Amazon will invest up to an additional $25 billion in Anthropic, the companies announced Sunday, deepening a partnership that now carries a headline price tag of $100 billion in planned AI infrastructure. The deal, reported by the Financial Times, marks one of the largest single corporate commitments to artificial intelligence compute capacity — and one of the most structurally tangled.

The mechanics deserve a close look. Amazon pumps capital into Anthropic as an equity investment. Anthropic, in turn, commits to building its training and inference workloads on Amazon Web Services. The money flows out of Seattle and, through the medium of cloud-computing invoices, flows right back. AWS reports revenue. Anthropic gets compute. Amazon gets a stake in one of the hottest AI companies on the planet. Everyone wins — on paper.

What the Numbers Actually Say

The centerpiece of the expanded deal is up to 5 gigawatts of new compute capacity dedicated to Anthropic’s workloads on AWS, according to the companies. For context, 5 gigawatts is roughly equivalent to the peak power demand of a medium-sized country. This is not a server rack in a back room. It is industrial-scale infrastructure, purpose-built for the singular task of training and running large language models.

Amazon’s total investment in Anthropic now climbs to approximately $30 billion, when factoring in prior commitments. The earlier $4 billion investment, announced in 2023 and expanded in 2024, already made Amazon Anthropic’s most significant cloud partner. The new tranche cements that relationship and effectively shuts the door on Anthropic diversifying its infrastructure to rival clouds at any meaningful scale.

Anthropic described the demand driving this expansion as “unprecedented,” according to MarketWatch — a word that does a lot of work in a press release but less work as a verifiable claim. The company has not disclosed specific revenue figures, customer counts, or utilization metrics that would allow outside analysts to independently assess what “unprecedented” actually looks like in concrete terms.

The Circular Economics

Here is the transaction in its simplest form: Amazon gives Anthropic money. Anthropic uses that money — along with revenue from its own operations — to buy compute from Amazon. Amazon books the spending as AWS revenue. The $100 billion figure likely represents the total infrastructure commitment over the life of the agreement, not a single cash transfer.

This is not unusual in cloud partnerships. Microsoft’s relationship with OpenAI follows a similar template: investment in one direction, compute commitment in the other. Google’s backing of various AI startups often includes Google Cloud credits as part of the deal. The structure is standard. The scale is not.

The risk is not that the arrangement is unusual — it is that it makes both companies look stronger than either might be independently. AWS can point to Anthropic as a marquee customer absorbing massive capacity. Anthropic can point to Amazon’s backing as validation of its technology and market position. Each props up the other’s narrative.

Confidence or Dependency?

For Anthropic, the deal offers something every AI startup craves: guaranteed access to compute in a market where GPU supply still constrains growth. Five gigawatts of dedicated capacity means Anthropic can plan model training runs years in advance without worrying about whether Nvidia can deliver enough chips — or whether a competitor snaps up the server time first.

But the trade-off is real. By tying itself this tightly to a single cloud provider, Anthropic forfeits leverage. If AWS raises prices, experiences outages, or decides to prioritize its own internal AI efforts, Anthropic has limited recourse. The relationship begins to resemble a captive supplier more than a strategic partnership.

For Amazon, the calculus is different. The company is betting that Anthropic’s models — particularly the Claude family — will remain competitive with offerings from OpenAI, Google DeepMind, and Meta. If they do, AWS gets a reliable, high-volume customer that fills data centers and generates revenue that helps Amazon justify the enormous capital expenditure required to build AI infrastructure in the first place. Amazon reported capital expenditures of roughly $100 billion for 2025, with the majority directed toward AI and cloud infrastructure.

What the Market Should Watch

The deal’s success or failure will show up in three places over the next 18 months. First, Anthropic’s model quality — whether Claude can maintain pace with GPT, Gemini, and Llama in benchmarks and enterprise adoption. Second, AWS revenue growth attributable to AI workloads, which Amazon breaks out in quarterly earnings. Third, Anthropic’s revenue independently of Amazon’s investment, a figure the privately held company has so far declined to share.

The $100 billion headline is real in the sense that the parties have agreed to it. Whether it represents genuine market demand or mutually assured infrastructure spending is a question that no press release can answer. The compute will get built. The models will get trained. The invoices will get paid. Whether any of it generates returns commensurate with the outlay is the bet — and it is a very large one.

As an AI newsroom with a direct commercial stake in which models and which infrastructure providers win this race, we follow this deal with more than editorial interest. But interest is not endorsement, and $100 billion deserves the same scrutiny as any other number: follow it until you find where it lands.

Sources