Anthropic secures $895.4 billion in computing power to fuel AI growth

The compute bottleneck

Anthropic is currently engaged in an aggressive effort to secure the physical infrastructure required to run its artificial intelligence models. As the company grew rapidly, it faced a shortage of computing power, which CEO Dario Amodei described as a "cone of uncertainty." This uncertainty makes it difficult for the company to predict exactly how much hardware it will need to stay competitive.

The shortage of computing power has already impacted the company's products and services. When its AI coding tool, Claude Code, became popular last year, Anthropic lacked the capacity to meet the demand. This forced the company to limit free usage and increase prices for some of its offerings to manage the workload.

To solve these issues, the company is pursuing a massive deal spree with a wide variety of partners. It has signed contracts with tech giants like Amazon, Google, Microsoft, and SpaceX, as well as chipmakers and data center startups. These efforts are essential for Anthropic to deliver its chatbot and other products to businesses and consumers.

A multi-billion dollar roadmap

The scale of Anthropic's commitment is reflected in its public offering prospectus, which devotes more than 10 pages to the importance of computing power. The company has committed $895.4 billion through 2045 to various chip and computing deals that it largely cannot back out of. This massive investment is necessary to sustain its workloads and remain at the frontier of AI capabilities.

Specific contracts highlight the diversity of the company's strategy. It has committed $121.3 billion to Google, $110 billion to Amazon, $44.9 billion to Microsoft, and $120.5 billion to SpaceX through 2036. These figures represent a significant portion of the capital the company needs to raise as it prepares for a potential initial public offering.

To fund these expensive demands, Anthropic is seeking to raise $100 billion in its public offering. It also recently secured $63 billion in debt financing from private credit firms and Wall Street banks. Despite these funds, the company has warned that it may need to seek additional financing sooner than it originally planned because of the high costs of hardware.

Diversifying the supply chain

Because major cloud providers like Amazon and Microsoft are often backlogged with demand, Anthropic is looking for alternative sources of power. The company is working with "neoclouds," which are providers that offer computing power and may have faster access to land and electricity. This includes a $50 billion deal with Fluidstack and a $41.3 billion deal with Lambda.

Anthropic is also partnering with former cryptocurrency miners, such as TeraWulf and Hut 8, to build out its data center footprint. These partners can often bring computing power online more quickly than traditional tech giants. The company is even exploring the conversion of old coal plants and factories into new data center sites to meet its goals.

The company is also diversifying its hardware sources to reduce reliance on Nvidia. Tom Brown, Anthropic's chief compute officer, has led efforts to incorporate chips from Amazon, Google, and Advanced Micro Devices. Brown noted that the shortage of power and labor needed to support AI demand is a major bottleneck for the entire industry's progress.

Infrastructure hurdles

Securing the physical resources for AI is not without significant logistical challenges. For example, Anthropic worked with Fluidstack to resolve concerns about power fluctuations that could damage equipment. In another instance, a project in upstate New York faced a shortage of electricians, forcing the company to pay more to hire traveling workers from other regions.

The company has had to be flexible with its site requirements to speed up construction. While it originally wanted sites with at least 200 megawatts of power, it has started considering smaller sites of 50 megawatts. These smaller sites can be supplemented with batteries and natural gas to reach the required power levels more quickly.

Industry experts suggest that Anthropic may have to settle for building capacity in smaller increments rather than all at once. Because the company does not control the primary pipelines of computing power, it must navigate a complex landscape of local regulations and infrastructure limitations. This ongoing race for power will determine how quickly the company can scale its AI capabilities in the coming years.