Anthropic has struck a $35 billion compute agreement with Lambda, the Nvidia-backed GPU cloud provider, according to The Information. The report, which cites people familiar with the arrangement, puts Anthropic among the most aggressive buyers of AI compute on the market. If the numbers hold, this is one of the largest infrastructure commitments any AI lab has made public.
Here’s what we know and why it matters.
What happened
The Information reports that Anthropic, the company behind the Claude family of models, agreed to a deal worth roughly $35 billion with Lambda. Lambda runs GPU cloud infrastructure and counts Nvidia among its backers, which means this arrangement keeps a huge slice of AI training and inference demand flowing through Nvidia’s ecosystem.
A deal this size isn’t about a single model launch. It’s a multi-year bet on having enough chips to train the next generations of Claude and to serve the growing wave of enterprise customers running Anthropic’s models in production.
Why this is a big deal
Compute is the constraint that shapes everything in frontier AI right now. Model quality, release timelines, and how many customers you can serve all trace back to how many GPUs you can secure and power.
What stands out here is the diversification. Anthropic already leans on Amazon, its largest investor and cloud partner, plus Google Cloud and Google’s custom TPU chips. Adding a $35 billion Lambda commitment tells you Anthropic doesn’t want to depend on any single provider for capacity. When chips are scarce, spreading your bets across multiple suppliers is how you avoid getting boxed in.
It also signals confidence in demand. You don’t lock up tens of billions in compute unless you expect to fill it. Anthropic’s enterprise business, driven heavily by Claude’s strength in coding and agentic workflows, has been climbing fast. This deal reads as the company building ahead of that curve.
The context you need
Big AI labs have been signing eye-watering infrastructure deals all year. OpenAI has committed to massive capacity through Microsoft, Oracle, and its Stargate project. Meta and Google keep pouring capital into their own data centers. The pattern is clear, the leaders are racing to secure compute before their rivals do.
Anthropic’s Lambda deal fits that pattern, but with a twist. Lambda is smaller and more specialized than the hyperscalers. Landing a customer of Anthropic’s size is a major validation for Lambda, and it shows that GPU-focused clouds can compete for the biggest contracts, not just the leftovers the giants don’t want.
For Nvidia, the arrangement is another win. Every dollar routed through a Nvidia-backed provider reinforces the company’s grip on the AI hardware layer, even as Google pushes its own TPUs and Amazon builds its Trainium chips.
What to watch next
A few things worth tracking as this develops:
- Confirmation and terms. The Information’s report is based on sources, and neither company has laid out the full structure publicly. Watch for how the spending is phased and over how many years.
- Power and timing. Compute deals live or die on electricity and data center availability. The real question is when this capacity actually comes online.
- Funding. Commitments of this scale usually pair with fresh capital. Anthropic has been raising at a rapid clip, and a deal like this hints more may be coming.
- Ripple effects for smaller clouds. If Lambda can win Anthropic, expect other specialized GPU providers to pitch themselves harder to the frontier labs.
The takeaway
Anthropic just made a statement about how it plans to keep pace at the frontier, buy compute early, buy it big, and don’t rely on one supplier. A $35 billion commitment to a Nvidia-backed cloud is both a growth bet and an insurance policy against scarcity.
For anyone building on Claude, the practical read is simple. Anthropic is investing to make sure capacity keeps up with demand, which should mean more headroom for the models and products you depend on. For the wider industry, it’s one more sign that the compute arms race is nowhere near cooling off.
More details are available in the original report from The Information.