▲ Anthropic AI
Anthropic announced that under its cloud, computing, and infrastructure agreements with six partner companies, it will spend at least USD 518 billion (approximately KRW 701 trillion) over the next 10 years, explaining that 80% of this amount consists of commitments or non-cancellable agreements where the full committed amount must be paid regardless of actual usage.
Reuters reported this, citing a draft initial public offering (IPO) prospectus that Anthropic confidentially submitted to the U.S. Securities and Exchange Commission (SEC) last June.
Anthropic's investment scale is comparable to competitor OpenAI's USD 500 billion Stargate project.
According to the prospectus, Anthropic has entered into massive long-term contracts with partner companies including Broadcom, Google, Amazon, and Microsoft (MS).
It has signed long-term infrastructure service contracts to spend at least USD 111.1 billion with Google, USD 110 billion with Amazon, and USD 31.4 billion with MS.
In particular, even if actual spending falls short of the committed amounts, Anthropic must pay the difference to Google or Amazon.
The contract with MS is also non-cancellable barring a material breach by the counterparty.
In addition, it has signed an equipment lease contract worth USD 161.2 billion with Broadcom.
Most of this agreement is also non-cancellable.
For Anthropic, this means it will be obligated to pay massive sums even if the AI bubble bursts in the future and demand for computing resources declines.
Explaining the rationale behind entering into such contracts, Anthropic stated, "In the future, demand for advanced AI systems will exceed supply, and ultimately the availability of computing resources will be the biggest constraint."
There were also new spending agreements.
Anthropic signed an agreement with Elon Musk's xAI for up to USD 84.5 billion worth of Nvidia-based computing capacity through 2029.
This contract can be cancelled with a 90-day prior notice.
An agreement was also reached in which AMD will purchase USD 5 billion worth of Anthropic stock, and Anthropic will receive over USD 20 billion in AMD computing resources.
Anthropic pointed out that Google, Amazon, and MS are both infrastructure providers and competitors building their own AI models, citing its reliance on these big tech companies as a "core risk."
It added, "Situations may arise where the interests of partner companies may not entirely align with those of Anthropic."
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.