Anthropic faces $800 bn compute burden before Nasdaq IPO
Claude’s S-1 reveals rapid revenue growth, customer concentration, and governance limits that leave ordinary shareholders with little influence.

Anthropic disclosed more than $800 billion in data center commitments in an Oct. 26 S-1 filing as it moved toward a Nasdaq IPO next month under ticker ANT.
Anthropic’s Oct. 26 S-1 showed revenue of more than $33 billion in the nine months ended Sept. 30, up from $4.6 billion last year, the Financial Times reported.
The compute commitments are large even against Anthropic’s own balance sheet. They are more than seven times the company’s $112 billion cash pile, based on Atlas360 calculations, making infrastructure spending the central financial issue in the planned listing.
The filing asks public investors to buy into more than a fast-growing software company. Anthropic’s founders urged prospective shareholders “not to think of your purchase in purely transactional terms,” the S-1 said.
Anthropic, founded in 2021 by a seven-person team that left OpenAI, presents its mission as part of the investment case. The company said its work aims at the responsible development of advanced artificial intelligence for the long-term benefit of humanity.
The filing also warned that the same technology could pose “catastrophic or existential risks to humanity” without safeguards. That disclosure places safety, governance and capital intensity at the center of the company’s pitch to public markets.
The revenue base is growing quickly but remains concentrated. Five customers generated 25% of revenue in the first nine months of the year, while only 250 customers spent more than $10 million on an annualized basis, the filing said.
Growth also slowed during 2026. Revenue more than doubled from the first quarter to the second quarter, then rose almost 50% from the second quarter to the third quarter, the FT reported.
Compute contracts dominate
Compute, meaning chips, servers and data center capacity used to train and run AI models, is the largest financial strain. Anthropic spent just over $2.50 on computing power and infrastructure for each $1 of revenue in the first quarter of 2025, compared with less than 50 cents in the second quarter of 2026.
That improvement helped margins. Anthropic reported an 8% operating margin and $1 billion of operating income for the three months ended June 30, compared with a $1.2 billion operating loss in the same period a year earlier.
The company cautioned that results may vary sharply from period to period because of compute spending. It also said rapid adoption to date may reflect early-stage demand, a risk for investors extrapolating current growth.
Anthropic said compute represented 55% of revenue in the first half of 2026, down from 201% a year earlier. That ratio will be a key test of whether revenue growth can outrun infrastructure costs.
The largest disclosed commitment is with Broadcom, which co-develops TPU AI processors with Google and leases them to Anthropic under deals worth $161 billion over five years. From 2028, Anthropic expects as much as $150 billion of additional spending on TPUs.
Anthropic also expects more than $20 billion of spending on AMD chip systems, with an option that could roughly double the total. AMD will invest as much as $5 billion in Anthropic as chips are delivered, the FT reported.
Cloud and infrastructure contracts extend the exposure. Google, Amazon and Microsoft are set to provide at least $269.4 billion of cloud compute over the next decade, while Nscale and Lambda agreements total as much as $86 billion.
The governance structure adds another test for public shareholders. The FT reported that ordinary shareholders would have almost no power, while the founding team said decisions may conflict with investor expectations because profit maximization will be weighed against the public good.
Anthropic’s SpaceX agreements are worth as much as $120.5 billion through 2029 and will cost about $3.1 billion a month once all capacity is available. The company expects that capacity to come online in December.
Anthropic’s planned Nasdaq IPO next month will test whether public investors accept a fast-growing AI business with heavy compute obligations, concentrated customers and limited ordinary shareholder control.
In December, investors can check whether the SpaceX capacity comes online as Anthropic expects. Once fully available, those agreements are expected to cost about $3.1 billion a month.
From early next year, either Anthropic or SpaceX can walk away from the agreements on 90 days’ notice. From 2028, Anthropic expects up to $150 billion of additional TPU spending.
Source: Financial Times article citing Anthropic S-1, Oct. 26, 2026.
More stories
- Anthropic faces $265.6bn purchase commitments in IPO push
- Anthropic seeks IPO after $16.3bn revenue, $9bn AI spend
- Brent crude falls more than 6% as Trump weighs Iran deal
- Ireland's transport budget deal targeted 20% fare-cut reversal
- EU faces budget fight before France and Spain votes
- Hyperscalers face up to $1 trillion AI depreciation by 2030
- Fed renovation fight puts Powell's board role in focus again
- UK bond yields hit 6% as global debt selloff widens again
Latest news
- GE Vernova lifts 2026 free-cash-flow outlook to $16bn-$17bn
- South African parties split over Cape Town's 30,000-home annual gap
- Kinross Gold faces Q3 test after 2%-3% output cut
- Ruto seeks Russian investment in Kenya's $380m trade push
- Sudan teachers seek 20% budget share with 8m children outside school
- Spain rail network faces scrutiny after 64 killed near Toledo
- Private equity firms face scrutiny over third of methadone clinics
- Spain reports at least 30 dead in high-speed train crash

