Key takeaways
- Anthropic cannot secure adequate cloud capacity from AWS, Google, and Microsoft, which now view the AI startup as a competitor rather than a customer.
- The company estimates it could spend over $750 billion on cloud contracts but is forced to invest in its own data centers and third-party operators.
- Despite Q3 revenue of $16.8 billion, Anthropic's H1 net loss reached $80.7 billion, with only two customers accounting for one-fifth of sales.
- Anthropic faces potential liability from harmful AI behavior exceeding available insurance coverage, adding risks ahead of its planned November IPO.
Anthropic faces a fundamental challenge to its growth trajectory: the cloud providers powering its AI operations increasingly view the company as a competitor rather than a customer, according to details from an S-1 filing leaked to The Wall Street Journal.
The company cannot secure adequate computing capacity from AWS, Google, and Microsoft, the three providers that dominate the cloud market. Anthropic estimates it could spend more than $750 billion on cloud contracts over time, with some agreements still under negotiation, yet even that level of spending falls short of what the business requires.
The Cloud Capacity Squeeze
Why Hyperscalers Are Withholding Capacity
The shortage reflects a strategic shift among the technology giants. AWS, Google Cloud, and Microsoft are not merely utilities delivering infrastructure to customers; they are increasingly competitors building their own AI products and services. Anthropic’s explosive growth and Claude’s market prominence have shifted it from being a consumer of cloud services to a potential threat to these companies’ AI ambitions.
The result is a misalignment between Anthropic’s needs and hyperscaler willingness to supply. According to the Wall Street Journal’s reporting on the leaked filing, capacity constraints are direct and measurable, forcing Anthropic to look elsewhere.
Alternative Infrastructure and Trade-offs
Anthropic is responding by diversifying its computing sources. The company is leasing its own data centers and contracting with third-party data center operators to supplement capacity from the major cloud providers. But this strategy carries risks. According to the filing, Anthropic is concerned that independent operators have limited operating history and unproven reliability compared to the established hyperscalers, potentially creating vulnerabilities in the infrastructure that powers Claude’s inference.
Revenue Growth Masks Structural Imbalances
Concentration and H1 Performance
The company reported $16.3 billion in revenue during the first half of 2026, but this figure conceals a critical vulnerability. Only two customers accounted for approximately one-fifth of total sales, meaning Anthropic depends heavily on a narrow base of high-value relationships. The concentration represents significant revenue risk if either customer changes its purchasing patterns or relationship with Anthropic.
Against this revenue backdrop, the company’s net loss reveals the mathematics of building a frontier AI company. Anthropic posted an $80.7 billion net loss in H1, a figure that underscores the enormous gap between sales and the costs required to develop, train, and operate Claude at scale.
Q3 Momentum and IPO Timeline
The trajectory shifted in the third quarter. Revenue reached $16.8 billion in Q3 alone, a 45% increase from the previous quarter. Anthropic communicated this result to prospective investors as the company prepares for an expected November IPO, signaling both momentum and urgency around going public.
Undisclosed Liability Exposure
The S-1 filing surfaces additional risks beyond infrastructure constraints. Anthropic disclosed that harmful outputs from Claude could generate legal liability for damages, with potential liability exceeding the coverage available through existing insurance policies. This disclosure adds to earlier warnings Anthropic has made public about catastrophic risks from AI systems.
The Strategic Bind
Anthropic’s predicament exposes a tension at the heart of AI infrastructure economics. Cloud providers have built trillion-dollar businesses by treating compute as a neutral commodity. The shift to AI, where training and inference have become existential to their own competitive position, has changed the calculus. When your customer is also your competitor, the relationship becomes transactional and fragile.
Resolving this bind will be central to Anthropic’s success as a public company. The company must either negotiate more favorable terms with hyperscalers, scale its own or third-party infrastructure to reliability parity, or diversify its customer concentration beyond the current top two accounts.
Source: TechCrunch
Frequently Asked Questions
Why can't Anthropic get enough cloud capacity from AWS, Google, and Microsoft?
AWS, Google Cloud, and Microsoft increasingly view Anthropic as a competitor rather than a customer, since they are developing their own AI products and services. The companies are prioritizing their own AI initiatives over providing unlimited capacity to a rival.
What is Anthropic doing to solve its cloud capacity problem?
Anthropic is leasing its own data centers and contracting with third-party data center operators to supplement capacity from the major cloud providers. However, the company is concerned that independent operators have limited operating history and may be less reliable than established hyperscalers.
How does Anthropic's customer concentration affect its business?
Only two customers account for approximately one-fifth of Anthropic's total sales. This heavy concentration on a small number of high-value customers creates significant revenue risk if either customer changes its relationship with Anthropic.
