Anthropic's cloud capacity squeeze reveals AI's infrastructure problem
A leaked filing says AWS, Google and Microsoft are increasingly treating Anthropic as a rival while the Claude maker hunts for more computing capacity.
Anthropic is running into a problem that is becoming harder for the AI industry to hide: the companies supplying the computing power also increasingly want to compete for the same customers. A leaked S-1 filing says Anthropic has struggled to secure enough cloud capacity from AWS, Google and Microsoft as those providers see the Claude maker less as a straightforward customer and more as a rival, according to TechCrunch 1.
The capacity squeeze
The filing reportedly says Anthropic has sought more capacity than its cloud partners were willing to provide. That matters because training and serving frontier AI models requires an extraordinary amount of infrastructure, and demand for Claude is growing at the same time as Anthropic's own ambitions.
The company is responding by leasing its own data centres and working with third-party operators. The filing also raises a less comfortable concern: some of those operators have a limited operating history, which could make them less reliable than the hyperscalers Anthropic is trying to supplement. In other words, building a more diverse infrastructure base may reduce dependence on one provider while introducing new operational risks.
The reported numbers are enormous. Anthropic believes it could spend more than $750 billion on cloud contracts, including deals still under negotiation, while the company posted an $80.7 billion net loss in the first half of the year against $16.3 billion in revenue, according to the reporting summarised by TechCrunch 1. Its third-quarter revenue was later reported at $16.8 billion, up 45 per cent from the previous quarter 3.
Why the partnerships are changing
Cloud providers have traditionally benefited when fast-growing AI companies rent their chips, data-centre capacity and associated services. But the relationship becomes more complicated when those customers build competing models and products. Google, Microsoft and Amazon Web Services are not passive landlords: each is investing heavily in its own AI platforms and wants the infrastructure economics as well as the software upside.
Anthropic has tried to spread its bets. In its own account of an expanded Google Cloud partnership, the company said its compute strategy uses Google's TPUs, Amazon's Trainium and Nvidia's GPUs, while describing Amazon as its primary training partner and cloud provider 2. That diversification can provide negotiating leverage and reduce the danger of a single supplier becoming a bottleneck. It also leaves Anthropic managing a more complex technical stack across several platforms.
Our opinion
The striking part of this story is not simply that Anthropic needs more servers. It is that the AI industry's supply chain is becoming a competitive battlefield. The model makers need hyperscalers to expand quickly, but hyperscalers increasingly have the money, chips and distribution to build rival products themselves.
That tension will not be solved by a single giant contract. Anthropic can lease more of its own capacity, add specialist operators and spread workloads across chips, but every extra layer brings cost and reliability questions. The company's growth may be impressive, yet the filing shows why revenue alone does not make an AI business financially comfortable: the infrastructure required to keep the models available can expand almost as quickly as demand.
For readers, the practical consequence is simple. The next phase of AI competition will be fought not only in model benchmarks and product launches, but also over who gets access to scarce computing capacity — and on whose terms.
Sources: TechCrunch reports on Anthropic's leaked filing and cloud relationships; Anthropic's own account of its Google Cloud expansion; and Reuters' reporting on the company's revenue growth.