Lambda Shifts Toward Few Giant Clients

Lambda is becoming less like a broad self serve GPU marketplace and more like a contract infrastructure vendor whose growth depends on winning and keeping a handful of giant accounts. The clearest sign is its shift away from selling hardware and inference tools, and toward multi year training clusters for Nvidia, Microsoft, and frontier labs, where a single contract can fill thousands of GPUs and drive a large share of revenue.1 2 3

• The business mix has moved upmarket fast. Lambda shut down its workstation and server business in August 2025, then deprecated its inference products to free GPU capacity for long duration training deals. That makes utilization more predictable, but ties growth to fewer counterparties with much bigger bargaining power.1
• The comparable is CoreWeave. Its revenue scaled on the back of Microsoft, which accounted for about 60% of revenue in 2024, showing how neoclouds can grow extremely fast once a hyperscaler or frontier lab outsources overflow demand. Lambda is now following the same playbook with Nvidia and Microsoft anchor deals.1 3
• At the product level, Lambda still wins many training workloads because it can offer cheaper reserved clusters, strong interconnects, and engineering flexibility that AWS could not match for some buyers. But as customers get larger, decisions start to look less like developer tool purchases and more like negotiated capacity procurement.2 4

The next phase is likely even more concentrated. As AI labs and hyperscalers lock up the biggest training clusters, Lambda's upside comes from becoming a trusted overflow and specialty build partner for those buyers. That can drive rapid revenue growth, but it also means a few renewal decisions will matter more than thousands of small customers ever did.1 3