Anthropic lost $80.7B in six months. The real story is the $750B compute tab.
Anthropic filed a draft S-1 showing explosive revenue growth, but locked-in infrastructure debt proves AI is a hardware game.
Business · Source: ZeroHedge
What happened
Anthropic is preparing for a massive initial public offering. A leaked draft of their S-1 prospectus shows staggering numbers for the first half of 2026. The company generated $16.3 billion in revenue in just six months. That is more than triple its total revenue for all of 2025. But the bottom line is a bloodbath. Anthropic recorded an $80.7 billion net loss over the same period.
Do not let the headline loss fool you. About $72 billion of that deficit is a non-cash accounting artifact. As the private valuation of Anthropic surged, early convertible notes turned into massive paper losses. Strip away the accounting noise, and the actual operating loss is $8.7 billion for the half. That still eclipses their entire operating loss for 2025. The culprit is direct compute and infrastructure spending, which hit nearly $9 billion before paying a single employee.
The future looks even more expensive. Anthropic is now on the hook for roughly $600 billion in cloud computing contracts. They are actively negotiating another $150 billion in capacity. This pushes their total compute obligations to a potential $750 billion. These agreements are rigid and noncancelable. Anthropic must pay them regardless of actual usage. They owe $111 billion to Google Cloud, $110 billion to Amazon, and $31 billion to Microsoft.
Key facts
- $16.3 billion — Revenue generated by Anthropic in the first half of 2026
- $80.7 billion — Total net loss recorded in the first half of 2026
- $72 billion — Non-cash accounting charge tied to rising valuations
- $750 billion — Potential total compute obligations including active negotiations
- 20 percent — Share of first-half revenue coming from just two customers
Why it matters
The frontier AI race is no longer just a software competition. It is a heavy industry. Building foundational models requires capital at the scale of nation-states. Startups cannot compete at the base layer without mortgaging their future to big tech cloud providers. If you are building an AI product, you must assume the underlying cost of intelligence will be dictated by these massive, locked-in infrastructure debts.
Cloud computing used to be flexible and pay-as-you-go. Now it is a rigid debt structure. Anthropic is locking in ten years of liabilities while relying heavily on a fickle user base. The filing notes that 20 percent of their revenue comes from just two customers. If those enterprise buyers churn, the noncancelable compute bills still come due. This fragile duration mismatch could trigger a massive collapse if AI revenue growth stalls.
For builders
Build at the application layer
Do not try to train frontier models from scratch. The infrastructure costs are insurmountable for almost everyone. Let companies like Anthropic burn billions on compute while you build profitable software on top of their APIs.
Beware of cloud lock-in
Anthropic is forced to look beyond the big three hyperscalers because they are viewed as competitors. If you scale massive AI workloads, diversify your compute providers early. You do not want your landlord to become your biggest rival.
Prepare for catastrophic risk
Anthropic explicitly warns that a serious AI security event could exceed their insurance cover. Enterprise builders must factor in the cost of catastrophic failures. Clients will eventually demand strict liability protection that foundational models cannot afford to guarantee.
My take
Winning the AI race is not about building the smartest model anymore. It is about affording the bill. I look at Anthropic taking on a mountain of noncancelable debt just to stay in the game, and it proves one thing. AI is ultimately a hardware monopoly disguised as a software revolution.
Original reporting: ZeroHedge. This is my rewrite and opinion.