Anchor Contracts Enable Financeable GPU Infrastructure

Anchor contracts turn a GPU cloud from a spot market seller into a financeable infrastructure business. When Lambda signs multi year deals for dedicated clusters, it can buy GPUs and build out power, networking, and storage against committed usage instead of hoping short term demand shows up. The tradeoff is that a few very large customers start to matter more than thousands of smaller ones, so winning or losing one account can move revenue growth materially.1 2 3 4

• Lambda has already moved in this direction. By April 2026 it had landed NVIDIA and Microsoft as anchor customers, including a $1.5B four year NVIDIA deal for 18,000 GPUs and a separate multibillion dollar Microsoft deployment agreement for tens of thousands of GPUs. Those are the kinds of commitments that justify ordering hardware at supercluster scale.1 2 5
• CoreWeave shows both sides of the model. Its long term contracts helped it scale to $2B of 2024 revenue and finance more GPU purchases, but Microsoft also made up 67% of 2025 revenue, and the company said concentration with a small number of top customers is likely to continue. The same structure that improves visibility also creates counterparty risk.3 6
• This marks a step up from Lambda's earlier business. Customer interviews describe reserved 18 month plus training clusters where buyers want fixed access, custom networking, and lower per GPU pricing than AWS. That is very different from selling single instances on demand, and it naturally favors a smaller set of large research labs, model companies, and hyperscalers.4 7

The market is heading toward a split. A few neoclouds will look more like data center landlords for frontier labs and hyperscalers, while smaller providers stay focused on self serve and startup demand. For Lambda, the upside is faster scaling and cheaper capital. The next challenge is broadening beyond a handful of giant tenants without giving up the predictability those tenants bring.1 3 4 7