Artificial intelligence company Anthropic has disclosed that roughly 80 per cent of its planned $518 billion infrastructure commitments are either non-cancellable or require payment regardless of how much computing capacity is actually used.
In its prospectus, the company told investors that securing sufficient computing power has become a major constraint on AI development. Anthropic expects demand for advanced AI systems to grow beyond available computing capacity, making access to infrastructure a key factor in future expansion.
Under long-term agreements covering the next seven to 10 years, Anthropic expects to spend at least $111.1bn with Google, $110bn with Amazon and $31.4bn with Microsoft, with payments required regardless of actual usage under the terms disclosed.
The company also has around $161.2bn in equipment lease obligations linked to Broadcom, most of which are described as non-cancellable.
Anthropic said its commitments to Google run from April 2026 through July 2033, while its agreement with Amazon extends from May 2026 to April 2036.
Under the Google arrangement, Anthropic said it would have to cover any shortfall if its actual spending does not reach the agreed level. Similar provisions apply to its Amazon deal.
The company’s Microsoft commitment, covering November 2026 to May 2033, is described as non-cancellable except in cases where Microsoft fails to remedy a material breach of its obligations.
Anthropic added that its Broadcom equipment leases are also generally non-cancellable, except in the event of a default.
The filing also revealed a growing relationship with Elon Musk’s xAI, under which Anthropic could spend as much as $84.5bn on Nvidia-based computing capacity through 2029. Unlike some of its other commitments, these arrangements can largely be cancelled with 90 days’ notice.
Anthropic also disclosed an expanded partnership with Advanced Micro Devices (AMD). The chipmaker has agreed to purchase up to $5bn of Anthropic shares and provide computing capacity expected to exceed $20bn.
The scale of Anthropic’s infrastructure commitments is broadly comparable with OpenAI’s $500bn Stargate initiative, which involves OpenAI, SoftBank, Oracle and MGX.
Anthropic said it is gradually moving away from relying solely on cloud providers and is investing more heavily in dedicated data centres and directly leased computing hardware.
The company identified its dependence on Amazon, Google and Microsoft as a potential business risk. All three provide infrastructure and distribution services to Anthropic while simultaneously developing their own competing AI models.
Anthropic noted that these companies occupy several roles at the same time including investors, customers, cloud providers, distributors and competitors which could create competing interests.
The company warned that any reduction, repricing or termination of third-party computing access could negatively affect its operations, financial position and business performance.
