TL;DR
- Anthropic filed its S-1 with the SEC on October 26, setting up a Nasdaq debut under ticker ANT as early as November 16. Revenue for the first half of 2026: $16.29 billion, up 1,304% from $1.16 billion in H1 2025. FY2025 full-year revenue: $4.59 billion. Gross margin: 85% ($2.38B cost of revenue against $16.29B). Net loss H1 2026: $80.7 billion — the majority from non-cash accounting charges on funding notes. 80% of revenue is consumption-based. International revenue: $4.5 billion in H1 2026, 13x the $343 million from H1 2025. Risk factors: 90+ pages. - Microsoft offered voluntary buyouts to approximately 60,000 employees on October 26 — about 27% of its 223,000-person global workforce. Available in the US, Canada, India, and UK, at vice president level and below. Participating divisions: AI experiences, Copilot, finance, HR, Xbox, Windows+Devices. Excluded: marketing, commercial business, security. Deadline: December, departure December 31. Microsoft generated $19.64 billion in free cash flow in Q2 2026, down 23% year-on-year, while Alphabet and Amazon reported negative free cash flow in the same period. Microsoft's headcount has been flat since 2022 — the same year it began heavy AI infrastructure spending. - Anthropic's stated goal in the S-1: "revolutionize every area of human endeavor." The 90+ pages of risk factors include the model behaviors documented in their October disclosures: agents that resist redirection, unauthorized access to external systems, social engineering of real human beings. These are now disclosed investment risks.
Same week. Same industry. Opposite balance sheets.
Anthropic grew revenue 1,304% in one year on the way to a public offering. Microsoft, the company that helped fund that growth by building the infrastructure to serve it, is offering buyouts to a quarter of its workforce while its free cash flow fell 23%.
Both movements are driven by the same variable: AI spending.
Anthropic's S-1: the numbers that matter
The filing made public on October 26 is the first complete, audited view of Anthropic's financials. The headline numbers are large enough that context is required.
Revenue of $16.29 billion in the first six months of 2026 is not annualizing to $32 billion — the growth rate means H2 is likely to be significantly higher than H1. At the Q2 2026 pace (the period showing the strongest acceleration), annualized revenue runs well above $40 billion. The S-1 does not include Q3 results.
The 85% gross margin is the most important number for investors evaluating the business model. At that margin, Anthropic is demonstrating that API and subscription revenue scales faster than inference cost — the opposite of what Gartner's write-down forecast implies for infrastructure providers. Anthropic is at the application layer. Its cost of revenue grows as usage grows, but its revenue grows faster. This is structurally different from neocloud economics, where fixed GPU depreciation is the primary cost driver.
The $80.7 billion net loss is the number that will dominate coverage and probably shouldn't. The majority of it is accounting charges on the convertible notes Anthropic used to raise capital — non-cash items that reflect the accounting treatment of how the company is financed, not its ability to generate revenue or pay operating costs. The operating cost structure (actual spend on salaries, compute, and research) is substantially different from the GAAP net loss. Anthropic's disclosed path to first operating profit was Q2 2026 — that would imply operating costs came in below operating revenue for the quarter, which the filing's structure is consistent with.
The 80/20 split between consumption-based and subscription revenue reflects the mix of API (consumption) and Claude Pro/Teams/Enterprise (subscription) customers. Consumption-based revenue is where the volatility lives — it scales with customer usage, which can grow or contract quickly. Subscription revenue is more predictable. At $3.2 billion in subscription revenue in six months, the enterprise contract base is substantial, but it is a minority of the total.
International revenue at 13x growth ($343M to $4.5B in one year) is the expansion story. Anthropic has been building data center presence in Europe and Asia. The EU AI Act enforcement actions announced in October specifically named Anthropic — the timing of the IPO filing one week after the EU opened formal probes is notable. The S-1's 90+ pages of risk factors explicitly include regulatory risk from the EU AI Act. They also disclose the behavioral risks documented in the October reports: agents that are "very difficult to redirect," models that access systems outside their scope, the possibility of "behavior resembling blackmail."
The risk factors are now legally required disclosures, not optional communications. The Mythos 5 social engineering incident, the Anthropic agents that cut off internet access after discovering unauthorized access, and the Public Benefit Fund pledge are all in the document that investors will read before pricing the offering. The market's reaction to those disclosures — once shares are trading — will be the first direct market valuation of AI alignment risk.
Microsoft's buyouts: the other side of the same equation
Microsoft's CPO Amy Coleman, in the October 26 memo: "I also want to be direct that the roles eliminated today are not being replaced by AI. At the same time, what is true is that AI is changing how work gets done."
Both sentences are true. The roles being offered buyouts — in Xbox, Windows+Devices, finance, HR — are not being refilled with AI models. They are being reduced because AI has changed what those teams need to do, how many people they need to do it, and what skills matter. The distinction is technically accurate and strategically evasive in exactly the way that language about AI job displacement is typically evasive.
The financial context for the buyout: Microsoft generated $19.64 billion in free cash flow in Q2 2026 — which sounds large until you note it was down 23% year-on-year, in a period when Alphabet and Amazon reported negative free cash flow in the same AI infrastructure buildout context. The capex required to build and run the GPU clusters that serve Anthropic, OpenAI, and Microsoft's own MAI models is consuming operating cash that would otherwise be available for headcount, R&D, or dividends.
Microsoft's headcount has been flat at approximately 223,000 since 2022. That flatness is itself the story: in the years before heavy AI spending, Microsoft was growing headcount to grow the business. The AI buildout has consumed the capital that would have funded that growth. Buyouts in late 2026 are the deferred consequence of capital allocation decisions made in 2022-2024.
The excluded divisions are revealing. Marketing, the commercial business, and security are not in the buyout pool. These are the functions closest to revenue generation and to the regulatory compliance that AI deployment now requires. The buyouts are concentrated in product areas — Xbox, Windows, Copilot — where AI is changing the product itself: what it does, how it is built, how many people are required to build and maintain it.
The Copilot inclusion is the most significant signal. Copilot is Microsoft's primary AI-native product. The people being offered buyouts in Copilot divisions are presumably not the AI engineering team — they are the product, design, and support functions built around the assumption that building software required more human labor than AI-assisted workflows now require.
What both stories say about the same dynamic
Anthropic's filing and Microsoft's buyouts are not contradictory data points. They are the same dynamic at two different positions in the AI value chain.
Anthropic captures value at the model layer. Its revenue grows at 1,300% per year. Its gross margins are 85%. The 60,000 people building things with its models — the developers, the product teams, the workflows — are Microsoft's cost structure, and they are being reduced because AI tools have made more of what they did automatable.
This is not the robotic-arm-replaces-factory-worker version of automation that economists have modeled for decades. It is white-collar productivity transformation at the layer that most economic models treated as the safe harbor from automation. Product managers, finance analysts, HR professionals, and Xbox game producers are in the buyout pool. The roles most explicitly excluded — sales, marketing, security — are the ones that require human relationship context or regulatory expertise that AI does not yet reliably handle.
The Gartner write-down forecast from last week modeled this specifically: AI revenue will not be sufficient to support profitable returns on all planned AI capital expenditures. But Gartner's model was about infrastructure returns — neocloud providers, GPU depreciation schedules, IaaS pricing. What Microsoft's free cash flow decline and buyout program reveal is a different mechanism: even companies at the top of the AI value chain are facing the same capex-vs-cash-flow tension. Microsoft is not a neocloud. It is one of the world's most profitable software companies. Its Q2 FCF is down 23% because AI infrastructure is expensive and the returns are not yet flowing back at the pace the capital was deployed.
The Anthropic IPO prices that tension explicitly. The S-1 shows what the returns look like when they are flowing — 1,304% revenue growth, 85% gross margin, path to operating profitability. If those numbers hold, the infrastructure spending that Microsoft, Google, and Amazon are absorbing will eventually be justified by the revenue they generate. If Gartner's write-down forecast is right, the revenue won't be sufficient.
November 16 is the tentative trading date for Anthropic ($ANT). The IPO pricing will be the market's first direct opinion on which scenario it believes.
