Lambda prioritizes multi-year training contracts

This shows Lambda choosing the business with the highest lifetime value per GPU, not the broadest product surface. Inference APIs and chat products keep developers happy, but they fragment capacity into small, bursty jobs. Multi year training contracts pack thousands of GPUs into one customer, keep utilization predictable, and create the kind of committed revenue stream that can finance giant cluster buildouts and debt backed expansion.

• Lambda had already been moving upmarket before the shutdowns. Earlier research framed Lambda as a more flexible alternative to CoreWeave, but by late 2025 it was signing much larger reserved capacity deals and shifting away from self serve products toward fewer, bigger customers with steadier demand.
  2 sacra 3 sacra 4 lambda 5 tomshardware
• The products being cut sat in a different market from Lambda’s new core. RunPod and Modal focus on developers spinning up workloads quickly, often with simple interfaces and granular billing. That market was still large enough for RunPod at $120M annualized revenue and Modal at $50M, but it rewards software convenience more than raw cluster scale.
  6 sacra 7 sacra 8 sacra 9 sacra
• The closest comparable is CoreWeave. Its growth was propelled by giant anchor contracts, especially Microsoft, that justified aggressive infrastructure spending. Lambda’s NVIDIA leaseback deal for 18,000 GPUs and separate multibillion dollar Microsoft deployment agreement point to the same playbook, turning scarce GPUs into long duration contracted capacity instead of retail cloud inventory.
  3 sacra 4 lambda 5 tomshardware 10 sacra 11 sacra

The next phase is a sharper split in the GPU cloud market. A few neoclouds will look more like infrastructure lenders and builders for frontier labs and hyperscalers, while self serve inference consolidates around software first platforms. Lambda is now clearly aiming to be judged against CoreWeave and Crusoe, not against developer tooling companies.