Anthropic is preparing for an exceptionally large expansion of its AI infrastructure, with at least $518 billion in long-term commitments over the next decade, according to information disclosed in its confidential IPO filing. Roughly 80% of those obligations are binding regardless of actual usage, underscoring how competition for computing capacity is pushing AI companies toward increasingly large and long-duration infrastructure agreements.

Broadcom, Google and Amazon Carry the Largest Commitments

Broadcom represents the largest individual obligation, with approximately $161.2 billion in equipment lease commitments that are largely non-cancelable. Anthropic has also committed at least $111.1 billion to Alphabet’s Google and $110 billion to Amazon under long-term infrastructure service agreements.

The company’s commitment to Google runs from April 2026 through July 2033, while its Amazon agreement extends from May 2026 through April 2036. Under the reported terms, Anthropic must cover any shortfall if its actual spending falls below the committed amount. Microsoft represents another $31.4 billion commitment running from November 2026 through May 2033, with the agreement described as non-cancelable except in the event of an uncured material breach by Microsoft.

Together, the Broadcom, Google, Amazon and Microsoft obligations account for roughly 80% of the headline $518 billion figure, highlighting the concentration of Anthropic’s infrastructure strategy among a small number of major technology and semiconductor companies.

AI Growth Is Creating a New Infrastructure Model

The commitments reflect a fundamental constraint facing advanced AI development: access to computing capacity. Anthropic has argued that future demand for advanced AI systems could exceed available supply, making long-term access to processors, cloud infrastructure and specialized hardware increasingly important.

The company has expanded its relationships across different technology platforms. In April 2026, Anthropic announced an expanded partnership with Google and Broadcom for multiple gigawatts of next-generation TPU capacity expected to come online from 2027. The company also announced a separate agreement with Amazon for up to 5 gigawatts of new compute capacity, including Trainium technology.

This diversification gives Anthropic access to Google TPUs, Amazon Trainium and other AI hardware while reducing reliance on a single computing platform. At the same time, the long-term commitments create substantial financial obligations if AI demand, revenue growth or computing utilization does not develop as expected.

Scale Creates Opportunity but Also Balance-Sheet Risk

For Broadcom, Google and Amazon, Anthropic’s commitments provide substantial visibility into future demand for AI infrastructure, cloud services and specialized computing equipment. For Anthropic, however, contracted capacity does not automatically translate into equivalent revenue or profitability.

The company must continue converting AI demand into commercial growth while managing fixed infrastructure obligations. The risk is particularly significant because some agreements require payments even when actual usage falls below contracted levels. The company also faces dependence on major technology partners that can simultaneously act as suppliers, investors and competitors.

The next phase will depend on Anthropic’s ability to scale revenue and utilization alongside its infrastructure commitments. Investors will be watching the company’s eventual public filings, capital requirements, customer growth and spending levels closely. The scale of the commitments also offers a broader signal for the AI industry: as competition for computing capacity intensifies, access to chips and cloud infrastructure is becoming a strategic asset, but securing that capacity is creating financial commitments on a scale rarely seen in the technology sector.