Anthropic signs $895.4 billion in deals to secure AI computing power

The compute bottleneck

Anthropic is currently facing a significant shortage of the resources required to power its artificial intelligence models. The company's CEO, Dario Amodei, described the amount of computing power needed as a "cone of uncertainty" because the requirements are so difficult to predict. Without a steady supply of hardware and electricity, the company cannot reliably develop new models or serve its existing customers.

The shortage became a visible problem last year when the popularity of Anthropic's coding tool, Claude Code, outpaced the company's available capacity. To manage this, Anthropic had to limit free usage of its products and increase prices for certain services. These constraints forced the company to launch an aggressive campaign to secure enough infrastructure to sustain its growth.

As reported by indianexpress.com, the company's public offering prospectus dedicated more than 10 pages to explaining these stakes. The document notes that staying at the frontier of AI capabilities is substantially dependent on securing compute at the required scale. This underscores how vital physical infrastructure has become to the success of software companies.

A massive deal spree

To solve its resource problem, Anthropic has signed dozens of contracts with a wide variety of partners. These include tech giants like Amazon, Google, and Microsoft, as well as SpaceX and various chipmakers. The company is also looking beyond traditional cloud providers by signing deals with "neoclouds," which are companies that provide computing power but may have faster access to land and power.

Specific commitments from the company's prospectus include $121.3 billion from Google, $110 billion with Amazon, and $44.9 billion with Microsoft. Additionally, Anthropic secured a $120.5 billion deal with SpaceX through 2036. These large-scale agreements are intended to provide the backbone for the company's workloads as it prepares for a potential $100 billion initial public offering.

Anthropic is also diversifying its hardware sources to avoid over-reliance on a single supplier. The company's chief compute officer, Tom Brown, has led efforts to use chips from Amazon, Google, and Advanced Micro Devices in addition to Nvidia. Brown stated that the shortage of power and labor is "very clearly the bottleneck for all of our progress."

Infrastructure and logistics

The company is taking a hands-on approach to building the physical sites where its AI will run. Anthropic has hired experts in data center construction and is buying cooling systems and other equipment directly to speed up the process. They are even helping to design the data centers that other companies will eventually build and operate to ensure the hardware is installed quickly.

To find enough space, Anthropic is exploring unconventional options like converting old coal plants and factories into data centers. They have met with developers in states like Texas and Indiana to discuss permits and power grid connections. The company is also working with former cryptocurrency miners, such as TeraWulf and Hut 8, to construct new facilities.

These projects have faced logistical hurdles, including competition for specialized labor. In one instance, a project in New York had to hire traveling electricians from other areas because local workers were busy with a football stadium. These complications have forced Anthropic to be flexible, considering smaller data center sites that can be supplemented with batteries and natural gas.

The path to IPO

The massive scale of these computing commitments has created a significant financial burden for the startup. Because Anthropic is not a giant like Meta or Google, it lacks the deep credit histories and massive internal resources to build these sites alone. This has made the company's upcoming public offering and recent debt financing, including a $63 billion loan, essential for survival.

The company is currently in a race to secure enough capacity to satisfy both its internal research needs and its external customers. If it fails to secure these resources, it risks falling behind competitors who already have established infrastructure. The goal is to ensure that the company can deliver products like the Claude chatbot without the interruptions it faced last year.

Looking ahead, Anthropic plans to significantly increase its power capacity in the coming year. It expects to have more than 3.5 gigawatts of power this year, which is enough to power all the homes in Chicago. The company aims to double that amount next year to keep pace with the escalating demands of the AI boom.