Anthropic's Cloud Capacity Crunch Exposes the Cost of Scaling Frontier AI
Highlights
Anthropic is reportedly having trouble getting enough cloud computing capacity from AWS, Google and Microsoft.
A leaked draft filing for a possible stock market listing describes more than $750 billion in potential cloud contracts, including deals that are still being discussed.
The company reported $16.3 billion in revenue in the first half of 2026, along with an $80.7 billion net loss.
Anthropic is using more rented data centres and outside companies to help run its computing systems.
Its fast revenue growth and reported plans for a November stock market listing have put its infrastructure plans under closer review by investors.
Cloud Capacity Becomes a Strategic Constraint
Anthropic faces a major challenge as it grows its AI business: it needs enough computing power, but it depends on cloud companies that are also competing with it in the AI market.
Reports based on a leaked draft filing for a possible stock market listing say Anthropic has had trouble getting enough computing capacity from Amazon Web Services, Google and Microsoft. All three companies run major cloud platforms and are also building their own AI models and services.
This matters because building and running advanced AI models takes a huge amount of computing power. Training models, making them better and serving more customers all require special computer chips and the systems needed to support them.
Even companies prepared to spend enormous amounts of money may struggle to get the computing capacity they need on time. The supply of hardware, the building of data centres, access to electricity and the time needed to install equipment can all slow growth.
For Anthropic, this creates a difficult situation: the companies supplying its computing infrastructure are also trying to grow their own AI businesses.
Potential Cloud Contracts Exceed $750 Billion
The reported draft filing describes more than $750 billion in possible cloud contracts and infrastructure commitments, including agreements that are still being negotiated.
This does not mean Anthropic has already spent that money or signed contracts covering the entire amount. Reuters has separately reported about $518 billion in cloud and infrastructure commitments over ten years. It is important to tell the difference between commitments that already exist and possible future deals when judging how much money the company may need to pay.
Long-term contracts can help a company secure computing capacity. They also help infrastructure providers plan their investments with greater confidence. However, these contracts can create large financial obligations if demand changes or newer technology changes the type of computing power needed.
Anthropic's challenge is not just deciding how much money to spend. It must get the right kind of computing capacity, make sure the infrastructure is ready when needed and avoid paying for resources it cannot use effectively.
The reported shortages suggest that even plans to spend huge amounts of money may not guarantee that Anthropic can get all the computing power it needs right away.
Hyperscalers Are Both Suppliers and Competitors
Anthropic's relationships with AWS, Google and Microsoft show how the cloud industry is changing. These companies earn money by providing computing services to AI developers while also building their own competing AI models and platforms.
Not all cloud computing capacity can be used for the same tasks. AI systems need suitable computer chips, fast connections between machines, storage, compatible software and facilities with the right setup. A provider may be spending heavily on infrastructure and still lack certain resources that customers need.
Reports say Anthropic is responding by renting data centres and working with outside companies that operate them. This could give it more ways to get computing power and greater control over when new systems are put into use.
However, the reported filing raises concerns that some of these outside operators have limited experience running their businesses. Experience matters when handling large AI workloads. Reliable electricity, cooling systems, maintenance, network performance and uninterrupted service all affect whether a data centre can provide dependable computing power.
Using more suppliers may reduce Anthropic's dependence on any one company, but it also means managing more business relationships and coordinating work across different operators.
Revenue Growth Comes With Significant Financial Risks
Anthropic's plans to expand its computing infrastructure come as its revenue grows quickly and its reported losses remain large.
The leaked filing reportedly shows that Anthropic earned $16.3 billion in revenue during the first half of 2026 and recorded an $80.7 billion net loss. That loss needs more explanation. Reuters reported that about $72 billion came from an accounting charge linked to an increase in the estimated fair value of financial instruments, especially convertible notes. This accounting charge does not mean the company lost the same amount of cash through its normal business operations.
Spending on computing remains a major concern. Reuters reported that Anthropic spent nearly $9 billion on computing and infrastructure in the first half of 2026, compared with $2.3 billion during the same period a year earlier.
According to Reuters reporting published on October 14, the company later told potential investors that its revenue reached $16.8 billion in the third quarter, an increase of 45% from the previous quarter.
These figures show a business that is growing quickly but must keep investing in the infrastructure needed to run its products. Investors will need to judge whether future revenue and cash income can cover these costs and support the company's long-term commitments.
Customer Concentration and AI Liability
The reported filing also suggests that two customers provided about one-fifth of Anthropic's revenue in the first half of the year. Depending heavily on a small number of large customers can be risky if they reduce their spending or move to competing services.
Another reported disclosure concerns possible legal responsibility for harm caused by harmful AI behaviour. Anthropic warns that claims against it could be larger than the amount covered by its insurance.
This describes a possible risk, not proof that such costs have already occurred. The company's actual responsibility would depend on what happened, which laws apply and what its contracts say. Even so, this risk is another factor investors must consider when judging the long-term financial risks of building increasingly powerful AI systems.
IPO Plans Put Infrastructure Strategy Under Scrutiny
Anthropic is reportedly aiming to launch an initial public offering (IPO) in November. A stock market listing would bring more attention to its financial reports and infrastructure plans.
Investors will need to separate money already spent from signed contracts and possible deals that are still being negotiated. They will also need to judge how well Anthropic can turn its infrastructure commitments into reliable computing power and revenue.
Anthropic's situation reflects a wider problem across the AI industry. Companies that build AI models must compete on performance, price and reliability while securing the computing resources needed to provide their services.
Anthropic's future growth will depend not only on demand for Claude but also on how well it manages relationships with suppliers, expands its infrastructure reliably and keeps enough financial flexibility as its need for computing power grows.