The $750 Billion Bet Driven by the Computing Power Gap

According to Woofun AI, Anthropic PBC disclosed in the draft prospectus for its initial public offering (IPO) that despite having reserved over $600 billion in long-term computing power resources, the company still faces the harsh reality of being unable to obtain all the compute capacity it expects.

This contradiction highlights a core pain point in today’s artificial intelligence infrastructure sector: even with massive capital commitments, compute supply continues to experience a structural shortage. The prospectus details payment plans—Anthropic must pay Alphabet (GOOGL.US, GOOG.US) $111 billion, Amazon (AMZN.US) $110 billion, and Microsoft (MSFT.US) $31 billion even if it does not fully utilize the compute covered by the agreements. These major cloud computing companies are only part of the stakeholders involved.

Under agreements running through 2033 and 2034, Anthropic may pay up to $86 billion to two vendors—Nscale and Lambda—to rent computing resources powered by Nvidia (NVDA.US) chips. Due to a chip leasing agreement between Google (GOOGL.US) and Broadcom (AVGO.US), Anthropic will bear over $160 billion in Google chip rental costs, which could push its total long-term infrastructure expenditures past the $600 billion mark.

Woofun AI’s data compilation shows that the impetus behind such massive spending derives from explosive revenue growth coupled with uncertainty in acquiring compute. In the six months ending June 30, Anthropic’s revenue reached $16.3 billion, compared to just $1.1 billion in the same period last year;

At the same time, its spending on compute and infrastructure rose to nearly $9 billion, up from $2.3 billion for the first half of 2025. The total six-month expenditure has already surpassed Anthropic’s full-year 2025 spend on compute and infrastructure, which was $7.3 billion.

However, the risk disclosure section of the prospectus points out that access to compute resources remains uncertain. As more cloud service giants view Anthropic as a competitor, they are unwilling to provide excessive computing capacity. Google, Amazon, and Microsoft are all selling competing AI models or supporting rival labs, and those companies receiving the largest payments from Anthropic are also restricting their supply of compute to it.

In addition, Anthropic is exploring further agreements that could increase its total investment by another $150 billion, bringing the combined amount to about $750 billion. On paper, the company’s losses are substantial—of the $80.7 billion loss it incurred in the first half of the year, $72 billion came from accounting charges related to changes in the value of convertible notes and other financial instruments issued for fundraising purposes.

This S-1 prospectus draft was released at a particularly notable moment, as the same day’s quarterly report from Space Exploration Technologies (SPCX.US) drew much of the attention. The company’s superintelligent business achieved $7.6 billion in revenue through cloud services and compute infrastructure deployment.

Nevertheless, for the broader AI industry, Anthropic’s disclosures may be even more significant—a customer that has already committed $600 billion and is still seeking more resources means that in the coming years there will be massive demand for chips, data centers, and cloud compute capacity, with Google, Amazon, Microsoft, Nvidia, and Broadcom all standing to benefit. Investors will soon gain deeper insight, as Anthropic is preparing for a potentially blockbuster IPO that could see the company valued as high as $2 trillion. However, recent reports indicate that after the company disclosed its third-quarter results, some current and potential investors have lowered their valuation expectations to the $1.5 trillion range. It is this intense demand for AI that underlies the sense of urgency: the arms race for compute has evolved from a mere technological race to deep entanglement with capital and supply chains.

This series of data points suggests that the competitive landscape of the artificial intelligence industry is undergoing a fundamental change. Anthropic’s predicament is not an isolated case but a microcosm of the entire industry’s struggle under the compute bottleneck. When cloud service giants are both suppliers and competitors, the political economy of compute allocation becomes even more complicated. Anthropic is attempting to lock in long-term contracts as a hedge, but shifting market dynamics make this strategy fraught with uncertainty. As the IPO approaches, the market will reassess the company’s true value and associated risks. This concerns not only Anthropic’s fate but also the flow of capital and the path of technological evolution across the AI ecosystem. In the coming years, the scarcity of compute resources will continue to drive up industry costs and may also give rise to new business models and technological breakthroughs.