Anthropic Files Prospectus as North American Tech Giants Ramp Up AI Capital Expenditure

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

On October 26 local time, AI company Anthropic publicly submitted its S-1 prospectus to the U.S. Securities and Exchange Commission. Based on the previously estimated $100 billion fundraising, if the listing succeeds, Anthropic will become the largest IPO in human history. The listing is interpreted by outsiders as "raising ammunition" for further expansion of AI capital expenditure.

At almost the same time, other AI giants are also accelerating investment in AI infrastructure: Meta launched about $35 billion in equity financing and plans to set up a market-based additional offering project of about $60 billion; SpaceX announced a GPU equipment financing lease arrangement with a total value of about $40 billion.

From project investments that start at tens of billions of dollars per deal to forward commitments whose scale exceeds trillions of dollars, a group of North American AI giants represented by the "Magnificent Seven" are further accelerating the pace of capital expenditure growth. Behind this, is it the confidence of the "certain," or the reluctance of the "gambler" to leave the table?

Anthropic plans IPO, capital expenditure draws attention

Anthropic has officially filed, and the revenue and computing power costs of this leading large-model company are attracting global market attention.

In the first half of 2026, Anthropic achieved revenue of $16.29 billion, an increase of 13 times compared with about $1.1 billion in the same period last year. Anthropic's revenue rose from about $4.7 billion in the first quarter of this year to about $11.6 billion in the second quarter, and by the third quarter revenue had reached $16.8 billion, up about 45% quarter-on-quarter.

Compared with the impressive revenue data, the company's real-money investment in computing power infrastructure is equally eye-catching.

The prospectus discloses that in the first half of the year, Anthropic spent nearly $9 billion on computing capacity and infrastructure, exceeding last year's full-year level of $7.3 billion. Compared with revenue of about $16.3 billion in the same period, computing power and infrastructure spending was equivalent to about 55% of the former.

At present, Anthropic's model training and services rely on cloud computing and chip partners such as Amazon and Google, while also expanding computing power resources it directly controls. According to reports, Anthropic is in talks with AI cloud service provider Nscale for a potential new computing power contract worth about $45 billion. Of the approximately $103 billion in contracts disclosed by Nscale, Microsoft and Anthropic together account for 85%.

Anthropic's frenzy for computing power goes far beyond this. In the prospectus, the long-term computing power obligations disclosed by the company unveil the computing power layout of this AI giant.

The prospectus shows that the company has locked in future computing power through procurement contracts and leases, and some contracts include minimum purchase amounts, leases that have not yet begun, and optional expansion quotas. Based on a rough estimate by the reporter from the prospectus: Anthropic has signed computing power contracts worth at least $657 billion, with terms extending as long as 2045; if incremental amounts under negotiation are included, the potential total scale could rise to about $750 billion.

Anthropic is not the only case of aggressive AI capital expenditure. Since the beginning of this year, North American AI companies have repeatedly raised or maintained high capital expenditure. However, at the same time, some leading companies have seen rare quarterly negative free cash flow.

According to the reporter's review, Microsoft's capital expenditure in the quarter ending in June this year was $41 billion, and it is expected to exceed $50 billion in the next quarter. Google and Amazon have respectively raised their 2026 capital expenditure plans to about $200 billion and $220 billion. In addition, although free cash flow has turned negative, Meta's third-quarter capital expenditure still reached $34.91 billion.

Real demand is the most direct reason. In recent years, global Token consumption has continued to climb. According to a forecast by third-party institution IDC, from 2026 to 2030, the compound annual growth rate of global Token consumption will reach 4822.6%, while the number of global active intelligent agents will increase from 79.4 million to 2.216 billion during the same period, a compound annual growth rate of 129.8%.

In the first half of this year, massive computing power contracts among U.S. tech giants are accelerating delivery. According to Oracle's disclosure, cloud infrastructure revenue in the quarter ending in August increased by 121% year-on-year, and remaining performance obligations jumped to $664 billion; Microsoft, Google and Amazon all said that the AI cloud services market as a whole is still "in short supply."

Investment bank Bernstein believes that current cloud market demand still far exceeds supply, and "the bottleneck has shifted from GPU supply to the physical capacity of data centers that can be powered." The institution predicts that the returns on massive capital expenditure and the financing methods are becoming the new deciding factors.

Sustained growth in objective demand has also intensified competitive pressure among the giants. Google CEO Sundar Pichai has repeatedly said that during a platform transition period, the risk of underinvestment is greater than that of overinvestment. Meta Chief Executive Mark Zuckerberg said bluntly in a podcast that this is a "survival" race where if you are not at the table, you are out, and that the cost of missing the next-generation computing platform is higher than capital investment.

However, the funding base for the pace of spending is changing. Compared with the previous model of building data centers relying on their own cash flow, equity, corporate bonds, financing leases and project companies have been used more frequently by the giants this year. Take Oracle as an example: as of the end of August, it had sold $20 billion of shares through a market-based additional offering. According to incomplete statistics by the reporter, as of July 7, Amazon, Google, Oracle and Meta had issued about $194 billion in bonds this year, significantly higher than the full year of 2025.

Of course, new variables are also emerging. SpaceX founder Musk recently said that the bond market is tightening financing for "neocloud" companies lacking non-AI revenue, and he expects large-scale industry consolidation.

But in the eyes of many industry insiders, this is not a cause for concern. On the eve of the arrival of artificial general intelligence, no giant is willing to leave the table early because of insufficient chips. After all, once artificial general intelligence arrives, the money spent on computing power investment now may become a trivial issue. news_keyword_pub,stock, iframe below the main text iframe below the main text General statement General statement QR code at the end start QR code at the end start

QR code at the end start QR code at the end start Editor name and work code

Editor name and work code