Nvidia disclosed at a Non-Deal Roadshow on September 28 that Anthropic's contracted value across cloud providers and neoclouds now exceeds $180 billion, with 2.5 gigawatts of Nvidia AI infrastructure contracted for delivery through 2028. This is not a single contract, but a stack of cloud compute agreements, equity commitments and infrastructure deals — the first time the financial entanglement between a frontier AI lab and the chip giant has been laid out this explicitly.
What makes up the $180 billion
According to financial media coverage of the roadshow, the "contracted value" is mainly composed of the following pieces:
- Lambda: $35 billion. A six-year cloud compute agreement for roughly 350 megawatts, hosted in a Texas data center developed by Hut 8. The structure is worth noting: Hut 8 leases the site to Nvidia, Nvidia supplies chips to Lambda, and Lambda sells the compute to Anthropic — making Nvidia simultaneously a chip supplier, a cloud investor and a de facto "sub-landlord".
- Nscale: $45 billion. Also a six-year term, about 460 megawatts in West Virginia, set to run on Nvidia's Vera Rubin architecture.
- Equity bundled with cloud commitments. In November 2025, Nvidia committed to invest up to $10 billion in Anthropic, with Microsoft committing $5 billion at the same time; in exchange, Anthropic committed to $30 billion of Microsoft Azure compute.
- Separately, The Information reported in mid-September that Anthropic's total compute spending commitments could reach $517 billion — a broader figure over a longer horizon, of which the $180 billion is the portion counterparties have confirmed so far.
Why this matters
First, look at how Nvidia's role has changed. It no longer just sells chips: it invests in cloud providers, anchors data center projects as a tenant, and helps finance entire deal structures — the chipmaker's business model is evolving toward general contractor of the "AI factory". Nvidia also revealed at the roadshow that the revenue opportunity per gigawatt of AI infrastructure has grown from about $18 billion in the Hopper era to $25 billion for Blackwell and $40 billion for Rubin. What it sells is no longer just GPUs, but CPU racks, networking and the complete "AI factory" package.
Second, look at the supply side. Neoclouds are becoming the main engine of capacity expansion: Nvidia mentioned that Australian partners including IREN and Firmus are expected to bring roughly 2 gigawatts online by 2027. The geography and ownership of compute supply are diversifying, while the technology stack is converging on Nvidia's reference designs.
A note of caution for investors
The $180 billion is Anthropic's locked-in "contracted value", not Nvidia's recognized revenue. Multiple reports stress the distinction: contracts deliver over many years and are recognized as revenue annually. Oracle's earnings are the reference point — it carries $664 billion in remaining performance obligations but expects to recognize only about 13% over the next 12 months. When reading AI infrastructure numbers, watch the delivery cadence, not just the signing headlines.
For the AI industry, the real signal is this: compute is turning from a one-time purchase into a long-term financial asset, and the decisive moves in the model race are increasingly written on balance sheets. Anthropic, which is pursuing an IPO reportedly targeting a $2 trillion valuation, can count this $180 billion of locked-in contracts as a core asset in its listing story.
