In this week's issue
Anthropic Just Put a $895 Billion Bill on the Table
Every limited partner (LP) with real AI exposure, and every European general partner (GP) whose portfolio companies buy Claude, build on it or compete next to it, should sit with Anthropic's amended IPO filing this week.
Not because the revenue chart is pretty. Because it puts a price on the question behind half the venture stories in London right now: can the labs that own the models survive the cost of the intelligence they sell?
Anthropic has filed the first amendment to its IPO registration statement (Form S-1). The numbers that matter:
About $895.4 billion in purchase commitments and leases, "substantially all of which is related to compute and infrastructure." Roughly $447.5 billion of that lands before the end of 2030, and about $69.8 billion is due in the next 12 months.
$17 billion of revenue in the latest quarter, up from about $1.4 billion a year earlier. That is growth of 1,127% year on year, according to the filing.
$9.9 billion spent on compute and infrastructure in that same quarter alone.
Credit to Sara at @sarainwondertech, whose reels pushed me into the filing. Her "almost $900 billion" shorthand holds up against the document.
Anthropic also claims that knowledge work "represented more than $40 trillion in economic value globally in 2025." That is their own estimate of the market they are chasing, not a fact to underwrite.
See below to continue…
Invest Alongside a $12B Owner/Operator
Choose individual real estate investments backed by a $12B owner/operator that invests 20%+ alongside individual LPs in every offering.
Accredited Investors only. $100k minimum investment.
This communication is for informational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any securities. Any such offer or solicitation will be made exclusively through the definitive offering documents. All investments involve risk of loss, including the potential loss of principal. Past performance is not a guarantee of future results. Any targeted returns or projections are forward-looking statements, are based on current assumptions, and are not guarantees of future performance. Actual results may differ materially.
Continue…
OpenAI had the head start. Anthropic filed the numbers.
Sara's second reel asked the question every LP should be asking: OpenAI had the brand, so how is Anthropic posting numbers like these?
Focus. Anthropic says that shifting investment toward consumer products to compete for those users "would require significant time and resources beyond our current plans." The company is built around enterprise customers, with coding as the way in. Claude Code, the filing says, "most concretely defined what autonomous AI work looks like."
The figures from the filing worth printing out:
About 900% net revenue retention - existing customers now spend roughly nine times what they did in the comparison period (excluding fixed-price Claude Pro, Max and Ultra subscribers).
In the latest quarter, Claude Enterprise users who didn't use Claude Code spent under $75 a month on average. The top 5% of Enterprise users who did use it spent about 20 times that.
More than 11,900 customers paid $100,000 or more over the last twelve months, more than 1,900 paid $1 million or more, and around 250 paid $10 million or more.
The bet is selling intelligence into work, especially code, not selling chat to consumers. That shows up in the revenue. It also shows up in the costs.
Margins are not a free option
TechCrunch went straight to the cost line. Compute rose to about 58% of revenue in the latest quarter, up from 48% the quarter before. Anthropic bought $5 billion of short-term compute that, in its own words, "currently costs significantly more" than its multi-year deals, and that pushed it to a $567 million operating loss after an operating profit the quarter before (Business Insider).
Price cuts and discounts took about $600 million off revenue growth between the two quarters. Quarter-on-quarter growth slowed to around 47%, after 144% and 132% in the two quarters before (Data Center Dynamics). And nearly $70 billion of commitments fall due within a year, against about $132 billion of cash and short-term investments.
Right thesis. Real demand. Unit economics still being negotiated in public, every quarter. As I wrote in my last issue, having the right AI thesis is not the same as surviving the hold. Now we're watching the same lesson play out at company level.
What UK and European LPs and GPs should do
1. Treat frontier-lab exposure as a bet on capital structure. The demand is real. The question is who pays for the compute and when. If you hold this risk directly or through a fund, ask for the commitment schedule, the mix of short-term versus multi-year compute, and how revenue holds up if token prices fall by 20-40%.
2. Use the filing to sort your application-layer portfolio. The bull case is companies building on Claude or GPT and selling the finished work - agents, copilots, automated workflows. That 20x spending gap for Claude Code users is the proof. The bear case is anyone trying to compete on raw intelligence against a lab sitting on $895 billion of compute. Back the companies selling the work, not another thin model layer.
3. European fundraising is still a barbell, not a mid-market. Look at what closed in the last few weeks: Creandum at €668 million and Seaya at €600 million at one end, Evertrue and Robin Capital at €15 million each and Octopus Ventures' new pre-seed fund at the other, with AP Ventures and Entourage raising for industrial and physical AI. If your "balanced" European book still assumes the stage maths of the last cycle, these closes are the rebuttal.
✅ Know an LP who underwrites AI apps as if compute were someone else's problem? Forward this their way. The filing made the cost of intelligence a portfolio input, not a footnote.
New Funds Across Europe
Creandum has closed €668m for Creandum VIII, its largest fund to date, to back around 30 pre-seed to Series A tech companies over two years!
Seaya has reached a €600m first close for Seaya Nova, writing €20-50m cheques into European Series B to D companies, and is expanding into London!
AP Ventures has closed Fund III at $300m to back early-stage industrial and energy technology out of London!
Entourage has closed Fund II at €100m, beating its €75m target in about two weeks, for pre-seed and seed AI, robotics and deep tech!
BackingMinds has reached a €50m first close for its third fund, anchored by the European Investment Fund, to back early-stage Nordic startups!
Evertrue Capital has closed a €15m debut fund backed by the British Business Bank and Phoenix Court, writing £250k-£500k pre-seed cheques into UK and European frontier tech!
Robin Capital has closed a €15m second fund as a Berlin solo GP backing AI, robotics, space and deep tech!
Octopus Ventures has announced the first close of First Cheque II, targeting more than twice the £10m of its first pre-seed fund, and has taken on OnStage's pre-seed portfolio!
✅ We also share this list in our founder-focused newsletter. Want to be featured next time? Send me a note.
Upcoming Events: deal-flow and LPs
Rare Founders
The Legendary Startup Christmas Party
Our biggest night of the year. Founders, investors and operators in one room, and the easiest place to see most of London's early-stage scene before the year closes.
In-person event
The Ministry, Borough, 79-81 Borough Rd, London SE1 1DN, UK
Tue 8 Dec, 18:00 - 00:00 GMT
Other London events - see the full list here.
💌 Enjoyed this?
If you found this useful, share it with another investor who should be in the loop.
Win Reward
Vasily Alekseenko
Founder
@ Rare Founders
Rare Founders - building the bridge between founders and investors via regular in-person and online events, meetups, conferences.