Tech • AI • Robotics • Game
I Investigated Anthropic's Secret File (416 pages)
9.4/10TL;DR
Anthropic’s IPO filing portrays one of the fastest-growing software businesses on record, but also a company dependent on heavy infrastructure spending, a concentrated enterprise coding boom, and rising competition from cheaper open models.
KEY POINTS
Explosive revenue growth
Anthropic, founded in 2021, went from about $41 million in annual revenue in 2023 to roughly $17 billion in a single quarter in 2026. Quarterly revenue rose from $376 million to $17 billion over seven reported periods, a year-on-year jump of about 1,127%. With around 5,300 employees, that implies roughly $6.6 million in revenue per employee.
Enterprise API dominates the business
The company’s core business is not consumer subscriptions but enterprise token sales through its API. About 79% of revenue comes from usage-based API demand, while Pro, Max and Team subscriptions account for about 21%. Distribution through AWS, Google Cloud and Azure has also surged, reaching 44% of revenue from 11% in 2023.
Customer expansion is unusually intense
The number of customers spending more than $100,000 annually climbed from about 20 to more than 11,900. Net revenue retention is around 900%, meaning the same cohort is spending about nine times more than a year earlier. That level of expansion suggests rapid product adoption, but it also raises questions about whether spending can keep compounding at the same pace.
Coding products appear to be the growth engine
A key driver is Claude Code. Corporate users who do not use the coding product reportedly spend less than $75 a month, while the heaviest 5% of users spend roughly 20 times more. The pattern indicates that developers are acting as the entry point inside enterprises, with adoption spreading from individual engineers to wider team contracts.
Token-maxing helped, then faded
Some enterprises pushed developers to maximize AI usage, a phase described as token maxing. Prime-time costs reportedly rose from $20,000 to $50,000 per developer per month, helping quarterly growth accelerate to 132% and then 144%. That pace then cooled to about 47% as companies reviewed bills, some buyers pulled back, and coding competition improved.
Cheaper alternatives are a real threat
Revolut said it shifted most coding workloads for roughly 2,000 developers to GLM models from Z.ai, hosted on its own infrastructure. The company reported cutting its AI coding bill by about 90%, to roughly $0.30 per million tokens, while also improving development speed metrics. DoorDash was also cited as moving some usage to an Alibaba model, underscoring the pressure from open-weight and lower-cost rivals.
Prices are falling even as usage grows
Newer model releases have reduced costs per task. Opus 5.5 cut cost per task by about 40% versus Opus 5, while Fable 5.1 reduced it by about 25% overall and as much as 45% on coding. Those cuts reportedly reduced revenue by around $600 million quarter to quarter, but management appears to be betting that lower prices will drive enough extra usage to offset the hit.
Profitability remains volatile
Compute efficiency has improved sharply. In early 2025, the company spent about $2.50 on servers for every $1 of revenue; by the second quarter of 2026, that had fallen to about $0.48, and $0.58 in the third quarter. Anthropic posted its first profitable quarter in Q2 2026, then swung back to a $577 million loss in Q3, showing how sensitive results remain to utilization and fixed infrastructure costs.
The headline loss is not the whole story
The filing cites roughly $104 billion in losses over three months, but about $94 billion is described as non-cash, tied largely to financing arrangements that may convert into shares. That means the accounting loss does not reflect the same level of cash burn, though it does point to potential future dilution for shareholders. The company’s economics look better on a cash basis than the headline figure suggests.
The biggest number is future compute commitments
The most striking risk is about $895 billion in committed spending on compute and data centers, around 25 times annual revenue. Against that, the company has about $132 billion in cash and aims to raise roughly $100 billion in an offering. If enterprise demand weakens or customers self-host cheaper models, those commitments could become the central financial risk.
Governance is tightly controlled
Control is concentrated through a single Class F share with 50.1% of voting power held by the founder, alongside a trust that elects most of the board. Anthropic is also structured as a public benefit corporation, and the filing warns extensively about AI safety and broader systemic risks. That combination gives management unusual power even after a listing.
CONCLUSION
Anthropic combines exceptional growth, strong enterprise traction and improving compute efficiency with unusually large infrastructure obligations and intensifying competition. The IPO case rests on whether demand for premium AI coding and enterprise automation can keep outrunning price declines, open-model competition and the burden of massive capital commitments.
Ask a question