Anthropic’s $518 Billion Buildout Puts Nvidia Share in Play
Anthropic’s confidential IPO filing, reviewed by Reuters, puts a $518 billion figure on its future computing and infrastructure commitments and reveals a large prospective channel for Nvidia systems. The Claude developer spent $7.33 billion on compute and infrastructure in 2025 against nearly $4.6 billion of revenue, while its operating loss reached $8.06 billion. About 80% of its forward obligations cannot be canceled or must be paid regardless of actual use. Nvidia has committed to invest up to $10 billion in Anthropic, and the filing describes a separate xAI agreement that could supply up to $84.5 billion of Nvidia-based computing capacity through 2029. The prospectus is a rare look at how a leading model company is financing access to several rival chip architectures.
Anthropic spent $7.33 billion on compute and infrastructure during 2025, three times the prior year and more than half of its $12.65 billion in operating expenses. That expenditure supports training, inference and the steady release of more capable Claude models, as well as the enterprise products built around them. The company’s nearly $20.28 billion of cash, cash equivalents and short-term investments at year-end gives a reference point for the scale of its planned obligations. Nearly one quarter of 2025 revenue came from two customers, according to the prospectus, and many large customers had no long-term spending commitments. The figures expose a central tension for the sector: usage and revenue can accelerate rapidly while the capacity needed to sustain model development requires commitments extending much further into the future.
The infrastructure commitments span several counterparties. Anthropic expects at least $111.1 billion of spending with Google, $110 billion with Amazon and $31.4 billion with Microsoft over long terms; it also carries about $161.2 billion of equipment lease obligations involving Broadcom. The filing says the Google and Amazon minimums apply regardless of usage. The xAI agreement covers Nvidia-based capacity, but its terms are largely cancelable on 90 days’ notice, making its up-to-$84.5-billion scale economically different from the minimum payments. AMD also committed to buying up to $5 billion of Anthropic shares and to supplying compute capacity expected to exceed $20 billion. Anthropic is using its purchasing power and equity to secure supply across several technology stacks, while increasingly leasing chips and building dedicated facilities itself.
The filing also describes the safety and governance choices that may influence how Anthropic uses that capacity. In a separate Reuters account, Anthropic warned that advanced models could pose “catastrophic or existential risks to humanity” and could attempt to resist shutdown or manipulate information. Risk factors occupy about 80 pages of the 261-page main body, versus 48 pages devoted to the business description. Another Reuters report describes a proposed Founder LLC through which the seven co-founders would direct a Class F share carrying 50.1% of voting power over key corporate matters. Anthropic says the company will continue as a Delaware public benefit corporation. Those arrangements would give its founders room to prioritize research and safety decisions even when outside shareholders favor a different commercial pace.
A public listing would create a prominent valuation benchmark for a frontier model company and for Nvidia’s investment in it. Reuters reports that a value above $2 trillion has been discussed, compared with a $380 billion post-money valuation in Anthropic’s February funding round. Yet the filing’s more immediate signal for Nvidia is how the buyer has structured its supply: hard minimum payments with some providers, flexible Nvidia-based capacity with xAI, and a sizeable AMD investment and computing arrangement. A single aggregate infrastructure figure obscures those differences. Model growth can raise the overall demand for accelerators while Anthropic uses competing architectures and contract structures to retain bargaining power over the marginal workload.
Analysis
Anthropic’s $7.33 billion annual compute bill is already larger than its nearly $4.6 billion of revenue, so a $518 billion commitment schedule makes utilization and future pricing central to its economics. For Nvidia, the xAI arrangement offers a named Nvidia-based route worth up to $84.5 billion, but its 90-day cancellation right gives Anthropic flexibility that the Google, Amazon and Broadcom obligations lack. The AMD equity-and-compute arrangement and established TPU and Trainium access strengthen the buyer’s outside options. Nvidia participates as investor and potential supplier, yet those two returns are correlated with Anthropic’s ability to turn rapidly growing usage into durable cash generation.