Anthropic's IPO Filing Discloses Nearly $900 Billion in Forward Compute Spending

Related topics top start Related topics top end Futures promotion begin Futures promotion end Miaopai begin Miaopai end Video player start Video player end Market chart begin Market chart end

Anthropic's IPO filing shows that the company, in the course of expanding its AI business footprint, plans to invest nearly $900 billion through forward performance obligations in servers, facilities and power projects, an amount far higher than the liabilities currently shown on its balance sheet.

The AI lab behind the Claude chatbot submitted its registration filing on October 26 to prepare for a listing, with the market expecting its valuation to reach $1.5 trillion. The filing submitted to the U.S. Securities and Exchange Commission disclosed a number of details, including the size and health of its balance sheet: book liabilities of $117.5 billion and cash of $112 billion.

But investors need to study the notes to the financial statements to see the details of the company's planned spending—these are off-balance-sheet obligations covering forward leases and purchase agreements for data centers, power and graphics processing units (GPUs).

Peers such as Alphabet and Meta have likewise accumulated hundreds of billions of dollars in forward performance spending while building out AI infrastructure, and such payments are not yet included on corporate balance sheets. In recent months, the amounts of these commitments have risen sharply, triggering Wall Street concerns: this week, shares of Amazon, Alphabet and Microsoft all came under pressure, partly because investors are wary of massive AI spending.

David Trainer, chief executive of investment research firm New Constructs, said of Anthropic's committed spending (also called hidden debt) that it is an astronomical figure. "They have to invest enormous sums across every computing link—training, inference and energy."

Anthropic declined to comment further on the scale of its capital commitments, saying the relevant information is all in the registration filing, which sets out forward spending in the financial statements, with the rest scattered throughout the 300-page document. The company said in its prospectus that it will rely on revenue growth and financing to cover this investment.

The filing states: "Computing investment is a prerequisite for long-term growth," referring to the computing resources used to train and run AI models. The related investment includes capital expenditures and lease obligations, and "we must bear these expenses long before the corresponding computing facilities are put into operation and generate revenue."

Hidden debt

Anthropic's off-balance-sheet purchase agreements quadrupled from last December to June this year, estimated at $265.6 billion. Most of these service contracts are tied to cloud hosting and other infrastructure partnerships, with terms of up to 20 years.

The company also has a guarantee obligation of up to $800 million to compensate for data center development costs. Leases not yet commenced total $321 billion.

Recent new commitments have further pushed up the total forward cost. In July, Anthropic agreed to purchase GPUs and related hardware from AMD, spending more than $20 billion.

The forward spending corresponding to the remaining computing arrangements (including leases and amended agreements) is $245 billion. The company has also earmarked $21 billion for short-term cooperation projects.

But Anthropic's off-balance-sheet debt may actually be higher than the currently estimated scale of committed spending over the coming decades.

The company warned in its prospectus that payment amounts under some computing and infrastructure agreements will be very substantial, but some agreements contain variable costs, making it difficult to lock in the final amount.

For example, GPU prices, future electricity prices and fuel surcharges may all fluctuate over time.

Information blind spots

The balance sheet alone cannot provide the full picture, partly because Anthropic is changing its business model and beginning to build its own data centers. Such facilities, costing billions of dollars, can reduce reliance on third-party operators.

Patrick Badolato, an accounting professor at the University of Texas, said the shift in business model makes its performance no longer comparable with that of previous years.

Major tech companies are racing to build new AI data centers, which require expensive servers and consume enormous power, driving a spending spree among leading tech companies such as Meta, Amazon and Microsoft. Tech giants are laying off staff, issuing additional equity and selling corporate bonds, all to support the development of advanced AI models and meet expected market demand for AI technology.

Badolato said these forward obligations will ultimately turn into expenses, eroding revenue, and are pre-registered items of future cash outflows.

Similarly, leases that have not yet commenced will in the future also turn into assets and liabilities and appear on the balance sheet.

Ravi Gomatam, an accounting analyst at Zion Research Group, said: "Whether or not the company can earn revenue, these committed payments must be made." This makes Anthropic's investment risk higher than that of ordinary companies.

Gomatam said that for most companies, committed spending accounts for only a small portion of total assets.

Moreover, unlike tech startups in earlier years that mostly relied on talent, Anthropic needs expensive infrastructure such as server farms to generate revenue.

At the same time, communities across the United States are increasingly dissatisfied with the water and power consumption of data centers, threatening to block new data centers from being built. Anthropic also faces increasingly fierce competition from low-cost open-source AI models, which may seize market share. These and other risks may make it difficult for the lab to earn enough revenue to cover expected costs and recoup its investment.

Gomatam said: "But it is not an ordinary company. Betting on this company is essentially betting on its development over the next few years." news_keyword_pub,usstock,MSFT&AMZN&AMD Account opening QR code start

Account opening QR code end iframe below main text iframe below main text General statement General statement QR code at end of article start QR code at end of article start

QR code at end of article start QR code at end of article start Editor name and work code

Editor name and work code