According to Woofun AI, Anthropic PBC revealed in its draft initial public offering (IPO) prospectus that even though it has reserved long-term computing power resources worth more than $600 billion, the company still faces the stark reality of not being able to obtain all the expected computing power.
This contradiction reveals a core pain point in the field of artificial intelligence infrastructure: even with huge capital commitments, there is still a structural shortage in the supply of computing power. The prospectus details the payment plan. Even without fully using the computing power covered by the agreement, Anthropic will have to pay $111 billion to Alphabet (GOOGL.US, GOOG.US), $110 billion to Amazon (AMZN.US), and $31 billion to Microsoft (MSFT.US). These large cloud computing companies are just some of the parties involved.
Anthropic is also likely to pay Nscale and Lambda vendors up to $86 billion to rent computing power resources powered by Nvidia (NVDA.US) chips, under agreements up to 2033 and 2034, respectively. Due to the chip lease agreement between Google (GOOGL.US) and Broadcom (AVGO.US), Anthropic is required to bear more than $160 billion in Google chip rental costs, and its total long-term infrastructure spending is likely to break the $600 billion mark.
Data compiled by Woofun AI shows that the driving force behind this huge investment comes from explosive revenue growth and uncertainty in computing power acquisition. In the six months ending June 30, Anthropic's revenue reached $16.3 billion, compared to only $1.1 billion in the same period last year;
Meanwhile, its spending on computing power and infrastructure rose to nearly $9 billion, up from $2.3 billion in the first half of 2025. Total spending over these six months has already surpassed Anthropic's total spending of $7.3 billion on computing power and infrastructure for the full year of 2025.
However, the risk disclosure section of the prospectus indicates that access to computing power resources remains uncertain because as more and more cloud service giants view Anthropic as a competitor, they are unwilling to provide too much computing power resources. Google, Amazon, and Microsoft are all selling competitive AI models or supporting the development of competitive labs, and the companies that get the most money from Anthropic are also limiting the supply of computing power to it.
In addition, Anthropic is also exploring further agreements, which may increase its total investment by another 150 billion US dollars, so the total amount may reach about 750 billion US dollars. Judging from book data, the company's losses were quite serious — of its $80.7 billion loss in the first half of the year, $72 billion stemmed from accounting processing fees associated with changes in the value of convertible notes and other financial instruments issued to raise capital.
The timing for this draft S-1 prospectus was quite special because the quarterly report released by Space Exploration Technologies (SPCX.US) on the same day attracted much attention. The company's super-smart business segment achieved $7.6 billion in revenue through the deployment of cloud services and computing power infrastructure.
However, for the AI industry as a whole, Anthropic's disclosure is probably even more important — a customer that has committed to investing $600 billion and is still seeking more resources means there will be huge demand for chips, data centers, and cloud computing power over the next few years, and Google, Amazon, Microsoft, Nvidia, and Broadcom will all benefit. Investors will soon be able to take a deeper look, as Anthropic is preparing for a potentially sensational initial public offering (IPO), which could value the company as high as $2 trillion. However, the latest report shows 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. Strong demand for AI is the root cause of this sense of urgency. The computing power arms race has evolved from a simple technical competition to a deep binding of capital and supply chains.
This series of data shows that the competitive landscape of the AI industry is undergoing a fundamental shift. Anthropic's plight isn't an isolated phenomenon; it's a microcosm of the entire industry struggling with computing power bottlenecks. When cloud service giants are both vendors and competitors, the political economy of computing power allocation becomes more complex. Anthropic tried to avoid risk by locking in long-term contracts, but changing market dynamics made this strategy full of variables. As the IPO approaches, the market will re-evaluate the company's true value and potential risks. This not only concerns the fate of Anthropic itself, but will also affect the capital flow and technology evolution path of the entire AI ecosystem. In the next few years, the scarcity of computing power resources will continue to drive up industry costs, and may also spawn new business models and technological breakthroughs.