARA is the first to move at scale.

Action: Founders building at the crypto/AI infra intersection pitch to crypto-native funds before generalists catch up.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

03 — Stablecoins crossed into real B2B

Tag [CRYPTO / PAYMENTS]

Source: Theta Capital (today), supported by week’s deal flow per Castle Island

What: Theta Capital published today arguing stablecoins have moved past speculation into B2B settlement infrastructure. The week’s raises support the thesis: Belo raised $14M from Tether to expand stablecoin payments in LatAm (April 28), Fence raised $20M from Galaxy to overhaul the back-end of the $6T credit market (April 28), Meta started paying creators in stablecoins via Stripe.

Numbers: Belo $14M, Fence $20M, Squads $18M - all stablecoin payments infra, all this week.

Byblos signal: The data is loud. When Tether is leading a Latin American consumer wallet round and Galaxy is leading a $6T credit market back-end, stablecoins aren’t a thesis anymore - they’re a category.

Action If you’re payments-adjacent, Galaxy and Tether are now the most active stablecoin investors. Outbound this week.

05 — Galaxy is gaming out the next Fed chair

Tag [MACRO / RATES]

Source: Galaxy Weekly Research Brief (today)

What: Galaxy framed Fed chair succession as the dominant variable for crypto positioning into H2 2026, alongside notes on Tether’s growing role as a global liquidity provider, DeepMind, and Aave.

Byblos signal: Tether’s role keeps showing up in serious research from serious funds. Galaxy is the second mention this week. The story is shifting from “is Tether risky” to “how big is Tether’s monetary footprint.”

Why it matters: Crypto founders should expect VC term sheets to get more conservative if rates expectations drift higher.

Action: Build a 24-month runway plan. Funds that price for a hawkish chair will reward it.

For web3 founders, the deck that lands in May is the one that says "we are the application layer that turns crypto rails into AI services revenue." The MARA deal is the first big proof point that the merged trade is real.

Today Signal Question

Crypto miners are the most undervalued AI infrastructure plays in public markets. What’s your take on this?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the news. Now go build something.

— Byblos.digital

Subscribe now

current sentiment by type:

• crypto-native: cautious/neutral

• tradfi-adjacent: selectively bullish (tokenization, stablecoin infra)

• mega funds: deploying aggressively into AI, selectively into crypto

• operational: deploying into strong teams regardless of sector

1/ Accelerating: Institutional Tokenization

• invesco integrated with superstate’s tokenization stack ($967M fund). first global asset manager to use third-party rails instead of building

• SAFO (spiko + amundi) hit $400M AUM in 3 weeks. outpaced blackrock BUIDL at launch

• CARI network: 5 US banks launched tokenized deposits on zksync. led by former US comptroller

• DTCC: tokenized treasuries in production. SEC authorized

• state street: launching tokenized fund servicing from luxembourg by EOY

• ondo + broadridge: proxy voting for 250+ tokenized stocks and ETFs

• total RWA on-chain: $55.7B (excl. stablecoins). from $5B in 2022

• SEC innovation exemption “on the cusp.” 4 of 5 token categories not securities

status: accelerating toward consensus. 6 institutions in production this month.

2/ New Signal: KPI-Gated Token Launches

megaETH MEGA token launches today (10:00 AM UTC)

• TGE gated by on-chain milestones, not calendar: 10 apps needed 100K+ transactions each over 30 days

• triggered april 23 when all 10 apps cleared threshold simultaneously

• 53.3% of supply unlocks only as growth KPIs are hit. not time-based vesting

• coinbase pre-listing confirmed. dragonfly led. vitalik backed

• premarket on hyperliquid: $1.5-2B implied valuation (down from $6B+ in october)

status: this model succeeds, expect copycats. if it fails, stays an experiment. watching launch day data.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Consensus Forming: Agent Infrastructure

7 independent sources published on agent infrastructure this week (lemniscap, late checkout, a16z, QED, activant, GTM newsletter, 1inch)

• threshold for our consensus signal is 5. this crossed it april 25 and hasn't cooled

• lemniscap today: "agentic control plane" — service discovery, authorization, payments across agents

• a16z this week: tested whether AI agents can execute DeFi exploits. 50% success rate with transaction data, drops when sandboxed

• salesforce shipped "headless 360" — exposing all products as API-first for agent consumption. hubspot, workday, servicenow following

• prediction from late checkout: all major SaaS platforms go headless within 18 months

status: consensus. the thesis is priced in.

Today Signal Question:

Agent infrastructure hit consensus. Per-agent pricing is accelerating. Institutional tokenization is in production not pilots. Composability flipped from strength to liability. And a token launched today by hitting on-chain milestones instead of following a calendar.

Which of these should change what you're building or how you're pitching - this week, not next quarter?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the news. Now go build something.

— Byblos.digital

Subscribe now

Heaviest Tuesday inbox this month. HTX Ventures published the CLARITY Act coalition data. Activant Capital dropped enterprise AI survey data that changes how founders should think about pricing. Quantum Frontier covered NVIDIA's quantum lab launch. a16z published on HCM disruption.

1/ The CLARITY Act Either Passes in May or Dies

[REGULATORY]

120+ crypto firms, Coinbase, Ripple, Circle, Kraken, a16z, Paradigm, sent a joint letter demanding a markup of the CLARITY Act. Senator Moreno: passes by end of May or shelved indefinitely, possibly until 2030. Galaxy gives it 50/50. Polymarket odds dropped from 82% in February to 47%. The sticking point is stablecoin yield - banks say it'll drain deposits, the White House says banning it only helps lending by 0.02% while costing consumers $800M. The bill passed the House 294-134 in July 2025 and has been stuck in the Senate since.

Why it matters: Framework or void.

2/ Anthropic Withheld Its Best Model

[AI]

Anthropic had a model called Mythos that it chose not to release due to cybersecurity risks. Instead it shipped a neutered version as Claude Opus 4.7. First major case of a frontier lab withholding its best model from the public entirely. AI is splitting into three tiers: restricted frontier models for labs and governments, commercial models with guardrails for enterprise, and open-weight models for everyone else. If your product depends on the best available model, you're now in a tier system where access is a competitive variable, not a given.

Why it matters: Best models going closed.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Enterprise Buyers Are Pricing AI Agents Against Payroll

[PRICING]

Activant Capital surveyed 430 accounts receivable leaders. 90% are comfortable with autonomous AI agents. 96% prefer per-agent pricing over per-seat licensing. 48% of $1B+ companies still rely on spreadsheets alongside specialist tools. Enterprise buyers aren't comparing AI products to Salesforce - they're comparing them to the salary of the person they replace. Per-agent pricing against payroll shifts your TAM from the $300B software market to the $35T global payroll market. Same day, a16z published "Workday's Last Workday?" calling HCM the last major enterprise category without an AI-native challenger.

Why it matters: Priced against payroll, not software.

4/ Stablecoins Quietly Became Yield-Bearing Dollar Accounts

[STABLECOINS]

Global stablecoin market cap hit $319B. Treasury projects $3.7T by 2030. The real shift: stablecoins are evolving from trading tools into yield-bearing dollar accounts competing directly with bank deposits. Issuers pass Treasury yields to holders - effectively savings accounts without a bank. This is why the CLARITY Act's stablecoin yield provision is the most contested point. Banks see it as existential. McKinsey's data (99% of volume isn't payments) supports the reframing: stablecoins are financial infrastructure, and yield is the killer feature.

Why it matters: Yield-bearing dollar accounts are banks' problem.

Today Signal Question

120 companies gave the Senate a deadline. Anthropic withheld its best model. Enterprise buyers are pricing AI against payroll. And stablecoins became yield-bearing accounts competing with banks.

If the CLARITY Act dies in May, which of these trends survives without a framework, and which ones stall?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the news. Now go build something.

— Byblos.digital

Subscribe now

1/ McKinsey Just Killed the Stablecoin Payments Narrative

[STABLECOINS]

McKinsey partnered with Artemis and published what the industry was hoping nobody would dig into. Of $35T in stablecoin volume in 2025, only $390B, about 1%, was actual payments. The rest: trading, internal transfers, arbitrage, automated activity. The Federal Reserve found the same thing: 0.7% payments, 48.8% trading.

The "$33T stablecoin volume vs $25.5T credit card transactions" comparison is apples to oranges. We're comparing a settlement layer to a payment network.

But $390B in actual payments isn't nothing. It doubled from 2024. B2B dominates at $226B (up 733% YoY). Card spending hit $4.5B (up 673%). Asia is 60% of volume.

The Breakdown's framing: stablecoins aren't replacing Visa. They're replacing prime brokerage. For founders: if your deck says "replacing Visa", update it before McKinsey's data lands in your VC's inbox.

Why it matters: Settlement infrastructure, not payment network.

2/ The GPU Is Splitting in Two

[INFRA]

Chris Zeoli published "The Inference Unbundling" and it explains a $40B week. LLM inference has two workloads: prefill (processes your prompt, needs compute throughput) and decode (generates tokens one at a time, needs memory bandwidth).

Nvidia bought Groq for $20B to get bandwidth-optimized decode chips. OpenAI committed $20B to Cerebras for similar reasons. The single GPU doing everything era is ending. The future is routed inference stacks: different chips for different parts of the same request.

For founders: inference costs are about to restructure. Specialized silicon = cheaper tokens. The founders who understand this architecture make better infra decisions.

Why it matters: Specialized chips. Cheaper inference.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Composability Was the Contagion Channel

[DEFI]

Arca's Kelp aftermath analysis reframes DeFi's core selling point. Composability, everything talks to everything, was the mechanism that turned $292M into $13B.

The chain:

fake rsETH minted → deposited into Aave as collateral → $196M drained → Aave TVL collapses 38% ($10B) → borrowing costs spike 4x → pools hit 100% utilization → nine protocols freeze.

Every step happened because protocols trusted each other's outputs without independent verification.

Compare to 2008: CDO interconnectedness spread Lehman's collapse. Nobody called interconnectedness a feature after that. Aave v4 introduces compartmentalized lending. DeFi United coordinated without a bailout, genuinely mature. But the lesson: design for isolation first, composability second.

Why it matters: Composability is a contagion channel.

4/ Perps Are Eating Every Asset Class

[DERIVATIVES]

Perpetual swaps hit $92.9T in 2025 volume. Decentralized venues grew 346%. This week: Hyperliquid listed an S&P 500 perp. Polymarket and Kalshi announced perps expansion. Phoenix launched a Solana perp DEX with instant settlement and zero gas.

The pattern: perps started with crypto pairs, then equities, commodities, indices, now RWA. The perpetual swap is becoming the default derivative wrapper for everything.

Hyperliquid's HIP-4 shift is the tell: they outsourced 94% of listings to deployers, now they're pulling high-volume canonical markets (gold, S&P, crude) back in-house. Keep the blue chips, outsource the long tail.

For founders: synthetic exposure to any asset is becoming cheap and permissionless.

Why it matters: $92.9T. Perps are the default wrapper.

Today Signal Question

McKinsey says 99% of stablecoin volume isn't payments. The GPU is splitting into specialized chips. DeFi's composability turned out to be its biggest vulnerability. And perps are quietly wrapping every asset class.

Which of these changes your roadmap this week?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

1/ Model Companies Are Eating the App Layer

SpaceX announced Tuesday that it secured the right to acquire Cursor for $60 billion. The deal gives Cursor a $10B “collaboration fee” paid over time, and SpaceX gets the option to buy the company outright after its IPO this summer. Cursor halted a $2B funding round to take the deal. TechCrunch reported that SpaceX preempted the raise deliberately — they didn’t want Cursor raising at $50B and becoming harder to acquire.

Cursor was valued at $2.5B in January 2025. $9B by May. $29.3B after its Series D in November. Now $60B. That’s 24x in fifteen months.

But here’s the part that matters for founders: Cursor doesn’t have its own model. It sells access to Claude and GPT. The entire $60B valuation is distribution — millions of developers who open it every day. SpaceX bought an audience.

The reason SpaceX needs Cursor is embarrassing and Bloomberg didn’t sugarcoat it: xAI’s own engineers were using Anthropic’s Claude instead of Grok for coding. Musk spent billions building an AI company and his own team preferred the competitor. Two senior Cursor engineers left to join xAI last month, reporting directly to Musk. When you can’t build the product, you buy the company and poach the team simultaneously.

The same week, OpenAI shipped GPT-5.5, six weeks after GPT-5.4. The model advantage window is compressing from months to weeks. Anthropic shipped Claude Opus 4.7 the week before. Google is iterating Gemini constantly.

And then Chamath dropped his annual letter and put numbers on it. His argument: 60-80% of equity value in tech companies lives in terminal value — estimated earnings beyond the forecastable period. If AI can replicate your core product in weeks, that terminal value is zero. Duolingo fell 67.5% from its mid-2025 peak. Adobe fell 24.6%. Private equity bought over 1,900 software companies for $440B between 2015-2025. Software now makes up 25% of the entire private credit market. If AI makes those products replaceable, the cash flows disappear and the loans default.

Chamath’s exact framing: “Private credit is the canary in the coal mine.”

Newcomer’s analysis of the Cursor deal frames the right question: if the most successful AI application of the cycle ends up as a subsidiary of a model company, what does that mean for every other AI application startup?

If your moat is “we use the best model” — that advantage expires in six weeks. If your moat is “we have a better UX”, model companies will ship that as a feature. The only durable advantages left are distribution (Cursor’s $60B answer), workflow integration so deep that switching is painful, and proprietary data that models can’t replicate.

2/ Your 3-Person Team Can Now Outship Their 10-Person Team

The Cursor deal is actually two stories. One is about model companies eating apps. The other is about what these tools do to founder economics.

Start with the data from this week alone:

• Intercom doubled engineering velocity in 9 months using Claude Code. 400+ person company. Measured output, not vibes. Lenny's Newsletter published the case study.

• NVIDIA has 10,000 employees on GPT-5.5 Codex. Their engineering leads said debugging cycles that took days now close in hours. They called it "mind-blowing."

• OpenAI reported 85% of its own employees use Codex weekly — not just engineers. Finance, marketing, data science, everyone.

• Bank of New York tested GPT-5.5 alongside Anthropic's models. Their CIO called the hallucination resistance "a step change."

UpRound and Brinc published analysis this week showing AI now writes the majority of production code. NFX went further and declared "functional AGI is here." Whether you buy the AGI claim or not, the productivity data is getting hard to argue with.

Here's what this means if you're a founder with a five-person team and $800K in the bank.

Your competitor just hired three developers who use Codex or Claude Code. They're shipping at twice your velocity with 60% of your headcount. And when they raise their next round, they'll show VCs a shipping cadence that looks like a team three times their size.

This is a competitive weapon. The founder who picks the right dev stack ships faster, burns less, and raises at better terms. The founder who doesn't is paying ten people to do what three people with better tools accomplish in the same sprint.

Cursor is worth $60B not because of its model, it doesn't have one, but because millions of developers open it every day. Distribution to productive developers is worth more than the model itself.

For founders: audit your dev stack this week. The difference between your team on basic tools versus your team on Claude Code or Codex is the difference between raising your Series A and running out of runway.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ The Founders Who Survived This Week All Did the Same Boring Thing

This week gave us a controlled experiment. Multiple protocols got stress-tested at the same time. The difference was architecture decisions made before anyone needed them.

The ones that broke:

Kelp DAO lost $292M on April 18, the biggest DeFi exploit of 2026. The attacker poisoned two RPC nodes, DDoS'd the clean ones to force failover, and the bridge released 116,500 rsETH to an attacker-controlled wallet. Kelp's emergency pause kicked in 46 minutes later. Too late. LayerZero confirmed on April 20 that North Korea's Lazarus Group was behind it — the same unit that drained $285M from Drift on April 1 using a completely different method (social engineering — they posed as a quant trading firm for six months). That's $575M from one state actor in 18 days.

Volo Protocol on Sui was exploited for $3.5M this week, days after Kelp. Their team froze $500K and blocked a bridge attempt of 19.6 WBTC. Better response than Kelp. But still exploited.

The aftermath: Aave's TVL dropped from $26.4B to ~$20B in 48 hours. Nine protocols froze rsETH markets. Jefferies warned this week that banks may pause their blockchain initiatives.

The ones that survived:

• Fluid had dynamic withdrawal rate limits built from day one. Zero losses.

• Spark had no rsETH exposure for ETH lenders. Isolated by design. Zero losses.

• Lido's stETH was completely unaffected. Isolated architecture.

• Circle's USDC didn't depeg. Cross-chain settlement kept working. The stablecoin plumbing passed its stress test.

Every survivor made the unglamorous choice early: rate limits, exposure isolation, architectural separation.

0xResearch called it a "risk reckoning" in their newsletter this week — and the language matters. It signals a narrative shift from "growth at all costs" to "sustainability or die."

For founders: your incident response time IS your security. Kelp's 46-minute pause was an eternity, and $292M was already gone. Your security architecture decisions in month three are the ones that save your company in year two. Build the rate limits. Isolate the exposure. Do the boring thing now. It's the only thing that matters when the stress test arrives.

4/ The Fundraising Bar Mid-Round

If you're raising right now, the game changed while you were writing your deck.

Blockchain Capital is raising $700M across two new funds — we confirmed it this week. Their LPs: university endowments, sovereign wealth funds, US pension plans. Their portfolio: Coinbase, Kraken, Circle, Tether, Polymarket. They manage $2B+ and have a $6B+ portfolio.

That's the top. Now look at the middle and bottom.

Messari data shows average crypto deal size is up nearly 50% in 30 days — but total April funding dropped to $466M, down from $3B in March. That's an 85% drop. Capital is concentrating in fewer hands backing fewer companies.

The pattern: established firms with institutional LPs are raising aggressively. Newer and smaller funds are struggling. The money is flowing to infrastructure plays that bridge TradFi and crypto. Blockchain Capital raising $700M while smaller funds scramble tells you exactly where the market is heading.

And then Chamath's letter landed on every VC's desk.

His thesis reframes how every investor evaluates software companies. If AI can replicate your product in weeks, your terminal value is zero. If your terminal value is zero, 60-80% of your equity value disappears. The opening question in every pitch meeting just changed.

Here's what your pitch deck needs now that it didn't need three months ago:

The "LLM can't kill this" slide. Why an AI agent can't replicate your product in six weeks. Chamath's letter is on every investor's mind.

Your AI tooling strategy. Which tools your team uses and how it affects velocity. Intercom 2x'd output. What's your number?

Distribution in one sentence. Cursor is worth $60B because of distribution. What's yours?

Security architecture. Not "we got audited." What did you build?

Compliance plan. The SEC just published a token taxonomy saying 4 of 5 categories aren't securities. An innovation exemption for on-chain trading is imminent. Tether has frozen $4.4B across 340+ agencies. The regulatory landscape is shifting fast and founders who aren't ready will get left behind.

Today Signal Question

The hottest AI app got absorbed by a model company. Terminal value is collapsing. A 3-person team with the right tools outships a 10-person team without them. The protocols that survived the biggest exploit of 2026 all made the same boring decisions two years ago. And the biggest crypto VC just raised $700M while total deal volume dropped 85%.

What are you building that survives all of that?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

53 newsletters from 35 VCs. Sentiment: split — AI side is euphoric, crypto side is cautious. GPT-5.5 dropped yesterday and dominated the conversation.

• a16z deployed twice — invested in Petual (AI audit platform) and Glif in the same day. Deployment pace accelerating despite “monitoring the situation” last week

• $10B options expiry tomorrow — BTC + ETH combined. Traders watching closely

1/ OpenAI Shipped a New Model in 6 Weeks

[AI]

OpenAI released GPT-5.5 yesterday, six weeks after GPT-5.4. The API went live today. The numbers: 900 million weekly active ChatGPT users. Over 50 million paying subscribers. 4 million active Codex users. 9 million paying business users. 85% of OpenAI employees use Codex weekly across engineering, finance, marketing, and data science. NVIDIA has already been testing it internally for weeks — 10,000+ NVIDIA employees are using GPT-5.5-powered Codex and calling the results "mind-blowing." Bank of New York is testing it alongside Anthropic models and says the hallucination resistance is "a step change."

OpenAI claims GPT-5.5 outperforms Claude Opus 4.5 on coding benchmarks, 82.7% vs 69.4% on Terminal-Bench 2.0. It scored nearly double Anthropic's model on FrontierMath Tier 4 (39.6% vs 22.9%). These are OpenAI's own benchmarks so take them directionally. But the real story isn't the benchmarks — it's the cycle time. Six weeks between major releases. Anthropic shipped Claude Opus 4.7 last week. The model advantage window is compressing from months to weeks.

For founders building on AI: any product differentiation based on "we use the best model" is now a feature with a half-life measured in days. The only durable advantage is what you build around the model — distribution, workflow integration, proprietary data. The model itself is a commodity that gets replaced every six weeks.

Why it matters: Your moat isn't the model.

2/ The SEC Is About to Let You Trade Tokenized Securities On-Chain

[REGULATORY]

SEC Chair Atkins said on April 21 that the agency is "on the cusp" of releasing an innovation exemption for trading tokenized securities on-chain. The SEC also published a crypto token taxonomy that distinguishes five categories of digital assets: four of which are explicitly NOT securities. The innovation exemption will let companies facilitate on-chain trading of tokenized securities within defined guardrails, without full compliance with existing rules that were written for paper-based markets in the 1930s.

The SEC and CFTC signed a memorandum to coordinate on product definitions, clearing, collateral, and reporting — meaning both regulators are aligned on making this work. This is the regulatory unlock the RWA sector has been waiting for. DTCC already went to production on tokenized US Treasuries earlier this month with SEC authorization. Now the exemption framework opens the door for broader tokenized securities: equities, ETFs, potentially anything with a CUSIP.

For founders building in tokenization, RWAs, or on-chain financial infrastructure: the window just opened. The exemption is speech-level guidance, not final rule, so it's still early. But the direction is clear and the regulatory coordination is real. If you've been building tokenization infrastructure waiting for regulatory clarity, this is it arriving.

Why it matters: Tokenized securities trading on-chain is regulator-approved.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Regulators Are Suing Prediction Markets for Being Casinos

[REGULATION]

Polymarket and Kalshi are both eyeing perpetual futures markets - a direct move into the territory currently owned by Hyperliquid, dYdX, and centralized exchanges like Coinbase. The expansion was flagged by Lark Davis this week and covered across multiple VC newsletters.

The timing is chaotic. The same week these platforms announce perps expansion, New York's Attorney General sued Coinbase and Gemini over their prediction market offerings, calling them illegal gambling. SEC Chair Atkins told CNBC on Monday that the SEC, DOJ, and CFTC are all examining prediction market manipulation — specifically around trades that appear to front-run announcements. Interactive Brokers analysis from Emerging Moats suggests traditional finance is watching prediction market growth closely, which means the institutional money is interested but the regulatory risk is escalating simultaneously.

The Q1 data we've been tracking tells the same split story: $1.7B raised, but 85% of US activity is sports betting. Kalshi raised $1B. Polymarket closed $600M. a16z led $2.5M into Kairos and published two deep-dives in a month. The capital is flowing but the identity crisis isn't resolved. Now adding perps into the mix complicates it further — prediction markets becoming derivatives platforms while regulators can't agree on whether they're financial infrastructure or gambling.

Why it matters: Expanding product or expanding lawsuits.

4/ AI Is Writing Most of the Code Now.

[DEV TOOLS]

UpRound and Brinc published analysis this week showing AI now writes the majority of production code. NFX went further — declaring "functional AGI is here." Whether you buy the AGI claim or not, the underlying data is hard to argue: 85% of OpenAI staff use Codex weekly. NVIDIA has 10,000 employees on GPT-5.5 Codex and says debugging cycles that took days now close in hours. Intercom doubled engineering velocity in 9 months using Claude Code. SpaceX just paid $60B for Cursor because Grok couldn't compete.

The coding layer of the AI stack is consolidating fast. OpenAI has Codex with 4 million users. Anthropic has Claude Code. Google has Gemini Code Assist. Cursor is being absorbed into SpaceX/xAI. GitHub Copilot sits underneath it all. The question UpRound raises is the right one: if AI writes most code, who controls the market? The answer is shifting from "the best model" to "the best distribution." Cursor had distribution — millions of developers — which is why it's worth $60B even though it doesn't have its own model.

For crypto founders specifically: your dev team's productivity is now directly tied to which AI coding tool they use. If your competitor's three-person team ships at the velocity of a ten-person team because they're on better tooling, your headcount advantage is gone. The dev stack isn't a nice-to-have anymore. It's a competitive weapon.

Why it matters: AI writes the code. Distribution controls the market.

Today Signal Question

OpenAI ships a new model every six weeks. The SEC is opening the door for tokenized securities on-chain. Prediction markets raised $1.7B but can’t decide if they’re finance or gambling. And AI writes most of the code now.

If models are commodities, regulation is shifting, and your developer’s productivity depends on which AI tool they use — what’s the one thing you’re building that can’t be shipped by someone else in six weeks?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

59 newsletters from 40 VCs. Sentiment flipped from defensive to opportunistic — first time this week. The biggest AI acquisition of 2026 landed and every VC newsletter pivoted from "monitoring the situation" to "where does the money go now."

• AI/ML mentions: 22 — highest single-day count we’ve tracked. Cursor deal + Chamath letter + a16z continual learning piece

• DeFi/security: 16 — Kelp aftermath cooling but Volo exploit on Sui reignited it

• Stablecoins: 8 — 51 Insights dollar weaponization piece + UK Treasury framework

• Crypto VC mood: capital concentrating. $466M raised in April so far vs $3B in March. big funds eating, small funds starving

• Notable shift: “infrastructure” vs “application” framing appeared in 11 newsletters simultaneously. the Cursor deal made it the consensus thesis overnight

1/ SpaceX Paid $60B for a Code Editor

[AI]

SpaceX announced Tuesday it secured the right to acquire Cursor, the AI coding editor, for $60 billion, or pay $10 billion for their partnership. Cursor was valued at $2.5B in January 2025. $9B by May. $29.3B after a $2.3B Series D in November. Now $60B. That's 24x in fifteen months.

Bloomberg reported the same day that SpaceX engineers have been slow to adopt Grok for coding because it's not good enough. Some xAI staffers have been using Anthropic's Claude instead. Musk's own employees prefer a competitor's product. Two senior Cursor engineers (Andrew Milich, Jason Ginsberg) already left to join xAI last month, reporting directly to Musk. The deal combines Cursor's product and developer distribution with xAI's Colossus supercomputer, reportedly a million H100-equivalent chips.

For founders building at the AI application layer: this is the "infrastructure eats applications" thesis playing out live. Neither Cursor nor xAI has proprietary models that match Anthropic or OpenAI — the same companies now shipping competing coding tools. Cursor still sells access to Claude and GPT. The $10B breakup fee is a massive win for a16z, Thrive, and Accel. But the message is clear: if you're an AI application startup, your exit path is increasingly "get acquired before the model company builds your feature."

Why it matters: Model companies are coming for your layer.

2/ AI Is Killing SaaS Moats

[AI]

Three signals converged this week that confirm the same thesis: traditional software moats are dissolving faster than anyone expected.

First: Cursor couldn't stay independent.

Second: Chamath's annual letter (published April 22) puts numbers on it. Duolingo fell 67.5% from its mid-2025 peak. Adobe fell 24.6%. His thesis: 60-80% of equity value in tech companies lives in terminal value. If AI can replicate your core product in weeks, that terminal value is zero. Private equity bought 1,900+ software companies for $440B between 2015-2025. Software is now 25% of the entire private credit market. If those products become replaceable, the loans default.

Third: this builds on our past signal — three independent VCs (Marcus Rein, Vestigo Ventures, Angular Ventures) all published the same warning about moat erosion on the same day. Traditional defensibility is eroding because AI agents bypass lock-in, abstract workflows, and synthetic data replicates scale. For founders: every VC just read Chamath's letter. The opening question in your next pitch meeting is "how is this not unbundled by an LLM?" Have an answer that isn't "network effects."

Why it matters: "How does an LLM not kill this?" is the new bar.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Blockchain Capital Is Raising $700M

[FUNDING]

Bloomberg reported today that Blockchain Capital is raising $700M across two new funds, a seventh early-stage fund and a second growth fund. The firm manages $2B+ in assets, has a $6B+ portfolio, and has already started deploying the new capital. Expected to close within five to six months. Their LPs: university endowments, sovereign wealth funds, US pension plans. Their portfolio: Coinbase, Kraken, Circle, Tether, Polymarket. This is old-money TradFi capital allocating fresh dollars to crypto.

Now zoom out. Messari data shows average crypto deal size is up nearly 50% in the last 30 days — but total April funding dropped to $466M, down from $3B in March. Capital is concentrating: larger checks into fewer companies. Core Scientific's $1B debt financing (Morgan Stanley) and Polymarket's $600M round account for a massive share. Most activity is still sub-$10M.

The pattern: established firms with institutional LPs are raising aggressively. Newer funds are struggling. Capital is flowing to infrastructure plays that bridge TradFi and crypto. Blockchain Capital raising $700M while smaller funds scramble tells you exactly where the crypto VC market is heading: consolidation at the top, higher bars for everyone else.

Why it matters: Big funds raising. Small funds struggling.

4/ Another DeFi Exploit Hit This Week

[SECURITY]

The DeFi security crisis didn't end with Kelp, that we described in our article earlier. Sui-based Volo Protocol was exploited for ~$3.5M in WBTC, XAUm, and USDC this week, just days after the $292M Kelp exploit. Volo's team froze ~$500K and successfully intercepted a bridge attempt of 19.6 WBTC, keeping those funds out of the attacker's hands. $28M in TVL across other Volo vaults was unaffected. Meanwhile, CertiK flagged that the Kelp DAO attacker is actively laundering stolen funds through THORChain to Bitcoin. The $292M isn't sitting in a wallet — it's being moved through cross-chain infrastructure in real time.

This continues a pattern we've tracked all month: $575M to Lazarus Group across Drift and Kelp in 18 days. $482M in total Q1 losses. 63% from social engineering. And now Volo makes it clear the exploit wave isn't over. Total DeFi losses in 2026 have now crossed $600M and it's only April. Jefferies warned this week that traditional financial institutions may pause their blockchain initiatives to reassess security risks after Kelp. If institutional adoption timelines get pushed back because of preventable exploits, the whole sector pays the price. The protocols that survive all share one trait: security architecture decisions made before they were needed.

Why it matters: $600M+ lost in 2026. It's April.

Today Signal Question

The hottest AI app of the cycle is being absorbed by a model company. SaaS terminal value is collapsing. The biggest crypto VC is raising $700M while smaller funds struggle. And DeFi lost $600M in four months.

If model companies eat the app layer, crypto VC concentrates at the top, and DeFi can't stop getting hacked — where does a founder go?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

• DeFi risk reckoning — 0xResearch, Coin Metrics, and our own digest all converged on the same framing: the yield-chasing era of DeFi needs a reset

• a16z Speedrun — applications open for Jul-Oct 2026 cohort. 60 founders. signal that a16z is still deploying despite “monitoring the situation”

• $11M into Hedera ETPs — DeFi Technologies making institutional alt-L1 treasury bets. not just bitcoin anymore

1/ Three VCs Published on AI Moats the Same Day

[AI]

Marcus Rein, Vestigo Ventures, and Angular Ventures all independently published pieces on competitive defensibility in the AI era on the same day. The consensus is uncomfortable: traditional moats (network effects, switching costs, data advantages) are eroding faster than anyone expected. Angular's framing is the sharpest: they call it the "AI-Native Animal," arguing that companies incrementally adding AI features will be outcompeted by companies rebuilt from the ground up around AI. This isn't the generic "AI will change everything" take.

It's specific: if your moat depends on complexity that AI can simplify, your moat is dead. If your switching cost depends on workflow lock-in that agents can abstract, your switching cost is dead. If your data advantage depends on scale that synthetic data can replicate, your data advantage is dead.

Three VCs arriving at the same conclusion independently is the kind of signal we track. The last time this happened was five newsletters converging on AI compute scarcity. For founders: the question isn't "are you using AI." It's "would you build your company completely differently if you started today." If the answer is yes, your competitors already are.

Why it matters: AI makes your moat irrelevant.

2/ DeFi's Risk Reckoning Is Just Starting

[DEFI]

The kelp aftermath shifted from crisis to autopsy today. 0xResearch called it a "risk reckoning": the language matters because it signals a narrative turn from "growth at all costs" to "sustainability or die." Coin Metrics published protocol health analysis. Our own investor digest covered the contagion mechanics. But the most interesting data point came from Circle: their scenario analysis shows stablecoin infrastructure held stable throughout the weekend stress. USDC didn't depeg. Circle's contracts didn't freeze. The cross-chain settlement layer that everything depends on actually worked.

That's the bifurcation story from yesterday getting sharper — DeFi's yield layer broke, but the stablecoin plumbing underneath it didn't. For founders, the implication is practical: protocols that survived (Fluid, Spark, the stablecoin rails) all shared one trait: they made architecture decisions two years ago that prioritized safety over speed. The protocols that broke made the opposite choice. The market is now repricing which approach it values. Expect "security-first" to become the new "community-first" in every pitch deck for the rest of 2026.

Why it matters: Yield broke. Plumbing didn't.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Blockchains are the Substrate for AI

[INFRASTRUCTURE]

Anand Iyer published a piece arguing that blockchains will serve as foundational infrastructure for physical AI systems, not as the application layer everyone keeps trying to build, but as the substrate underneath. Verifiable compute. Provenance. Coordination between autonomous systems. Data Gravity's Chris Zeoli published a complementary piece on verifiable reinforcement learning in security domains on the same day.

Two different authors, same thesis: the real crypto × AI opportunity isn't "AI tokens" or "AI-powered trading bots." It's using blockchain's core properties (immutability, verifiability, permissionless coordination) to solve problems that physical AI systems can't solve without them.

How do you prove an autonomous vehicle's decision model wasn't tampered with? How do you verify that the training data for a medical AI hasn't been poisoned? How do you coordinate a fleet of delivery drones without a single point of failure?

Blockchains. Not because they're trendy but because no other architecture provides verifiable, permissionless coordination at scale. This is the most technically credible crypto × AI thesis we've tracked — and it has nothing to do with token prices.

Why it matters: Blockchain's real job is verification.

4/ Prediction Markets Are Either the Future of Finance or America's Gambling Problem

[PREDICTION MARKETS]

Pomp published a piece this week positioning prediction markets as essential institutional portfolio infrastructure — not gambling, not speculation, but a core allocation tool for sophisticated investors hedging real-world outcomes. The same week, The Breakdown published a piece on the "gamblification of America," calling out prediction markets specifically. Both are right and that's the problem.

Kalshi processes institutional-grade event contracts regulated by the CFTC. Polymarket runs offshore markets where you can bet on whether a downed pilot gets rescued. The regulatory tension here is identical to what we covered at the CFTC hearing two weeks ago — a congressman showed two betting lines and asked the federal regulator to tell them apart. He couldn't. The prediction markets sector raised $1.7B in Q1. a16z wrote two deep-dives in one month. Kalshi hit a $1B raise. But sports events are 85% of all US activity, and the "is this gambling" question hasn't been answered.

For founders building in this space: the window is open but the regulatory risk is real. Build with compliance infrastructure from day one because it's the only thing that survives the inevitable crackdown.

Why it matters: Regulation will decide.

Today Signal Question

Three VCs said your moat is dying. DeFi proved infrastructure can break in 46 minutes. Prediction markets raised $1.7B but can’t agree on what they are. And the most credible crypto × AI thesis turned out to have nothing to do with tokens.

What are you building that still matters in 3 years?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the news. Now go build something.

— Byblos.digit

Subscribe now

• DeFi crisis dominated — 22 of 54 newsletters mentioned Kelp or contagion6

• Aave bank run — $6.6B in panic withdrawals in 48 hours

• Nuclear convergence — 3 VCs published same thesis on same day. AI needs power.

• a16z said “monitoring the situation” — when they go risk-off, the rest follows

• $1.02B deployed — across 18 Silicon Valley startups last week despite everything

The other half of the inbox pretended nothing happened and wrote about AI productivity and tokenized Treasuries.

1/ Kelp DAO Lost $292M

[DEFI]

Kelp DAO's cross-chain bridge got exploited for $292M on Saturday, the largest DeFi hack of 2026, overtaking Drift by a few million. The attacker tricked LayerZero's messaging layer into approving a fake transaction, draining 116,500 rsETH (18% of circulating supply) across 20+ chains. Kelp's emergency pause kicked in 46 minutes later but too late.

The stolen rsETH got deposited into Aave as collateral, borrowed WETH against it, and left ~$196M in bad debt. Aave's TVL dropped $6.6B in 48 hours as users panic-withdrew, creating classic bank run dynamics. Nine protocols froze rsETH markets. LayerZero blamed Kelp's 1-of-1 verifier setup. Kelp fired back saying that was LayerZero's own default configuration. The Spiderman-pointing-at-Spiderman meme is having its crypto moment.

LayerZero linked the attack to North Korea's Lazarus Group, the same unit behind Drift's $285M exploit on April 1. That's $575M drained by one state actor in 18 days through two structurally different attack vectors. Ledger's CTO warned 2026 is shaping up as one of the worst years for DeFi security. The counterpoint from protocols like Fluid and Spark, who had rate limits and isolated exposure from day one, is that this was preventable.

Why it matters: $575M stolen in 18 days.

2/ Wall Street Is Tokenizing Itself

[TRADFI]

While DeFi was having a bank run, DTCC, the company that settles virtually every US securities trade, ismoving to production on tokenized US Treasuries. The SEC gave DTC a No-Action Letter authorizing a three-year tokenization service on pre-approved blockchains.

Eligible instruments: Russell 1000 equities, major index ETFs, and US Treasury products. DTCC partnered with Digital Asset to mint tokens on the Canton Network, targeting an MVP in H1 2026 with expansion based on client demand. The tokens carry the same legal protections and ownership rights as conventional entries. 51 Insights declared "Wall Street is tokenizing itself", which is the kind of sentence that would've gotten you laughed out of a boardroom in 2022.

The timing is ironic and instructive: DeFi's trustless infrastructure lost $292M because one verifier wasn't verified. TradFi's tokenization service launches with SEC authorization, quarterly reporting requirements, strict access controls, and the full weight of existing market infrastructure behind it.

Why it matters: DTCC tokenizing Treasuries is production

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Intercom Doubled Engineering Output With AI in 9 Months

[AI]

Intercom published a case study showing it 2x'd engineering velocity using Anthropic's Claude Code over nine months. Not a vague "productivity gains" claim but actual measured output doubling. Madrona's latest piece frames the shift as "AI buyers want outcomes, not demos”, and Intercom just provided the outcome. This matters for crypto founders because it answers the question every VC is quietly asking: "does AI tooling actually work or is this another hype cycle?" If a 400+ person company can double engineering output with a coding assistant, the implied savings and speed improvements cascade through every startup's burn rate model.

Five newsletters covered this in 24 hours. The thesis connecting to last week's narrative: AI raised $242B because the ROI is becoming provable. The companies that can show "we shipped twice as fast with half the team" will raise at better terms than the ones showing "we integrated GPT into our landing page."

For founders: if you're not using AI coding tools yet, you're competing against people who build twice as fast as you.

Why it matters: 2x output in 9 months

4/ Three VCs Published on Nuclear Energy on the Same Day

[ENERGY]

AlleyCorp spotlighted its investment in Valar Atomics.

James Riney interviewed Founders Fund's Scott Nolan on nuclear fuel and the US-Japan energy alliance.

NZVC pitched a co-investment on AI inference infrastructure backed by Bezos, Gates, and Zuckerberg through Village Global.

Three separate VCs, same day, same thesis: AI compute demands are creating unprecedented energy requirements, and nuclear is the only clean baseload option that can scale fast enough. This is the physical-world version of the "AI needs crypto rails" narrative from last week — except instead of payment rails, it's power rails. AI data centers consume 10-50MW each. The new generation coming online needs 100-500MW. Wind and solar can't deliver baseload at that scale. Nuclear can. The US Department of Defense budget crossing $1T and global defense spending hitting $2.7T adds a national security dimension.

For crypto founders specifically: DePIN networks connecting distributed energy assets to AI compute facilities is the intersection where all three of these capital flows converge. If you're building there, the thesis capital is forming right now.

Why it matters: AI needs power. Nuclear is getting VC checks.

Today Signal Question

DeFi lost $6.6 billion in panic withdrawals this weekend while DTCC quietly started tokenizing US Treasuries with SEC authorization. Both are putting assets on-chain. One had a bank run. One had a filing number.

Which version of on-chain finance do you think wins the next decade?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

58 newsletters from 39 VCs hit the inbox in the last 24 hours — up 26% from yesterday, which means either the market is heating up or every fund manager simultaneously discovered that Thursday is content day.

AI infrastructure dethroned stablecoins for the first time in three weeks with 16 mentions.

Prediction markets held strong at 12 — a16z published another deep-dive on the sector, which at this point is less "thought leadership" and more "we're about to write a lot of checks, please form an orderly queue."

Stablecoins came third at 11 mentions but with a new flavor: European banks are dominating the conversation now, not Circle and Tether. Solana got 8 mentions on bullish Q1 data.

The overall mood: VCs are deploying like it's 2025 while prices act like it's 2023. Somebody's going to be right and somebody's going to feel silly. We'll find out who by Q3.

1/ a16z Raised $15 Billion and Spent Some of It on a Submarine Company

[VENTURE]

a16z announced $15 billion in new funds, the largest venture capital fundraise in history, representing 18% of all US venture dollars in 2025.

The breakdown: $6.75B for growth, $3B for other venture strategies, $2B targeting crypto (Fund V, closing mid-2026), $1.7B each for apps and infrastructure, $1.176B for American Dynamism, and $700M for bio. Then they backed Ulysses, a physical-world AI company under the American Dynamism practice — the team compared it to the submarine in Disney's Atlantis, which is either a brilliant cultural reference or the most expensive nerd joke in venture capital history.

For crypto founders specifically: Fund V is targeting $2B, down from $4.5B in 2022, but still the largest crypto fund currently raising. Recent crypto deployments include Jito ($50M), Kairos ($2.5M prediction market terminal), and The Better Money Company ($10M stablecoin clearinghouse). Chris Dixon's thesis: crypto is in its "financial era" and finance is "the foundation for everything else." If you're building infrastructure that connects financial plumbing to the real world, you're in the strike zone. If you're building another DEX fork, the line for rejection starts to the left.

Why it matters: Largest VC raise ever. is about submarines.

2/ Twelve European Banks Built a Euro Stablecoin

[STABLECOINS]

While the US debates stablecoin yield rules, Europe did the thing. Twelve banks (ING, BNP Paribas, UniCredit, BBVA, CaixaBank, Danske Bank, SEB, KBC, DZ Bank, Raiffeisen, DekaBank, and Banca Sella) formed Qivalis, a consortium launching a MiCA-compliant euro stablecoin in H2 2026. Incorporated in the Netherlands, supervised by the Dutch Central Bank, and run by Jan Sell, formerly head of Coinbase Germany, because apparently the best way to compete with American stablecoin dominance is to hire someone who used to work for it.

Reserves: 40% bank deposits, the rest in eurozone sovereign bonds.

Use cases: 24/7 cross-border payments, programmable settlement, on-chain settlement for tokenized securities.

A separate group (ING, UniCredit, BNP Paribas) is also building a Swiss-franc stablecoin. Societe Generale and Oddo BHF already have live MiCA-compliant stablecoin products. EU USDC volume climbed 109% in six months. The digital euro CBDC won't arrive until 2029 at the earliest. These banks are building around their own central bank, which is the most European power move imaginable.

Why it matters: Europe's stablecoin stack is being built (without you).

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Base Broke Up With Optimism

[L2]

Coinbase's L2 Base did two things that matter.

First, it launched a Creative Studio initiative, signaling a hard pivot from "DeFi chain" to consumer/creator platform — think onchain art, loyalty programs, gaming, and social apps.

Second, and more structurally: Base announced in February that it's leaving the OP Stack and moving to its own "base/base" stack, severing a three-year dependency on Optimism. That means Base keeps its sequencer revenue (previously shared with Optimism's treasury), ships faster without Optimism's release cycle, and controls its own roadmap.

The numbers justify the confidence: Base is now the largest Ethereum L2 with $9.15B in DeFi TVL, 880K daily active addresses, 3.7B cumulative transactions, and $1.7B+ bridged TVL. USDC dominance on the chain is 90%. Base is also exploring a native token for 2026, which would be one of the most anticipated launches in crypto given Coinbase's 110M verified users as a distribution moat.

For founders: if you're building consumer crypto (social, creator tools, gaming, loyalty) Base just made it very clear they want you. The creative studio, the Solana bridge, the split from Optimism — it's all one message: we're building the consumer layer and we're doing it alone.

Why it matters: The biggest L2 went independent.

4/ One Founder Spent $15K to Prove You Don't Need the Cloud

[AI INFRA]

Kerman Kohli published a deep-dive on building his own AI inference machine, a $15K box with an NVIDIA RTX 6000 Pro (96GB VRAM), 128GB DDR5 RAM, and 4TB NVMe storage, assembled by hand over a weekend with zero prior hardware experience.

The thesis behind the build: compute is getting harder to access, the big labs will steal your ideas and train on your data, and the real prices of LLMs are about to get very real now that Anthropic has session limits and OpenAI is killing Sora. Kohli burns 100-200M tokens per day and decided the math favored owning hardware over renting it.

The contrarian insight buried in the piece: "worst case scenario if I don't use it, I can resell it for more in 6 months given the way compute prices are heading" — which is basically going long on compute scarcity as an asset class. This connects to a broader theme in the inbox: Village Global published on AI investing theses, Illuminate Financial flagged "the missing link for AI agents," and Dror Berman's Innovation Endeavors piece from yesterday argued "the era of building software for humans is ending."

Five newsletters in 24 hours focused on AI inference, compute economics, and decentralized AI stacks. If you're a crypto founder building at the AI × crypto intersection (DePIN, decentralized compute, agent infrastructure) the VC attention on this vertical is the highest we've tracked.

Why it matters: Own your compute or rent your future.

Today Signal Question

For today’s issue, it’s simple. What are you building that doesn’t need permission?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

• 47 newsletters, 32 VCs across fintech/defense/crypto/AI

• Applied Intuition $15B discipline example

• RWA $27.6B on-chain data point

• Shield AI & Anduril execution narrative

• SEC Reg Crypto regulatory certainty thesis

1/ SEC Reg Crypto Safe Harbor Reshapes Fundraising

[REGULATORY]

The SEC submitted "Regulation Crypto" to the White House on April 6, expected to publish within 2-4 weeks. Two explicit safe harbors: $5M for four years (seed/Series A), $75M for twelve months (Series B+). An Innovation Exemption lets startups test products without pre-approval. The SEC has already dropped 60%+ of pending enforcement cases and the March 17 SEC-CFTC interpretive release clarified that Bitcoin, Ethereum, Solana, and XRP are not securities.

Why it matters: Regulatory certainty compresses legal spend and lets VCs size checks upward.

2/ RWA Infrastructure Gets Funded, RWA Issuance — Doesn’t

[FINTECH]

On-chain tokenized assets hit $27.6B (BlackRock's BUIDL fund, JPMorgan's JPM Coin on Base, sovereign bond tokenization from Canada and France). Yet capital flows to infrastructure, not issuers. Nava ($8.3M seed for AI agent escrow), OpenFX ($94M Series A for fintech rails), and Keyrock ($1.1B valuation) signal the pattern. Franklin Templeton's acquisition of 250 Digital shows institutional buyers want execution, settlement layers, custody, compliance, not another tokenization wrapper.

Why it matters: RWA issuance became a regulated commodity.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Applied Intuition's $15B Wins Through Embedded Discipline

[DEFENSE TECH]

Applied Intuition closed $600M Series F at $15B valuation. CEO Younis revealed the unglamorous moat: when the company was 12 people, it documented why early success happened and codified 10 values into 50 observable behaviors tied to every hire and promotion. No breakthrough algorithm. No AI hype. Just consistent institutional discipline. $2.3B ARR in 2024, 60% YoY growth, 89% retention, 31+ founders on staff. IPO expected 12-24 months.

Why it matters: Execution discipline in observable behaviors beats technology novelty.

4/ Kleros Court V2 Technical Win, Economic Loss

[DEFI]

Kleros completes Court V2 contract reviews with Futarchy UI for jury selection and improved Scout analytics. But PNK token crashed to $0.012-0.032 (97% from ATH), $8.8-24.3M market cap, episodic volume. The protocol solved the technical problem, decentralized dispute resolution works. It did not solve the economic problem, token incentives can't bootstrap sustainable juror participation at scale. No protocol-wide dispute flow, no viable exit for jurors.

Why it matters: Stress-test any protocol’s incentive design at 10x TVL before committing engineering.

Today Signal Question

If VCs see you as just wrapping bonds or stocks on-chain with zero institutional relationships and zero platform stickiness, they won’t fund Series A.

What they will fund: settlement layers, compliance oracles, custody infrastructure, escrow systems for autonomous agents.

What would it take for RWA issuance startups to get Series A funded again?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the news. Now go build something.

— Byblos.digital

Subscribe now

• 58 newsletters, 40 VCs — highest count this week

• Stablecoins dominated for the third day running, but the tone shifted: yesterday was regulatory wins, today is structural questions

• Physical AI showed up as its own category for the first time — distinct from robotics and enterprise software

• DeFi security spiked — the Drift Protocol hack was a governance failure, not a smart contract bug. That distinction matters.

• Sentiment is genuinely split — regulatory clarity pulling one way, geopolitical risk and security incidents pulling the other

Byblos Founder Network — Early Access

Know what investors are thinking before you walk into the room.

We track 600+ VC newsletters daily and surface what investors are actually thinking — deal flow, thesis shifts, sector conviction — in one feed built for founders raising or preparing to raise. We're keeping the first cohort small and founder-focused.

Apply for early access to be added to a private community →

Byblos digital early access

1/ Investors are questioning whether stablecoins actually need blockchains

[STABLECOINS]

Ethereum processes 20–30 transactions per second on average. Visa processes 1,700. Global card networks handle 25,000+ combined. That gap doesn’t close on current architecture: even with all the Layer 2 solutions, the practical ceiling before fees start rising is around 350 TPS.

The result: stablecoins have moved $300 billion in market cap since 2020, but research tracking organic economic transfers found only 10–20% of on-chain stablecoin volume represents real payments. The rest is bots, internal exchange transfers, and automated trading. Ten31 Research highlighted this today.

Meanwhile, bridges between blockchains, the infrastructure needed to move stablecoins between chains, have lost over $3 billion in user funds over the past several years. And Ethereum gas fees have historically averaged $1–5 per transfer, spiking to $50+ during congestion. Legacy payment rails charge less for more volume.

The biggest tell is what stablecoin issuers are actually building. Both Circle and Stripe have launched their own permissioned payment networks - closed systems controlled by a small consortium of validators. They function like databases. Tether, which generates over $10 billion in annual free cash flow, built a toolkit that abstracts the payment rail entirely. Tether can route a transaction across an internal ledger without touching a public blockchain at all.

The economics are clear: every transaction routed internally is a validator fee kept rather than paid out. At $10 billion in annual earnings, Tether has every incentive to cut out the public chain.

Why it matters: The stablecoin infrastructure layer is being repatriated by issuers.

2/ Physical AI is becoming its own investment category

[INNOVATION]

The AI conversation has been almost entirely about cognition: reasoning, language, writing, code. A new thesis is forming around physical agency.

The architectural unlock happened in 2023. Vision-Language-Action models take a camera feed and a plain language instruction and produce physical movement as output.

The investment gap is also a geopolitical one. South Korea runs 1,220 robots per 10,000 manufacturing workers. The US runs 307. China shipped roughly 90% of the world’s humanoid robots in 2025.

The opportunity for founders is in the software layer: the orchestration tools, simulation environments, and training data infrastructure that makes physical AI deployable at scale.

Why it matters: The next investment frontier is models that can pick things up.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Banks are losing and most of them don't know it yet

[TRADFI]

US bank branches have fallen by more than 20% since 2008. Meanwhile 60–80% of banking customers now cite digital or mobile as their primary way of managing money, including 61% of baby boomers.

The NPS gap tells the deeper story. Fintechs like Chime and Robinhood regularly score 70–90 on customer satisfaction. Traditional banks average below 30. Customers are comparing their local branch to Venmo.

Robinhood has reached a valuation in the tens of billions, rivalling institutions that took decades to build. Wise and Felix are handling cross-border transfers that Western Union built an empire around. The return on equity for most banks is now barely above their cost of capital.

The response from smart incumbents is not internal transformation: experience shows that’s slow, expensive, and culturally wrong. Fifth Third Bank went from 2–3 app updates per year to over 400. They acquired Rize Money, partnered with Stripe, and bought fintech lenders Provide and Laurel Road. That’s what adaptation looks like.

For founders building in fintech or payments, the structural moment is clear: incumbents are increasingly willing to partner or acquire rather than build. The banks that engage with the fintech ecosystem will survive.

Why it matters: AI just made the succession gap accessible for small businesses.

4/ Your AI moat looks bigger than it is. Here's proof.

[AI]

When the dominant player in Indian container glass manufacturing went bankrupt in 2021, the second-largest company watched its margins double, from 11% to 24%, and return on capital jump from 3% to 20%. Because the biggest competitor was out.

In September 2025 that competitor was acquired by new owners backed by Cerberus Capital Management. They committed 2.5× the market leader’s annual earnings in fresh investment on day one.

The AI version is the same story playing out faster. Harvey built $200M ARR in legal AI while Thomson Reuters was slow. Cursor grew to hundreds of thousands of paying developers while Microsoft Copilot sat at 3.3% activation. Glean owned enterprise search while IT teams were still figuring out their strategy.

The incumbents are coming back: Thomson Reuters launched CoCounsel. Microsoft is retooling Copilot. Salesforce rebuilt Einstein. They have enterprise distribution, compliance infrastructure, and legal relationships no startup has.

Founders who built proprietary training data, deep integrations, and load-bearing workflows will survive. Founders who built clean UI on top of a generic model and collected margin while the incumbent was.

Why it matters: A temporary competitive gap is only a moat if you built something inside it.

Today Signal Question

Stablecoin issuers are building their own closed payment networks — Circle, Stripe, and Tether are all quietly reducing their dependency on public blockchains.

If that trend continues, what does the web3 payments stack actually look like in three years?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

• 57 newsletters hit the Byblos feed today from 39 unique VC publishers — the highest daily count this week

• Blockchain infrastructure had its strongest day of the week, with LayerZero, Polymarket, and Circle all generating independent newsletter mentions

• Security emerged as a quiet new narrative: Frankly Speaking opened a series on AI-enabled product security that directly challenges how existing tools work

Narratives today: Developer toolchain breaking under AI pressure · Activation as the real B2B AI bottleneck · Blockchain scalability reaching institutional grade · AI product security becoming its own category · GTM spend no longer enough to win

1/ a16z backs GitButler — version control is the next thing AI breaks

[AI]

Git was designed 20 years ago for humans working linearly. Modern developers now run multiple AI agents in parallel: one fixing a UI bug, another optimising a database query, a third updating documentation. The problem: Git’s core assumption is that your local copy represents a single, atomic change. It breaks under parallel editing. GitButler is rebuilding version control from scratch with this in mind. It’s native support for parallel branches, agent-specific commands, and rich metadata designed for how AI-augmented development actually works. a16z led the Series A. The CEO is Scott Chacon, who co-founded GitHub and literally wrote the book on Git. His entire career has been about making Git accessible. GitButler is finishing that mission for the AI age. a16z’s framing: “The developer’s role is shifting rapidly from programmer to system architect and agent orchestrator.” The toolchain hasn’t caught up yet.

Why it matters: Every AI coding tool is constrained by version control infrastructure.

2/ Zapier went from 10% to 97% AI feature adoption with one internal hackathon

[PRODUCT]

One of the most useful product insights from this week’s newsletter feed: Zapier’s AI feature usage sat at 10% across their user base until they ran a single internal hackathon. It jumped to 97% immediately after. The GTM Newsletter covered this as a case study in activation strategy.

The GTMnow Newsletter (by GTMfund)

GTM: How Zapier got to 97% company-wide AI adoption

Listen now

a month ago · 12 likes · GTMnow and Sophie Buonassisi

What changed was that people actually experienced what the product could do. This is the most common problem in B2B AI right now: availability is not the bottleneck. Activation is. Most B2B SaaS companies building AI features assume that shipping is the hard part. Zapier’s data says the hard part is getting one person in a team to have the experience that makes them go back and tell everyone else.

Why it matters: Activation, not availability, is what determines whether an AI product gets used.

3/ The CEO of LayerZero just said "the scalability trade-off is dead"

[BLOCKCHAIN INFRA]

Bryan Pellegrino, CEO of LayerZero, the cross-chain messaging protocol backed by Citadel and NYSE, sat down with 51 Insights this week and made a claim that would have been dismissed two years ago: the scalability trade-off is over. The blockchain trilemma: the idea that you can only have two of decentralisation, security, and scalability. It has been the defining constraint of the industry since Ethereum launched. Pellegrino’s argument is that the infrastructure has matured to the point where the trade-off no longer exists in practical terms. The fact that Citadel Securities and NYSE have backed LayerZero matters as context: these are not companies that bet on “almost there” infrastructure. If institutional capital at that level is validating the scalability claim, it’s worth taking seriously.

Why it matters: Enterprise-grade blockchain throughput removes the last infra objection to building.

4/ A new security category is forming

[SECURITY]

Frankly Speaking, a VC-backed security analyst newsletter, opened a new series this week on what they're calling "AI-enabled product security" — and the framing is relevant for any founder shipping AI features. The core argument: traditional security tools like Snyk and Semgrep were built to scan for known vulnerabilities.

Frankly Speaking

AI-enabled product security (part 1)

Disclaimer: Opinions expressed are solely my own and do not express the views or opinions of my employer or any other entities with which I am affiliated…

Read more

25 days ago · 6 likes · Frank Wang

They generate alert noise that buries engineering teams. AI agents now ship code faster than any human security team can review it. The solution is not more security headcount — it's AI reviewing AI. "You have to combat AI velocity with AI velocity." The newsletter identifies three customer types:

• AI-native companies (who will just build their own security tools), legacy companies (who will stick with old tools)

• AI-progressive companies - the ones actively adopting AI but still figuring out execution.

• The market.

Every early-stage AI startup with a real codebase is an AI-progressive company right now. The newsletter also makes a harder claim: "The era of winning through GTM spend is over. Product superiority is everything now."

Why it matters: Every team shipping AI features at speed is accumulating security debt they cannot yet audit.

Today Signal Question:

Git was built for humans. Traditional security tools were built for humans. Version control, code review, vulnerability scanning — the entire developer toolchain assumes a human is doing the work.

Which tool in your stack is most likely to break first as your team shifts to AI-assisted development?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

• Sentiment split for the first time in days: neutral leads at 15, bullish at 12, mixed at 7, bearish at 2. The market is digesting a lot of conflicting signals at once.

• The open vs. closed AI infrastructure bifurcation emerged as a cross-newsletter theme that nobody was explicitly tracking as a thesis going into the week. It formed bottom-up from the data.

Byblos Founder Network — Early Access

Know what investors are thinking before you walk into the room.

We track 600+ VC newsletters daily and surface what investors are actually thinking — deal flow, thesis shifts, sector conviction — in one feed built for founders raising or preparing to raise. We're keeping the first cohort small and founder-focused.

Apply for early access to be added to a private community → Byblos digital early access

1/ SpaceX confidentially files for the biggest IPO in history

[FUNDING]

SpaceX confidentially filed for an IPO targeting a $1.75T valuation and a $75B raise, which would make it the largest public offering ever, nearly 3.5× the size of Saudi Aramco. The company merged with xAI in February setting a combined $1.25T valuation. Now the June 2026 listing is taking shape under the codename Project Apex with 21 banks lined up.

The twist: banks wanting access to the deal must first commit to using Elon Musk’s xAI Grok, bundling AI ecosystem adoption into the most coveted IPO allocation in a decade. Retail investors reportedly get 30% allocation, triple the usual. SpaceX hit $15–16B in revenue last year. Cathie Wood projects the company reaches $2.5T by 2030.

Why it matters: The SpaceX IPO sets the public benchmark for space infrastructure.

2/ Anthropic kills 135K developers’ access

Anthropic blocked 135,000 OpenClaw users from running the open-source agent framework on Claude subscriptions with less than 24 hours notice, sending monthly costs from $200 to $675+. A single day of OpenClaw on Opus reportedly cost $109 in tokens versus $6 for Claude Code.

Two days earlier, Google released Gemma 4 under Apache 2.0 — four model sizes from 2B to 31B parameters, zero restrictions, no access revocation possible. The 31B model ranked third on Arena AI’s leaderboard. AIME 2026 math performance jumped from 20.8% to 89.2%. The contrast is the story: one platform tightened the gate, the other removed it entirely.

Robot Wave

AI Waves #4. Platform Risk: Anthropic shut the gate on 135,000 developers. Google opened it.

AI Waves #4…

Read more

a month ago · 28 likes · Steven Waterhouse

Why it matters: If you're building on a proprietary API without an exit route, you're renting your product.

3/ OpenAI closes $122B at $852B valuation

[AI]

OpenAI closed the largest funding round in tech history, $122B led by Amazon ($50B), NVIDIA ($30B), and SoftBank ($30B), reaching an $852B valuation. The company now generates $2B in monthly revenue with 900M weekly active users. An advertising pilot launched six weeks ago already exceeds $100M ARR. ARK Invest is adding OpenAI to multiple ETFs ahead of a reported late-2026 IPO.

The number that matters: $122B buys roughly 18 months of runway at current burn. Jensen Huang claimed this week that AGI has been achieved. Mark Gubrud, who coined the term, agreed.

Why it matters: Staying in the frontier AI race is expensive.

4/ FDIC approves GENIUS Act stablecoin framework

The FDIC Board approved a proposed rulemaking implementing the GENIUS Act, establishing the first federal prudential framework for stablecoin issuers, redemption protocols, capital requirements, and risk management. This brings stablecoin issuance inside the formal federal supervisory perimeter for the first time. The framework addresses custodial services, pass-through insurance for reserve deposits, and tokenized deposits. This lands the same week Morgan Stanley launched the first major bank Bitcoin ETF, and Stellar reports ~6M daily operations with $1.3B in distributed asset value — up 50% YTD. The GENIUS Act plus Franklin Templeton tokenized ETFs plus the Third Circuit prediction market ruling are three separate branches of the same tree: crypto becoming regulated financial infrastructure.

Why it matters: Stablecoin founders now have a clear compliance path.

Today’s Signal Question

Anthropic closed the gate on 135,000 developers overnight. Google opened one the same week with Gemma 4 Apache 2.0. SpaceX is bundling xAI adoption into the most coveted IPO allocation in years.

The infrastructure layer is consolidating fast, and the companies controlling access to it are using that control aggressively.

Are you building on rails you actually own?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

• 15 newsletters tracked in the last 24 hours — all from a single firm

• Sentiment: 67% bullish

• AI & ML led mentions (6), followed by Enterprise SaaS (5), Infrastructure (5), Payments & Stablecoins (4)

• Dominant narrative: institutional crypto moving from experiment to infrastructure

• The thread connecting everything this week: stablecoin rails, AI agent governance, and on-chain distribution are all converging on the same capital pool

1/ Franklin Templeton puts five ETFs on-chain via Ondo Finance

[RWA]

A $1.7T asset manager just made five ETFs available in crypto wallets, 24/7, no brokerage account needed. Franklin Templeton partnered with Ondo Finance to tokenize growth equities, gold, large-cap, high-yield credit, and income equity ETFs — the tokens trade on-chain via a special-purpose vehicle, with Ondo market makers providing liquidity even when traditional markets are closed. Ondo now controls roughly 70% of the tokenized equity market, with $700M+ TVL and $13B in cumulative volume. The market was $32M in January 2025. It’s approaching $1B now. Franklin Templeton’s Head of Innovation put it plainly: digital wallets will eventually hold “the totality of an individual’s financial life.” The NYSE picked BlackRock-backed Securitize as its tokenization partner the same week. The race is on.

Why it matters: Franklin Templetohe removed the biggest barrier to institutional tokenization - distribution.

2/ Sycamore raises $65M seed for the OS that governs enterprise AI agent fleets

[ENTERPRISE]

Former Atlassian CTO Sri Viswanath just raised one of the largest seeds in AI history, $65M, 14x the startup median, to build the operating system that manages enterprise AI agent fleets. Coatue and Lightspeed co-led, with the Databricks CEO, Okta co-founder, Intel CEO, and former OpenAI Chief Research Officer all writing personal checks.

The core thesis: companies are deploying dozens of AI copilots with no way to trust, audit, or control them. Sycamore builds the governance layer between the enterprise and its agents, using a progressive trust model where agents earn autonomy through demonstrated reliability. Gartner says 40% of enterprise apps will include task-specific agents by end of 2026, up from under 5% in 2025.

Why it matters: Investors are paying seed prices for the governance layer, not the agents.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

---

3/ Federal appeals court blocks states from regulating prediction markets

[REGULATORY]

The Third Circuit just ruled that Kalshi’s prediction market contracts are federal “swaps” which means states have zero jurisdiction over them. New Jersey tried to shut Kalshi down under gambling laws. The court said no. This is the first federal appeals court ruling on the question, and it comes as the sector is logging record numbers: Kalshi processed $13.07B in March, Polymarket $10.57B — $25.7B combined, the biggest month on record.

The problem is the Ninth Circuit went the other way, declining to block Nevada’s enforcement against Kalshi. That circuit split, plus 19 active federal lawsuits and criminal charges filed by Arizona, almost guarantees a Supreme Court case. The Trump administration already sued three states, Arizona, Connecticut, and Illinois, to block enforcement. On April 16 a consolidated Kalshi, Crypto.com, and Robinhood case goes to the Ninth Circuit.

Why it matters: One SCOTUS ruling determines whether it lives as a national product or dies.

4/ OpenFX raises $94M to make stablecoins the invisible layer under global FX

[PAYMENTS]

OpenFX just raised $94M Series A at a ~$500M valuation to settle cross-border FX through stablecoins in under 60 minutes, for sub-10 basis points. Traditional correspondent banking costs ~69bps and takes 2–5 business days. The model, founder Prabhakar Reddy calls it the “stablecoin sandwich”, converts sender fiat to stablecoins, moves value across borders instantly, converts back.

Neither party ever touches crypto, and the company just hit $45B in annualized volume, up from $4B a year ago. Accel, Atomico, Lightspeed Faction, Northzone, and Pantera backed the round. Reddy’s boldest line: “Within ten years, AI agents will be the largest category of FX users on the planet.” That’s a plumbing pitch. The $7T-per-day FX market still runs on rails built in the 1970s.

Why it matters: The founders building on top can inherit $200T in global capital flows.

Today's Signal Question: Institutional money is moving on-chain, stablecoin rails are replacing correspondent banking, and AI agents are being handed governance layers. The infrastructure is being built without most founders noticing.

Are you early to the right layer or are you about to build on top of someone else's moat?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

21 newsletters in. 3 VC firms publishing.

Sentiment split 10 bullish, 6 neutral, 3 mixed, 2 bearish — the most divided we've tracked this week, and the data explains why.

Two things are true simultaneously: AI coding capability hit a genuine threshold in November 2025 and supply chain attacks are now exploiting the exact same infrastructure. The security thread underneath that number is what most coverage missed.

1/ AI Coding Agents Just Crossed the Autonomy Threshold

[DEVELOPER TOOLS]

Simon Willison, writing in this week’s newsletters, identifies November 2025 as the month AI coding agents moved from “mostly works” to “actually works” — enabling what he calls dark factory patterns where AI writes, reviews, and QA’s code without any human in the loop. The trigger was the Claude Opus 4.5’s capacity for long-range task delegation, combined with tools like OpenClaw, which hit 250,000 GitHub stars in four months, surpassing React’s adoption curve.

The employment implication is counterintuitive. Junior developers are not the primary target. Mid-career engineers — the people responsible for architectural judgment, code review, and cross-system orchestration — face the highest automation risk. Those are exactly the tasks agentic systems now handle. Replit raised $400M at $9B this week betting on a different thesis: that human creativity still belongs in the loop, and the opportunity is AI systems that enhance rather than replace it.

Why it matters: The automation risk is the engineers doing code review and architecture.

2/ Axios Got Hacked. 100M Weekly Downloads

[INFRASTRUCTURE]

The popular HTTP library Axios, 100M+ weekly downloads, was compromised via a hijacked maintainer account. The attacker installed OS-specific remote access trojans that self-deleted after execution to evade detection. This followed the TeamPCP campaign, which demonstrated cascading supply chain failure across five separate ecosystems in eight days.

The a16z newsletter this week frames this as a new category of threat, not an evolution of an old one. The problem: frontier AI models now write and manage dependencies at machine speed. The average application contains 1,100+ open source components with 755+ transitive dependencies, most chosen by nobody on the development team. Attack surface is expanding faster than human monitoring capacity can track it. Prompt injection remains an unsolved security vulnerability sitting directly at the intersection of both trends.

Why it matters: Your AI is choosing dependencies you never approved, at a speed no human can monitor.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

---

3/ Anthropic Paid $400M for an 8-Month-Old Startup

[FUNDING]

Anthropic acquired Coefficient Bio in a stock deal valued at over $400M. The company was eight months old and operating in stealth. It joins Anthropic’s Health Care Life Sciences team, targeting AI models for biological research and a broader bet on artificial superintelligence in science. The exit generated a 38,513% IRR for Dimension, the fund that backed it.

The same week, Replit closed $400M at a $9B valuation, the same dollar figure, a completely different thesis. Where Anthropic is betting on AI replacing human judgment in scientific research, Replit is betting on AI amplifying human creativity, with humans explicitly kept in the loop. Two $400M bets on what AI’s role in creation should be, written in the same week.

Why it matters: AI × biotech is the category where the fastest capital cycles are currently happening.

3/ The Agent Economy Hit $600M Annualized Volume

[PAYMENTS]

OKX Ventures’ AI Agent Economy Infrastructure Report, published in this week’s newsletters, contains a number that didn’t make headlines: average daily transaction volume on agent infrastructure protocols declined more than 80% from the December 2025 peak as of March 2026. This is happening while the capability layer, the actual AI, continues advancing. The infrastructure and the use of it have decoupled.

Jamie’s Substack

The Agent Economy Awakens

Base, Coinbase’s Ethereum Layer-2 blockchain, has exploded as the go-to hub for AI agents—autonomous software entities that think, act, and transact without constant human input. Launched in 2023, Base’s low fees (often under a cent per transaction) and seamless Ethereum integration make it perfect for agents needing to zip through thousands of micro-ac…

Read more

2 months ago · 3 likes · 4 comments · Jamie

The underlying infrastructure, however, is real. Coinbase and Cloudflare’s joint x402 protocol processed 100M+ transactions with $600M annualized payment volume by end of 2025, embedding stablecoin micropayments directly into the HTTP protocol layer. ERC-8004 — a trust layer giving AI agents on-chain identity, reputation, and verification registries — went live on Ethereum mainnet on January 29, 2026, built with MetaMask, Google, and Coinbase. Virtuals Protocol deployed 18,000+ agents with an agent GDP exceeding $479M. The rails exist. The question is whether the activity follows.

Why it matters: Agent economy infrastructure is still in a correction.

4/ YC’s W26 Batch Quietly Marks the End of the “ChatGPT Wrapper” Era in Startups

Multiple newsletters this week reference a shift in the YC W26 batch: significantly fewer “ChatGPT wrapper” companies than prior cohorts. The focus has moved to enterprise AI infrastructure, industry-specific tooling in legal and finance, and robotics infrastructure. This is a market maturation signal, founders are building deeper, more defensible products because the wrapper category has been commoditised by the foundation models themselves.

The broader narrative emerging in the newsletters frames this as the Systems of Action era: AI moving from reactive assistance to proactive autonomous decision-making. The analogy cited is Taylor’s scientific management revolution — the moment when individual craft skill was replaced by systematic, optimised process. Knowledge work is at that inflection point now.

Why it matters: Pitch based on LLM is now about a W24 company to investors who are now funding W26.

Today Signal Question: If your product depends on open source dependencies, who is actually watching them?

The question for your team isn't whether you're using AI to write code. It's whether your security posture scaled alongside it.

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the news. Now go build something.

— Byblos.digital

Subscribe now

Today's 12 newsletters came in with a split mood:

5 bullish, 4 neutral, 3 mixed

The capital allocation tells a cleaner story. AI & ML dominated at 8 mentions, but the narrative underneath isn't about AI features. It's about AI replacing the fundamental logic of how software companies are built. Redpoint and Emergence both published this week, and both land on the same conclusion: the companies getting funded now grow faster with fewer people, compete for headcount budgets instead of software budgets, and own outcomes instead of selling tools. That framework maps directly onto what's happening in legal AI. Read more in today’s digest below.

1/ Harvey and Legora Together Raised $750M in Three Weeks

[VERTICAL AI]

Harvey closed $200M at an $11B valuation, its fourth raise in 14 months, up from $8B just months prior, led by Sequoia and GIC. 100,000+ lawyers across 1,300 organizations are already using the platform. Over 25,000 custom AI agents have been deployed on its infrastructure. Harvey is pushing into Europe.

Three weeks earlier, Legora, Swedish-born, now headquartered in New York, raised $550M at $5.5B led by Accel, tripling its valuation in five months. It grew from 40 to 400 employees in a single year and from 250 to 800 client firms in under 12 months. Legora is pushing into the US. Both companies have near-identical revenue trajectories according to Dealroom.

The threat neither controls: Anthropic launched a legal plugin for Claude in February 2026. Legal software stocks dropped on the announcement. Both CEOs waved it off. The market is less sure.

Why it matters: Legal AI is the live blueprint for every regulated vertical you're building in.

2/ Redpoint Says AI-Native Companies Are Breaking SaaS Unit Economics

[ENTERPRISE SAAS]

Redpoint’s 2026 market update — published this week in the newsletters we track — argues this AI cycle is backed by real demand, not hype, precisely because AI-native companies achieve faster revenue growth with materially smaller headcount. That breaks the core SaaS assumption: that revenue scales with people. Horizontal software is under existential pressure. Vertical software with proprietary data moats stays defensible.

Emergence Capital published a companion framework this week. The thesis: companies are shifting from selling software to owning outcomes. A new KPI is emerging in boardrooms: AI enablement spend per employee. Old GTM playbooks are being declared dead. Engineering is no longer the bottleneck; sales and marketing capacity at AI-native speed is.

Why it matters: If you're building vertical AI that competes for headcount budget, you're in the right frame.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

3/ Europe Just Got Its Proof-of-Concept Exit

[DEEP TECH]

Gleamer, Paris-based, founded 2017, radiology AI, was acquired by US imaging giant RadNet for €230M all-cash. The company had raised roughly €50M across three rounds in eight years. ARR CAGR exceeded 90% from 2022 to 2025. Expected $30M ARR in 2026. 700+ customer contracts across 44 countries. The exit creates the largest provider of radiology clinical AI solutions worldwide. That’s a 4.6x return on total capital raised.

The same week, Elaia Partners closed €254M across two funds: €120M for B2B startups from pre-seed to Series B, and €134M — double the previous fund — for deep tech seed in computing, industry, and life sciences.

Elaia’s Substack

Elaia Q1 2026 Newsletter

Welcome to Elaia’s Q1 2026 newsletter. This quarter was one of our most active on record: two fund closes, a new partner, a landmark portfolio exit, and a fusion energy deal that made headlines across Europe. We head into Q2 with more capital to deploy and a portfolio that continues to raise the bar…

Read more

a month ago · 1 like · Elaia

Built with institutional partners including PSL, INRIA, CNRS, and the Barcelona Supercomputing Centre. Also this week: Proxima Fusion signed a commercial fusion plant development agreement with Bavaria, RWE, and the Max Planck Institute - the first commercial fusion pathway beyond pure research.

The backdrop The Economist reported this week: tech talent flow is reversing from the US to Europe for the first time.

Why it matters: Gleamer is the template European founders have been waiting for.

4/ AI Scaled Past the Chips Bottleneck

[INFRASTRUCTURE]

Redpoint’s market update flagged energy infrastructure as the emerging AI constraint. The data behind that call: 70% of US transmission lines and transformers are over 25 years old. The ASCE gave US energy infrastructure a D+ rating. Morgan Stanley projects US data center demand could reach 74GW by 2028, with a 49GW shortfall in available power access. Gartner estimates 40% of AI data centers will face operational power constraints by 2027. Lead times for high-voltage transformers have stretched to 2–4 years.

The bottleneck has physically shifted. In 2021–2024 the constraint was hardware supply chains. By 2026 it’s grid access. A single AI training task uses up to 1,000 times more electricity than a standard web search. Data centers are projected to consume 6–12% of all US electricity by 2028.

Two signals from this week’s newsletters that connect directly: Drew Baglino, formerly Tesla’s SVP of Powertrain and Energy Engineering, founded Heron Power specifically to address industrial power electronics for affordable electrification at scale. And Proxima Fusion’s commercial fusion agreement isn’t just a science story — it’s an energy generation play that doesn’t depend on existing grid infrastructure.

Why it matters: if you're building anything touching power electronics, your buyer is now a data center.

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

Today Signal Question: What happens to your product when Anthropic ships a plugin for your category?

Foundation models are moving down the stack faster than most vertical AI roadmaps anticipated. The companies that are surviving it aren't doing so because their AI is better — they're surviving because they've built something the model itself can't replicate. So the question for your pitch isn't "is your AI good enough." It's: if Anthropic ships your use case tomorrow, what do you still have that they don't?

That’s the news. Now go build something.

— Byblos.digital

Subscribe now

Newsletters received: 20

Unique VCs publishing: 3

Period: 24 hours to 12:00 UTC

Sentiment breakdown: Bullish 50% · Neutral 25% · Mixed 20% · Bearish 5%

Top sectors by mention count:

• AI & Machine Learning (9)

• Infrastructure (8)

• Enterprise Software & SaaS (7)

• Payments & Stablecoins (6)

• DeFi (5)

Top trending projects: Ethereum, Solana, Visa (3 mentions each) · Claude Code, Figma, Polymarket, Meta, Stripe, Ripple, USBC, Canton, Tether (2 each)

The dominant narrative today is a structural split. Capital is flooding into AI infrastructure while traditional enterprise software gets repriced at decade lows. Two things can be true at once: AI is the best venture opportunity of the decade AND the first wave of AI is actively destroying valuations in the software layer beneath it. Today’s briefing is about both sides of that split, and what it means for founders raising right now.

1/ Coatue Bet $30B on Anthropic at $380B

[FUNDING]

Coatue co-led Anthropic's $30B Series G at a $380B valuation in February. A leaked LP presentation from January shows their math: $18B revenue in 2026, growing to $200B by 2031 with $48B EBITDA. At 41x forward EBITDA, that's a $1.995 trillion valuation by 2030, bigger than Apple today. The model assumes frontier AI is winner-take-most and compute costs compress fast enough to reach 24% margins at scale. One signal it might already be conservative: Anthropic is reportedly at $19B ARR now, ahead of Coatue's own year-end target.

Why it matters: AI infrastructure costs are dropping.

2/ The SaaSpocalypse

[ENTERPRISE SOFTWARE]

Revenue multiples went from 18.4x to 6x since September. The thesis driving it: AI agents replace human labor, so enterprises buy fewer licenses.

That creates a real problem for founders. Every software pitch gets sorted before you open your mouth. AI infrastructure looks resilient. Everything else looks like a seat count waiting to compress. The companies surviving it — Salesforce’s Agentforce at $800M ARR, Microsoft’s new AI-native pricing tier — all did the same thing: embedded AI into existing distribution instead of fighting it.

The infrastructure requirements: cloud dev environments, parallel workstreams, progressive agent access controls. But the insight that matters most came from Stripe's own framing of the project: the bottleneck was never coding speed. It was activation energy, that is, the friction of starting a task. Once that's removed, speed follows automatically. That reframes the whole developer tooling category.

Why it matters: Investors are already sorting you into a bucket before you pitch.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

---

3/ Banks Are Going On-Chain and Skipping Stablecoins

[PAYMENTS]

Four major US banks — Huntington Bancshares, First Horizon, M&T Bank, and KeyCorp — are deploying tokenized deposit products now. Visa just became a Canton Super Validator. Thunes integrated stablecoin payouts to 11,500 banks via Swift. USBC launched friend-and-family testing on a new Solana-based architecture with personhood-tied identity and push/pull payments. The EU is showing regulatory support for euro-denominated bank money on-chain. And a draft Congressional bill, the PARITY Act, would exempt regulated stablecoins from tax on gains, explicitly favouring institutional-backed digital money over permissionless alternatives. The institutional on-chain payments stack is being assembled quietly and it looks like bank-backed tokenized deposits, not DeFi stablecoins. This research is worth looking into if you want to find out more.

Why it matters: The institutional payments layer is built on bank-backed rails.

4/ Stripe Ships 1,300 AI Pull Requests Per Week

[AI AGENTS]

Stripe's internal AI coding agents "minions" shipped roughly 1,300 pull requests last week, all triggered via Slack emoji, all still human-reviewed. The infrastructure requirements: cloud dev environments, parallel workstreams, progressive agent access controls. But the insight that matters most came from Stripe's own framing of the project: the bottleneck was never coding speed. It was activation energy — the friction of starting a task. Setting up the environment. Finding the right context. Once that's removed, speed follows automatically. That reframes the whole developer tooling category.

Why it matters: The winning dev tool removes the reason people don't start.

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

Today Signal Question: AI agents are compressing enterprise seat counts. Incumbents are surviving by embedding AI into existing distribution — not by building new products.

If you’re an early-stage founder, you don’t have existing distribution yet.

So the question is: what do you build on top of when the floor is moving?

That’s the week. Now go build something.

— Byblos.digital

Subscribe now

Across our inbox today: newsletters from a16z crypto, Bankless, 0xResearch (Blockworks), Wintermute, Hex Trust, Animoca Brands, PANews, Apollo Crypto, and Social Leverage. The dominant themes showing up across multiple sources simultaneously:

• RWA / Tokenization — mentioned across a16z, Social Leverage, 0xResearch, Wintermute

• AI + Web3 / Agentic infrastructure — Animoca Brands, Bankless, Chamath, Apollo

• Prediction markets going institutional — 0xResearch, PANews, CoinDesk

• Ethereum L2 fragmentation fix — Bankless, 0xResearch, Ethereum Foundation

Weekly funding data (Mar 23–29): 15 disclosed deals, $767M+ total. The single biggest deal? ICE adding $600M to Polymarket. Strip that out and you’re looking at $167M across 14 deals, infrastructure and CeFi leading volume.

1/ a16z Says Founders Should Build What Wall Street Can't

[RWA / TOKENIZATION]

This week's a16z crypto essay is one of their most direct thesis statements in years. The argument: what's happening right now in capital markets is the largest infrastructure upgrade since electronic trading in the 1990s.

a16z crypto

Exchanges won’t build the next layer of finance. Founders will.

Wall Street isn’t just exploring blockchain anymore. It’s migrating to it…

Read more

a month ago · 56 likes · 8 comments · Jason Rosenthal

DTCC — the clearing house that processed $3.7 quadrillion in transactions in 2024 — received SEC authorization in December 2025 to tokenize real-world assets on approved blockchains, and is targeting a production-ready Treasury tokenization service for H1 2026. The NYSE announced a 24/7 on-chain trading platform in January. Tradeweb executed the first real-time on-chain US Treasury financing against USDC last August. And Nasdaq filed its own rule change with the SEC in September. a16z's conclusion: the incumbents are laying the regulated plumbing. They don't want to build the middleware, the compliance tooling, the distribution layer, the collateral management infrastructure, or the cross-border settlement interfaces. That's the founder's surface area. "Their margin is your opportunity to build the new rails."

Why it matters: RWA infra is the pitch framework for founders.

2/ ICE Just Put $1.6 Billion Into Polymarket

[FINTECH]

Read that again. Intercontinental Exchange, the company that owns the NYSE, has now deployed $1.6 billion into Polymarket across two tranches. The initial $1B came in October 2025. This week they exercised the remaining $600M. ICE also plans to purchase up to $40M in secondary stock from existing Polymarket shareholders. This is not an exploratory experiment. ICE operates core global financial infrastructure, exchanges, clearinghouses, data services, and they've decided on-chain prediction markets belong in that stack. Per PANews's weekly funding roundup, this was one of 15 disclosed deals in the week of March 23–29, representing the largest single allocation by a wide margin.

Why it matters: Founders of prediction/on-chain information markets have the most legitimate institutional signal.

All of this was pulled from the Byblos live feed - 600+ VC firms, 27500+ newsletters, one place.

See what else came in today at byblos.digital →

---

3/ Ethereum Is Fixing Its Own Fragmentation

[L2 INFRASTRUCTURE]

The first major announcement out of EthCC Cannes this week: the Ethereum Economic Zone (EEZ). Led by Gnosis and ZisK, backed by the Ethereum Foundation as "credibly neutral shared infrastructure," the EEZ is a framework for synchronous composability between Ethereum L1 and L2s.

The problem it's solving: your Arbitrum liquidity can't talk to your Base contracts without bridging. Every L2 right now is an island. The EEZ would let smart contracts on different chains atomically compose in a single transaction: no bridge required, no fragmented liquidity. Mechanically, it works via ZK-proven rollups using proxy contracts that let a contract on one chain represent a contract on another, with atomic state changes applied across both.

The whole thing runs without any Ethereum protocol changes. It's built on smart contracts and real-time ZK proving. The result, if it works: single-chain UX with multi-chain execution. Gnosis has a strong track record — CoW Protocol, Safe multisigs, the conditional tokens framework underlying Polymarket. ZisK's zkVM handles real-time proving. Ethereum Foundation is funding it as open source for all.

Why it matters: If the EEZ delivers, you can deploy once and access all Ethereum liquidity. But L2s that opt in must re-org when Ethereum re-orgs.

4/ Animoca's Chairman: Blockchain Is the Economic Substrate for AI Agents

Yat Siu, co-founder and executive chairman of Animoca Brands, published a long-form essay this week titled "Sovereign Multitude: The Internet After Users."

His argument: the primary actors on the internet are no longer going to be humans. AI agents — persistent, goal-directed software entities that can perceive, decide, and transact — are replacing the human user as the web's main demographic. The key implication for web3: agents need wallets, payment rails, identity, reputation, and programmable assets to transact autonomously. That's exactly what blockchain provides. Without blockchain, agents are still dependent on centralized intermediaries at every step. With blockchain, they become genuine economic actors, able to receive payment, commission services, manage capital, and settle with other agents entirely on-chain.

Siu names DeFAI (DeFi + AI), Bittensor, Autonolas, and Moca Network as examples of the emerging agentic economy. He also frames new roles that will matter: Agent Economist, Agent Orchestrator, Agent Governor, and argues that companies that hire for these now will have a decisive early advantage.

Today Signal Question: Wall Street is your customer now. They’re laying the foundation. They won’t build what runs on top.

If you knew your first real enterprise customer was a bank, a clearinghouse, or a stock exchange, what would you build?

Everything in today’s digest came from the Byblos live feed. The full newsletter archive, the VC directory, and tomorrow’s signal are all waiting for you at Byblos.digital - the investor interned, curated.

That’s the week. Now go build something.

— Byblos.digital

Subscribe now