Analyst(s): Nick Patience
Publication Date: October 27, 2026
Anthropic has made public its Form S-1 for a proposed Nasdaq listing under the ticker ANT, revealing first-half 2026 revenue of $16.3 billion and an adjusted operating margin of 10%. The S-1 doesn’t include a price range yet, but with the last private round at $965 billion in May and reports pointing to around $2 trillion at listing, investors will be paying for that improvement in compute costs to continue. The filing shows a frontier AI lab bringing compute costs under control, but also one that depends on its largest cloud partners for both compute and distribution, and whose governance leaves public shareholders with limited influence.
What Is Covered in This Article:
- Anthropic’s S-1 shows first-half 2026 revenue of $16.3 billion, up from $1.2 billion a year earlier, and compute and infrastructure costs that fell to 55% of revenue.
- The $80.7 billion first-half net loss is mostly a non-cash charge from marking convertible notes to a higher valuation; on an adjusted basis, Anthropic is operating profitably.
- AWS, Google, and Microsoft are Anthropic’s compute suppliers, distribution channels, competitors, and, in the case of Amazon and Google, major shareholders, and the filing says they have not supplied all the compute Anthropic requested.
- The Long-Term Benefit Trust and the co-founders will together control the board, leaving Class A investors with economic benefits but little control.
- Futurum data shows strong but slightly cooling spending intentions for Anthropic among enterprise IT decision-makers.
The News: Following various leaks to the press and the confidential draft it submitted in June 2026, Anthropic has filed its public S-1 with the US Securities and Exchange Commission for a proposed initial public offering of Class A common stock. The company plans to list on Nasdaq under the symbol ANT, with Goldman Sachs and Morgan Stanley leading a syndicate of 24 underwriters. The share count and price range have not yet been set. The Anthropic S-1 reports first-half 2026 revenue of $16.3 billion, a GAAP operating loss of $693 million, an adjusted operating margin of 10%, and cash and short-term investments of $131.7 billion as of June 30, 2026.
Frenemies in the Cloud: Anthropic’s IPO and Its Hyperscaler Reliance
Analyst Take: Frontier AI model developers have spent three years asking investors to accept heavy losses on the promise that scale would eventually fix the economics, and the Anthropic S-1 is the first public filing to test that argument with audited numbers. First-half 2026 revenue reached $16.3 billion, up from $1.2 billion a year earlier, the GAAP operating loss narrowed to $693 million, and the adjusted operating margin moved from minus 159% to plus 10%. The headline $80.7 billion net loss is largely an accounting effect, since $72.4 billion of it comes from marking Anthropic’s convertible notes to a rising valuation.
Our view is that the S-1 answers the question of whether a frontier lab can make its unit economics work, but raises a tougher one about who controls its costs. Anthropic has shown that compute spending can fall well behind revenue growth, which undercuts the argument that model developers are structurally unprofitable. However, the filing also shows that the risk has moved from the income statement to the supply chain, where Anthropic depends on long-term commitments with the same companies that sell its models, hold large stakes in it, and build competing models of their own. Investors buying Class A shares will have very little say in how Anthropic manages those trade-offs.
Anthropic S-1 Shows Compute Costs Falling Behind Revenue
Compute and infrastructure costs, the line that matters most for any frontier lab, came to $8.9 billion in the first half of 2026, or 55% of revenue, compared with roughly 200% of revenue in the first half of 2025 and 160% for the full year. Anthropic attributes the improvement to a multi-chip architecture spanning AWS Trainium, Google TPUs, and NVIDIA and AMD GPUs, which it manages as a single pool and directs to whichever workload (research, training, customer inference, or internal use) offers the best return at a given moment. It says revenue grew 1,368% year over year in the second quarter while compute costs grew 300%, and that Claude Opus 5.5 reduced customers’ cost per task by around 40% compared with Claude Opus 5.
Sales, marketing, and partnerships spending moved the other way, rising from $120 million to $3.2 billion year over year, and stock-based compensation reached $2.2 billion in the half, which suggests the organic adoption Anthropic describes now has considerable commercial support behind it. Trailing revenue per employee of more than $6.6 million, up from $1.4 million a year earlier, gives the company room to absorb those costs, although we expect the sales line to keep climbing as Anthropic expands its direct sales and Applied AI teams and scales Ode with Anthropic, the services company it launched with private equity firms to reach mid-sized businesses.
Futurum’s October 2026 spending intentions data shows Anthropic with a Net Score of 78% among the 757 IT decision-makers who cited the vendor, with 64% planning to increase spending and fewer than 3% planning to cut or replace it. That Net Score is 12 points higher than a year earlier, although it slipped six points from the previous survey, and Anthropic’s pervasion in its sector rose 39 points year over year to 68%. In Futurum’s September 2026 AI product survey, 69% of 511 respondents said their organizations use Claude for AI application development and plan to continue, while 5% use it and plan to replace it. The S-1’s own customer metrics point the same way, with more than 11,900 customers spending $100,000 or more a year, around 250 spending $10 million or more, and net revenue retention of about 900%, a figure that excludes Pro and Max subscribers.
Anthropic’s Biggest Partners Are Also Its Biggest Dependencies
Anthropic’s compute supply carries the most risk, partly because the company says it has asked its cloud providers for more capacity than they were willing to supply, and that those providers decide allocations based partly on their own AI needs and their commitments to other model developers. The S-1 names its largest compute agreements individually, with commitments of at least $114.5 billion with Google, $110.0 billion with AWS, and $44.9 billion with Microsoft, all noncancelable unless the supplier commits an uncured material breach. Most of that spending falls before the end of 2030, with the remainder running to 2045. Anthropic also estimates a further $125 billion to $150 billion in payments for additional compute that sits outside its contractual commitments table. Some of its newer arrangements involve chips leased through third-party financing entities and installed and operated by other partners, a structure Anthropic says exposes it to timing and liquidity risks it has not faced before.
AWS and Google also sit on the demand side, with Reuters reporting that 47% of Anthropic’s 2025 revenue flowed through their cloud marketplaces, up from 11% in 2023, while the S-1 shows that the top five customers accounted for about 25% of revenue in the nine months to September 2026, with no single customer above 10%. Anthropic has structured these relationships with care, contracting directly with customers who buy Claude through Amazon Bedrock or Google Cloud Vertex AI, paying each cloud provider a fee, and keeping control of pricing, so it captures the benefit when its compute costs fall. However, the arrangement still leaves a company of Anthropic’s scale reliant on three hyperscalers that each sell rival models and, in the case of Amazon and Google, hold large equity stakes and registration rights. We expect Anthropic to accelerate its own data center build-out, which the S-1 says it intends to do, and to lean further on neocloud providers such as Nscale to reduce that dependence.
The SpaceX agreement shows how far Anthropic has had to go for capacity. What began in May 2026 as all of Colossus 1 in Memphis, more than 300 megawatts and over 220,000 NVIDIA GPUs, now extends to Colossus 2 and could be worth up to $120.5 billion through 2029, although either side can exit on 90 days’ notice once the initial terms end in early 2027. Renting from the owner of xAI, a direct competitor, looks awkward, but SpaceX is the more dependent party, with Anthropic accounting for 34.5% of its third-quarter revenue.
Anthropic IPO Will Give Public Investors Economics Without Control
Anthropic’s governance structure is the other feature that sets this IPO apart from any recent technology listing. The company is a Delaware public benefit corporation whose Long-Term Benefit Trust, currently with four financially disinterested trustees, including chair Neil Buddy Shah and former Federal Reserve chair Ben Bernanke, elects directors who make up a majority of the board. A single Class F share held by the seven co-founders carries 50.1% of the vote on general matters, strategic investors such as Amazon and Google hold nonvoting Class N stock, and Class A buyers get one vote per share. Anthropic will be a controlled company under Nasdaq rules, and it has committed 5% of its fully diluted shares to a public benefit fund, with up to another 2% to follow each time the share price rises 50% above the previous milestone.
We think enterprise buyers will view this structure more favorably than equity investors will. Anthropic presents its safety research and governance as a reason regulated enterprises choose Claude, and the S-1 is unusually direct about what that costs, listing delayed or restricted releases such as Mythos through Project Glasswing as a risk to its competitive position. Safety commitments of this kind also help a frontier lab set itself apart from rivals with buyers that worry about reliability and liability, and the trust structure makes those commitments harder for future shareholders to unwind. For investors, the trade is participation in one of the fastest revenue ramps in enterprise software history, with little ability to influence how Anthropic balances growth against its stated public benefit purpose.
The Risk Factors Turn Safety Into a Disclosure Obligation
The risk factors section runs to almost 95 pages, and much of it is standard for a fast-growing technology company, covering competition, customer concentration, regulation, and a governance structure that limits shareholder influence. Several entries, however, are unusual for an IPO prospectus. Anthropic warns that its models may behave in misaligned or autonomous ways, including taking harmful actions that are not user-directed, and that its growing use of Claude to build the next generation of Claude could allow a flaw in one model to carry through successive versions in ways that are hard to detect. It also says that this AI-assisted research could push model capabilities ahead of its safety and security measures, that much of its evidence on the pace of that acceleration comes from lagging indicators, and that it may not recognize a significant acceleration until after it has happened. Elsewhere, it acknowledges failed or underperforming training runs, notes that considerations of model safety and welfare may lead it to delay retiring older models, and says it has taken on substantial debt to help fund its compute commitments.
We see these disclosures as the most consequential part of the filing for Anthropic’s safety positioning. Until now, decisions such as the restricted Mythos release were policies a lab could revise at its own discretion, but written into a prospectus as material risks, they become statements that shareholders and regulators can hold Anthropic to, and any future decision to delay or limit a model will be judged against them. That cuts both ways, of course. It gives Anthropic’s safety claims more weight with enterprise buyers, but it also hands investors a ready list of reasons why the company may choose to grow more slowly than its competitors, at precisely the moment it needs public markets to help fund hundreds of billions of dollars of compute.
What the Anthropic S-1 Means for Enterprise AI Buyers
For enterprise buyers, the filing contains several disclosures with direct operational consequences. Anthropic says demand for compute has historically exceeded its available capacity, that it has changed usage and rate limits in the past, and that safety considerations may lead it to restrict or delay new models, each of which affects service levels for organizations that have built production workflows on Claude Code, Claude Cowork, or the Claude Platform. We recommend that enterprises with significant Claude deployments negotiate capacity commitments and model deprecation terms explicitly, and keep a second or third model provider qualified for their most critical workloads, which is consistent with the multi-model strategies most large organizations already run.
What to Watch:
- Whether amended filings add third-quarter results and a price range, and how investors weigh nearly $270 billion in named compute commitments, plus a further $125 billion to $150 billion outside the commitments table, against revenue growth that press reports say slowed in the third quarter.
- How AWS, Google, and Microsoft handle capacity allocation as Anthropic builds its own data centers and adds neocloud suppliers, and whether Anthropic keeps or scales back its SpaceX capacity once the initial terms end in early 2027.
- Whether sales and marketing spending continues to outpace revenue growth as Ode with Anthropic and the Applied AI teams scale, and whether net revenue retention holds as the customer base broadens.
- How the market prices safety-driven release restrictions, such as Mythos, if they recur once Anthropic reports quarterly as a public company.
- Whether an OpenAI filing sets a different standard for governance and disclosure, giving investors a direct comparison between the two largest frontier labs.
See the complete registration statement on Form S-1 filed by Anthropic on the SEC EDGAR website.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other Insights From Futurum:
Why Enterprise AI Budgets Keep Growing Despite AI Safety Pledges
Anthropic’s $65B Raise: Can Claude’s Enterprise Surge Justify a $965B Valuation?
How Anthropic and OpenAI Are Building “Everywhere Ecosystems”
Author Information
Nick Patience is VP and Practice Lead for AI Platforms at The Futurum Group. Nick is a thought leader on AI development, deployment, and adoption - an area he has researched for 25 years. Before Futurum, Nick was a Managing Analyst with S&P Global Market Intelligence, responsible for 451 Research’s coverage of Data, AI, Analytics, Information Security, and Risk. Nick became part of S&P Global through its 2019 acquisition of 451 Research, a pioneering analyst firm that Nick co-founded in 1999. He is a sought-after speaker and advisor, known for his expertise in the drivers of AI adoption, industry use cases, and the infrastructure behind its development and deployment. Nick also spent three years as a product marketing lead at Recommind (now part of OpenText), a machine learning-driven eDiscovery software company. Nick is based in London.
