Anthropic’s confidential IPO paperwork, filed with the SEC in June 2026 and reviewed by Reuters, discloses a $518 billion decade-long anthropic infrastructure buildout spread across six partners: Google, Amazon, Microsoft, Broadcom, AMD and xAI. About 80% of that sum is non-cancelable or payable regardless of usage, the prospectus shows.
In plain terms, Anthropic has committed $111.1 billion to Google, $110 billion to Amazon, $31.4 billion to Microsoft, $161.2 billion in Broadcom equipment leases, up to $84.5 billion to xAI for Nvidia-based chips, and more than $20 billion in AMD compute capacity. The bet is simple: computing power, not cash, will decide who wins the AI race.
Anthropic told investors the scale is deliberate. Future demand for advanced AI systems will likely be “limited principally by the availability of compute,” the company wrote, according to the filing. That single sentence explains why a startup with $4.59 billion in 2025 revenue is locking itself into obligations more than 100 times that size.
How Anthropic’s $518 Billion Splits Across Google, Amazon and Microsoft
The anthropic google amazon microsoft spending breakdown forms the backbone of the deal. These three contracts run seven to ten years and require payment “regardless of usage,” a phrase that appears twice in the prospectus language.
| Partner | Commitment | Contract Window | Cancellation Terms |
|---|---|---|---|
| $111.1 billion | April 2026 to July 2033 | Anthropic pays the shortfall if actual spend falls short | |
| Amazon | $110 billion | May 2026 to April 2036 | Similar shortfall-payment terms to Google |
| Microsoft | $31.4 billion | November 2026 to May 2033 | Non-cancelable except for Microsoft’s uncured material breach |
| Broadcom | $161.2 billion | Multi-year equipment lease | Non-cancelable by either party except default |
| xAI (SpaceX) | Up to $84.5 billion | Through 2029 | Largely cancelable with 90 days’ notice |
| AMD | $20 billion+ compute, $5 billion stock purchase | Ongoing | Tied to AMD’s parallel equity investment |
“If our actual spend falls short, we must pay Google the difference,” Anthropic wrote in the prospectus obtained by Reuters, adding that similar terms apply to the Amazon agreement. That single line captures the entire risk profile of this anthropic 518 billion ai deal: the company owes the money whether or not it uses the servers.
Why Are Anthropic’s Cloud Commitments Non-Cancelable?
Anthropic structured most of its Google Cloud, Amazon Web Services and Microsoft Azure deals as binding purchase commitments rather than flexible service agreements. The company frames this as a hedge against a chip shortage it sees coming, not a mistake.
The logic tracks a broader industry pattern documented in research on AI infrastructure financing: labs that lock in supply now avoid getting squeezed later when GPU capacity tightens further. Anthropic’s own computing and infrastructure spending tripled in 2025 to $7.33 billion, up from roughly $2.4 billion the year before, evidence that the compute crunch it describes is already hitting its books.
The anthropic non-cancelable cloud commitments also reveal a strategic pivot buried in the filing. Anthropic said it is shifting “from a cloud-only model toward dedicated data centers and directly leased chips,” meaning it now wants to own more of the physical stack rather than renting all of it from Google, Amazon and Microsoft, the same three companies that also build competing AI models.
What Broadcom, AMD, xAI and SpaceX Bring to the Buildout
Not every partner in this deal supplies cloud capacity. Broadcom’s role is almost entirely hardware. Anthropic carries roughly $161.2 billion in Broadcom-related equipment leases tied to custom AI chips and networking gear, obligations the filing describes as non-cancelable by either side barring default, matching what industry coverage calls the anthropic broadcom lease obligations at the center of its chip strategy, tied to multi-gigawatt TPU capacity beginning in 2027.
The xAI arrangement looks different by design. Anthropic could spend up to $84.5 billion through 2029 leasing Nvidia-based capacity from Elon Musk’s SpaceX, which absorbed xAI in February 2026. That figure nearly doubles the roughly $45 billion SpaceX itself disclosed earlier this year, when it said Anthropic agreed to pay $1.25 billion a month for access to about 325,000 GPUs across its Colossus facility. Unlike the cloud giants, this anthropic xai spacex 84.5 billion deal comes with a 90-day cancellation window, giving Anthropic an exit ramp the Google and Amazon contracts do not offer, a distinction confirmed in reporting on the SpaceX filing.
AMD’s involvement blends supply and equity. The chipmaker agreed to buy up to $5 billion of Anthropic stock while supplying compute capacity Anthropic values at more than $20 billion, a structure that ties AMD’s own upside to Anthropic’s survival.
Anthropic Revenue vs Infrastructure Spending: Does the Math Work?
Anthropic’s 2025 revenue hit $4.59 billion, a jump of roughly 1,088% year over year. Against $518 billion in future commitments, that means the company has pledged over $100 in future infrastructure spending for every dollar it earned last year.
The operating loss tells the same story from a different angle. Anthropic’s operating loss widened to $8.06 billion in 2025 from $2.98 billion the year prior, even as revenue grew tenfold. The company held $20.28 billion in cash and short-term investments at the end of December 2025, a cushion that looks thin next to decade-long obligations measured in hundreds of billions.
Customer concentration adds another wrinkle. Two unnamed customers accounted for nearly a quarter of 2025 revenue, and Anthropic warned that most of its largest clients are not locked into long-term contracts. That combination, heavy fixed costs on one side and flighty revenue on the other, is the crux of whether anthropic’s 518 billion spending sustainable question that investors will have to answer before the IPO prices.
How Does Anthropic’s $518 Billion Compare to OpenAI’s Stargate?
The anthropic openai stargate comparison is inevitable given the timing and size. OpenAI’s Stargate project, announced in January 2025 alongside SoftBank, Oracle and the Abu Dhabi fund MGX, targets roughly $500 billion in AI infrastructure spending, a figure OpenAI later said would grow far larger when combined with its other compute deals.
The structural difference matters more than the dollar figure. Stargate is a joint venture with shared ownership and multiple capital partners splitting the bill. Anthropic’s $518 billion, by contrast, is a web of bilateral purchase contracts where Anthropic alone carries the payment obligation, even though Google and Amazon are also equity investors in the company. That distinction is why the anthropic 518 billion vs openai stargate 500 billion framing understates how concentrated Anthropic’s financial exposure actually is compared with OpenAI’s shared-venture model.
The Circular Financing Risk Behind Anthropic’s IPO Prospectus Details
Nvidia is reportedly negotiating to anchor as much as $10 billion of Anthropic’s IPO, on top of the $10 billion equity commitment it made in November 2025 alongside a $5 billion Microsoft investment. Nvidia’s total contracted value with Anthropic now exceeds $180 billion once cloud deals, equity and compute leases are combined.
That overlap is exactly what critics of AI build-out financing flag as circular: a chipmaker invests in a customer, the customer spends that capital on the chipmaker’s hardware, and the resulting revenue growth looks organic even though a slice of it originated from the investor itself. Documentation of this pattern across the sector notes that Nvidia has taken similar positions in OpenAI, CoreWeave, Nscale and other AI infrastructure firms, making Anthropic one node in a much larger interlocking network rather than an isolated case.
These anthropic ipo prospectus details matter for anyone weighing the anthropic ipo valuation 2 trillion dollars figure now circulating ahead of a listing expected after the U.S. midterm elections in November 2026. A valuation that size assumes revenue keeps compounding at a pace few software companies have ever sustained, let alone one already spending more on compute than most rivals earn in total.
When This Buildout Could Backfire
Every piece of this structure depends on demand outrunning supply for years, not quarters. If enterprise AI adoption slows, or a competitor undercuts Claude on price, the shortfall clauses in the Google and Amazon contracts turn from theoretical to real, forcing Anthropic to pay for capacity it does not need.
The dependency risk cuts deeper than cash flow. Google, Amazon and Microsoft supply the infrastructure Anthropic runs on while simultaneously building rival models, an arrangement Anthropic itself flagged as a risk factor in the filing. A landlord that also competes with its tenant has every incentive to prioritize its own product when capacity runs short, regardless of what the contract says on paper.
The xAI arrangement is the one flexible lever in the entire structure. Its 90-day cancellation window suggests Anthropic’s own leadership sees a scenario where scaling back is necessary, even as the company locks arms with Google and Amazon for the next decade with no such escape hatch.
Watch how the S-1 lands on SEC EDGAR and whether the IPO price reflects these fixed obligations or glosses over them, because the gap between the two will tell you more about Anthropic’s real risk than any valuation headline.
Frequently Asked Questions
How much is Anthropic spending on AI infrastructure over the next decade?
Anthropic expects to spend at least $518 billion over roughly ten years across six partners: Google, Amazon, Microsoft, Broadcom, AMD and xAI. About 80% of that total is non-cancelable or payable regardless of actual usage, according to its confidential IPO prospectus.
How much did Anthropic commit to Google, Amazon, and Microsoft?
Anthropic pledged $111.1 billion to Google through July 2033, $110 billion to Amazon through April 2036, and $31.4 billion to Microsoft through May 2033. All three run under long-term infrastructure agreements requiring payment regardless of usage.
What is Anthropic’s relationship with AMD and Broadcom?
AMD agreed to buy up to $5 billion of Anthropic stock while supplying over $20 billion in AI compute. Broadcom, meanwhile, holds roughly $161.2 billion in equipment lease obligations that neither party can cancel except in a default.
How does Anthropic’s $518 billion compare to OpenAI’s Stargate project?
OpenAI’s Stargate targets around $500 billion, funded jointly by OpenAI, SoftBank, Oracle and MGX as a shared venture. Anthropic’s $518 billion is structured as separate bilateral contracts, meaning Anthropic alone carries most of the payment obligation.
Why is Anthropic building its own data centers instead of relying only on cloud providers?
Anthropic said it is shifting from a cloud-only model toward dedicated data centers and directly leased chips to reduce dependence on Google, Amazon and Microsoft, which supply its infrastructure while also developing competing AI models.
What are the risks of Anthropic depending on Amazon, Google, and Microsoft for compute?
These three companies control critical infrastructure and distribution for Anthropic while building rival AI products themselves. Anthropic flagged this dual role as a specific risk factor in its prospectus, noting its growth is tied to partners with competing interests.
How much revenue did Anthropic generate in 2025 compared to its infrastructure spending?
Anthropic generated about $4.59 billion in 2025 revenue while committing $518 billion to future infrastructure, a ratio exceeding 100 to 1. Its operating loss widened to $8.06 billion the same year as computing costs tripled.
What is Anthropic’s expected IPO valuation and when will it go public?
Anthropic is reportedly targeting a valuation exceeding $2 trillion, more than double its $965 billion valuation from May 2026. The listing could occur after the U.S. midterm elections in November 2026, following several delays.
