Anthropic's $8B Loss and $54.6B Compute Bet
Anthropic's Big Reveal: Soaring Revenue, Soaring Losses
Anthropic, the company behind the Claude AI model, recently shared its IPO prospectus with select partners, aiming for a staggering $2 trillion valuation. The numbers inside tell a story of explosive growth—and equally explosive spending.
In 2025, Anthropic's revenue skyrocketed to nearly $4.6 billion, a 12-fold jump from the previous year. Nearly two-thirds of that came from the U.S. market. But here's the catch: operating losses more than doubled, surpassing $8 billion. Growth, it seems, comes at a hefty price.
The Business Model: Usage Over Subscriptions
How does Anthropic make money? Mostly by charging for usage. The Claude model operates primarily on a pay-as-you-go basis, pulling in about $3.8 billion. Subscription services, by contrast, brought in just $789 million—less than 20% of total revenue. It's a clear signal: businesses are willing to pay for AI, but they prefer to pay for what they use.

A Double-Edged Alliance with Tech Giants
Anthropic's ties to Amazon and Google run deep. In 2025, the company earned $2.16 billion through these cloud platforms—a share that leaped from 11% in 2023 to 47%. But that convenience comes at a cost: roughly $351 million in channel fees.
More importantly, Amazon and Google aren't just partners. They're also investors, compute suppliers, and—through their own AI models—competitors. That creates a potential conflict of interest, especially when it comes to pricing and negotiation. It's like renting space in a building owned by your rival.
Customer Concentration and Compute Commitments
Anthropic's revenue is also heavily concentrated. Its two largest anonymous customers accounted for 24% of revenue, and there are no long-term agreements in place. That's a risky bet if those customers decide to walk.
Meanwhile, the company has locked itself into $54.6 billion in non-cancellable compute hosting commitments by the end of 2025. Unpaid invoices total $909 million, with 60% collected through intermediaries. In plain terms: Anthropic has promised a lot of money for computing power, and it's on the hook for it.
What This Means for the AI Industry
Anthropic's prospectus offers a rare glimpse into the financial realities of a top AI player. The company has successfully scaled its commercialization—but it still faces serious hurdles. Limited control over distribution channels, rigid computing costs, and fluctuating customer retention are all part of the equation.
Can Anthropic turn its massive bet into sustainable profit? That's the trillion-dollar question.
Key Points
- Revenue surge, losses deepen: 2025 revenue hit $4.6B (12x YoY), but operating loss exceeded $8B.
- Usage-based billing dominates: Pay-as-you-go brought in $3.8B; subscriptions only $789M.
- Amazon and Google: partners and rivals: 47% of revenue flows through their clouds, with $351M in fees—and potential conflicts.
- Concentration risk: Two anonymous customers make up 24% of revenue, no long-term deals.
- Compute commitments: $54.6B in non-cancellable hosting obligations; $909M unpaid invoices.