Anthropic reveals $895.4 billion in cloud and data center commitments

Massive scale of AI infrastructure

Anthropic recently updated its Form S-1 filing with the SEC, providing a clearer picture of its financial health and infrastructure needs before going public. The company reported $17 billion in revenue for the third quarter, which represents a 1,127 percent increase compared to the same period last year. Despite this growth, the company reported a net loss of $23.29 billion during that same timeframe.

The most striking detail in the update is the scale of the company's infrastructure commitments. Anthropic revealed that its purchase commitments and lease agreements now total about $895.4 billion. This figure is significantly higher than the $417 billion previously listed in earlier filings, showing a rapid acceleration in how much hardware the company plans to secure.

The company noted that substantially all of this spending is related to compute and infrastructure. This includes the physical hardware and data centers required to train and run large-scale artificial intelligence models. As the demand for AI capabilities grows, the cost of the underlying electricity and silicon becomes a primary driver of corporate spending.

The cost of cloud partnerships

The filing breaks down how Anthropic plans to pay for this massive amount of computing power. The company has committed $121.3 billion to lease compute from Google, $120.5 billion from SpaceX, $110 billion from Amazon, and $44.9 billion from Microsoft. These partnerships ensure that Anthropic has the necessary resources to scale its operations across different cloud environments.

A critical detail for investors is the nature of these contracts with the major cloud providers. The deals with Google, Amazon, and Microsoft are mostly non-cancelable outside of uncured material breaches. This means Anthropic is legally obligated to pay for this capacity regardless of how much of it they actually use in their day-to-day operations.

In contrast, the deal with SpaceX is structured differently. It is cancelable with 90 days' notice starting in March of next year. This provides Anthropic with slightly more flexibility regarding its partnership with the space company compared to its long-term commitments with the three largest cloud hyperscalers.

Diversifying compute sources

Anthropic is also looking beyond the largest cloud providers to secure its supply chain. The company signed a $44.6 billion deal with the neocloud provider Nscale, which runs until 2033. Additionally, it entered into a $41.3 billion agreement with Lambda that lasts until 2034. These deals show a strategy of diversifying where the company's models are hosted and trained.

There are specific conditions attached to these smaller providers. Nscale is required to attempt to obtain qualifying financing for its Nvidia GPUs, while Lambda must secure its facility lease in a timely manner. These requirements suggest that Anthropic is carefully managing the risks associated with smaller infrastructure partners who may face their own scaling hurdles.

The company also entered into a $50 billion deal with Fluidstack and has contracts with TeraWulf and Hut 8. While the filing did not mention an $11.6 billion cloud deal with Akamai or a compute agreement with CoreWeave, the sheer number of partners indicates a complex web of infrastructure dependencies. This multi-provider approach helps mitigate the risk of being overly dependent on a single source of hardware.

Future hardware and equity

Looking further ahead, Anthropic is planning for even more compute capacity in the coming years. The New York Times reported that the company is expected to have more than 3.5GW of compute this year, with a similar amount expected in 2027. Furthermore, Anthropic is set to procure additional TPU compute, which could cost between $125 billion and $150 billion between 2028 and 2031.

This additional TPU compute is not included in the $895.4 billion calculation mentioned in the S-1 filing. It could be acquired through a direct purchase from Google with five-year financing, a new Google Cloud deal, or through a third-party partner. This indicates that Anthropic's long-term infrastructure roadmap extends well into the next decade of AI development.

The company also has a unique hardware supply agreement with AMD. Under this arrangement, the chip designer is on track to buy $5 billion in Anthropic shares. These shares will be purchased in stages based on specific milestones regarding the delivery and deployment of AMD GPUs. This structure aligns the interests of the hardware manufacturer with the AI company's growth and infrastructure needs.

Growth and spending trends

The financial data reveals a shift in the pace of Anthropic's growth. While the year-over-year revenue increase was massive, the quarter-to-quarter growth slowed to 47 percent. This is a notable decrease from the 144 percent growth seen in the second quarter and the 132 percent growth seen in the first quarter.

This deceleration in growth comes alongside a significant increase in infrastructure expenses. For the nine months ending September 30, the company's compute and infrastructure expenses rose to $14.4 billion. This represents a 329 percent increase over the same period last year, highlighting how quickly the costs of maintaining AI systems are escalating.

As Anthropic prepares for its public listing, these numbers provide a clear look at the company's burn rate and capital requirements. The massive commitments to cloud providers and hardware manufacturers show that the company is betting heavily on the continued expansion of AI. Investors will need to weigh the high revenue growth against the enormous costs and non-cancelable obligations required to sustain it.