MeanCEO: Tech Startups and Startup Ideas

Anthropic's Leaked IPO Filing: $16.3 Billion in Revenue, an $80.7 Billion Loss and a Compute Squeeze. What It Means if You Build on Claude

TL;DR: What does Anthropic's leaked IPO filing say, and should you worry if you build on Claude?

On 20 October 2026 the Wall Street Journal reported on a draft of Anthropic's IPO filing. It shows $16.3 billion in revenue in the first half of 2026, up from $1.1 billion a year earlier, and an $80.7 billion net loss, about $72 billion of which is accounting charges tied to how Anthropic raised money. The more useful detail for founders: Anthropic can't get as much cloud capacity as it wants from AWS, Google and Microsoft, and two customers bring in about a fifth of its revenue. Claude isn't going anywhere. But capacity limits, price moves and changing priorities are now a documented risk. Keep a second model tested, cap your spend, and grab the cheap prices and credits while they last.
Most AI companies keep their numbers private. Anthropic is about to lose that luxury: it's preparing to go public, and the drafts of its IPO filing keep leaking.
If your product calls Claude, this isn't just Wall Street gossip. The filing describes the company you depend on: how fast it grows, where its money goes, and what it struggles to get. Here's what was reported, what's confirmed and what it means for your stack.

What happened

On Tuesday 20 October 2026, the Wall Street Journal reported on a draft prospectus for Anthropic's initial public offering, the S-1 filing a company must make before listing its shares. Anthropic confidentially filed a draft with the US Securities and Exchange Commission in early June, so the public can't read the document itself yet. The figures come from the Journal's reporting, as relayed by TechCrunch, Reuters and Investing.com.
The headline numbers for the first six months of 2026, which ended on 30 June:
  • Revenue: $16.3 billion, up from $1.1 billion in the same period of 2025.
  • Net loss: $80.7 billion, up from $13 billion a year earlier. Reuters says about $72 billion of it came from accounting charges tied to changes in the value of financial instruments Anthropic issued to raise money. That part is a non-cash, paper loss.
  • AI infrastructure spending: almost $9 billion, nearly four times the year before, and more than the $7.3 billion Anthropic spent on it in all of 2025, according to Investing.com's summary.
  • Customer concentration: two customers accounted for about a fifth of first-half revenue, and the company expects a limited number of customers to keep generating a large share of sales.
Anthropic didn't respond to requests for comment from Reuters or TechCrunch, and Reuters said it couldn't immediately verify the report. Treat every number here as reported, not confirmed: these come from a draft, and drafts change before a filing goes public.

The numbers in one table

WhatReported figureFor comparison
Revenue, first half of 2026$16.3 billion$1.1 billion in the first half of 2025
Net loss, first half of 2026$80.7 billion$13 billion a year earlier
Of which: accounting charges on financing instrumentsAbout $72 billionNon-cash
AI infrastructure spend, first half of 2026Almost $9 billion$7.3 billion in all of 2025
Noncancelable cloud compute dealsAbout $450 billion over roughly ten yearsGoogle Cloud $111bn, AWS $110bn, Microsoft $31bn
Possible future compute spendingUp to $750 billionIncludes deals still being negotiated
Revenue from the two biggest customersAbout 20%Of first-half revenue
Revenue, third quarter of 2026$16.8 billionUp 45% on Q2's $11.6 billion
Sources: draft S-1 as reported by the WSJ, via TechCrunch, Reuters and Investing.com (20 Oct 2026); Q3 figures from Reuters (15 Oct 2026).

Growth is huge, but it's slowing

A week before the leak, Anthropic held an investor day in San Francisco. According to Reuters, an investor who attended said the company reported third-quarter revenue of $16.8 billion, up more than 45% from $11.6 billion in the second quarter, after $4.7 billion in the first. Anthropic also told investors it was profitable on an adjusted basis in the quarter.
Up 45% in a quarter would be a dream for any startup. For investors who expected even more, it wasn't enough. Reuters reported that investors at the event discussed a valuation of around $1.5 trillion, a big step down from the roughly $2 trillion previously expected. 22V Research analyst Dauvin Peterson wrote on 18 October, as quoted by CNBC, that the revised range "was viewed as disappointing". He also pointed to questions about slower growth in September and about pricing pressure on large language models.
That last point is the one to remember. Pricing pressure for Anthropic is cheaper tokens for you, and the market is already pushing in that direction. We tracked the cuts in our AI price war roundup.

The real story for builders: a compute squeeze

TechCrunch read the leak as a capacity story. According to the filing, Anthropic is having trouble getting enough cloud capacity from AWS, Google and Microsoft, which increasingly see it as "a competitor rather than a customer". Those three are Anthropic's suppliers, investors and sales channels, and they also sell rival AI models.
Anthropic's answer is to lease its own data centres and work with third-party operators. The filing worries those operators have a "limited operating history", so they could be less reliable than the big clouds. On 20 October, CNBC reported that Anthropic is in talks with Nscale, a UK-based neocloud, about a second multibillion-dollar deal to lease computing capacity in the US. Anthropic declined to comment. CNBC also noted that Anthropic said earlier this year that demand for Claude had caused "inevitable strain" on its infrastructure.
This isn't new, just better documented. When Reuters saw the first draft in September, the filing already said that relying on a few partners and suppliers "could give rise to conflicts of interest and adversely affect our access to compute". That same draft showed that 47% of Anthropic's 2025 sales went through Amazon's and Google's cloud marketplaces, and that the US made up nearly two-thirds of sales.
If you've seen errors or tight limits on Claude, the capacity story explains part of it. Claude's own status page logged several incidents in late September and early October, including elevated errors across claude.ai, Claude Code and the API on 29 September. A company that can't buy all the compute it wants has to ration it somewhere.

Why it matters if your product runs on Claude

None of this means Claude is going away. Anthropic made $16.8 billion in a single quarter, and its biggest problem is too much demand. But a filing like this tells you where the pressure points are.
What the filing reportedly showsWhat it can mean for your productWhat to do
Not enough cloud capacity from the big threeRate limits, errors at peak times, slower access to new modelsBuild retries and a fallback model into your app
Two customers bring in about 20% of revenueThe biggest accounts get the most attention when capacity is tightDon't assume your small API account gets priority
Pressure on revenue growth and pricingToken prices keep falling; plans and limits keep changingRecheck your model choice every quarter
Up to $750 billion of compute plansHuge fixed costs push the company to monetise harder after the IPORe-read pricing, terms and usage limits after the listing
Liability for harmful AI behaviour could exceed insuranceUsage policies may get stricterKeep your own logs and human checks for risky actions
Sources: draft S-1 as reported by TechCrunch, Reuters and Investing.com; CNBC, 19 October 2026. The middle and right columns are our analysis.
Replit has already shown what switching looks like: in our story on Replit making OpenAI the default for its agent, a company whose CEO says it was one of Anthropic's biggest customers switched its agent's default to OpenAI, while keeping Claude selectable. That's the right model for a small team too: one default, one tested fallback, and a router in between.

The good news: Claude is cheaper than ever right now

A company preparing to list wants growth, and that's showing up in prices and perks. On 7 October Anthropic launched Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and says it costs around 75% less to run than Haiku 4.5. The same day it halved the price of Sonnet 5.5 cache reads to $0.10 per million tokens. Haiku 5.5 is available on Anthropic's own platform and on AWS, Google Cloud and Microsoft Azure.
On 6 October Anthropic also expanded its Claude Startups programme, CNBC reported. Members can get up to $45,000 in discounts and credits on startup tools through its Claude Startup Stack, a one-time $1,000 API credit and a free year of Claude Team for up to five premium seats. Startups founded within the past five years, or funded within the past two, can apply. "Startups are often the first to push Claude to its limits," said Beth Robertson, Anthropic's head of startups.
For a bootstrapped team, that's real money. Just don't build your unit economics on the assumption that today's prices, credits and limits will hold after the IPO.

The European angle

  • Europe is the smaller market. In the September draft, the US accounted for nearly two-thirds of Anthropic's 2025 sales. When capacity is short, the biggest market and the biggest customers tend to come first.
  • Most European teams buy Claude through a cloud. Nearly half of Anthropic's 2025 sales went through AWS and Google marketplaces. If you buy Claude through your cloud provider, check which region your requests run in and what your data processing agreement says. Our guide on how European startups use AI without breaking GDPR walks through the checks.
  • European compute is part of the story. Nscale, the neocloud Anthropic is talking to, is British, but the capacity it discussed is in the US. CNBC also reported that Microsoft pulled out of a deal to anchor an Nscale data centre near London, amid delays connecting to the UK's power grid. Building AI capacity in Europe is still slow.
For more on the people side of Anthropic's growth, read our story on how Anthropic hires fewer juniors as AI agents do more work. Our AI news for October 2026 page tracks the rest of the month, and the startup funding news for October 2026 roundup follows the money.

What this means for you

  1. Test a second model on your real prompts. Not a benchmark, your own tasks. Keep the results, so switching is a config change and not a rewrite. Our checklist for evaluating AI vendors helps you compare them.
  2. Cap your spend. Set monthly limits and alerts on every API key, and rotate keys you no longer use. The OpenCode breach that exposed saved API keys showed how fast an unguarded key turns into a bill.
  3. Use the cheap tier. Move classification, extraction and simple support replies to Haiku 5.5, and turn on prompt caching for long, repeated context. That's where the October price cuts pay off most.
  4. Take the credits. If you were founded in the last five years or raised money in the last two, apply to Claude Startups. Then plan for what happens when the credits run out.
  5. Keep it portable. Store prompts, evals and tool definitions in your own repo, not only inside one vendor's console.
  6. Watch the IPO. Anthropic is reportedly aiming to list in November. When the filing becomes public, check the final numbers and any new risk factors, and re-read your plan's pricing, terms and limits.
Are you getting customers from AI search yet?

More and more traffic is coming from Perplexity, ChatGPT, Grok and other AI tools.

👉 Rank on ChatGPT

FAQ: Anthropic's IPO filing

How much revenue does Anthropic make?

According to the WSJ's report on its draft IPO filing, Anthropic made $16.3 billion in revenue in the first half of 2026, up from $1.1 billion a year earlier. Reuters reported that Anthropic told investors its third-quarter revenue was $16.8 billion, up 45% from $11.6 billion in the second quarter.

Why did Anthropic lose $80.7 billion?

Most of it isn't cash. Reuters reports that about $72 billion of the first-half net loss came from accounting charges tied to changes in the value of financial instruments Anthropic issued to raise money. The company also spent almost $9 billion on AI infrastructure in the same six months.

When is Anthropic's IPO?

No date has been announced. Anthropic confidentially filed a draft with the SEC in early June 2026, and press reports say it is targeting a listing in November. Reuters reported that the debut is likely to come after the US midterm elections in early November.

Is Anthropic's filing public?

Not yet. The figures come from drafts seen by the Wall Street Journal, Reuters and the Financial Times. Anthropic didn't comment on the 20 October report, and numbers can change before the final filing is published.

Will Claude get more expensive after the IPO?

Nobody knows. Right now prices are falling: Haiku 5.5 launched on 7 October at $0.10 per million input tokens, and Sonnet 5.5 cache reads were halved. But a listed company with huge compute commitments will be under pressure to improve margins, so don't build your pricing on today's rates.

Should I stop building on Claude?

No. The filing shows strong demand, not a company in retreat. The lesson is to avoid depending on any single model provider: keep a second model tested on your own prompts, use a router, and set spend caps on every key.

About the Author

Violetta Bonenkamp, also known as MeanCEO, is an experienced startup founder with an impressive educational background including an MBA and four other higher education degrees. She has over 20 years of work experience across multiple countries, including 5 years as a solopreneur and serial entrepreneur. Throughout her startup experience she has applied for multiple startup grants at the EU level, in the Netherlands and Malta, and her startups received quite a few of those. She’s been living, studying and working in many countries around the globe and her extensive multicultural experience has influenced her immensely.
2026-10-21 05:28 News and Trends