Lambda Competes for Reserved HGX Contracts

Lambda moving into HGX H100 and HGX B200 clusters means it is no longer just the cheaper place to rent a few top tier GPUs, it can now bid for the same reserved training contracts that used to default to CoreWeave. For big model training, the deciding factor is not only chip brand, it is whether hundreds or thousands of identical GPUs are wired together in Nvidia’s reference design with fast InfiniBand networking and sold under long term reservations.1 2 3 4

• In late 2023, an Iambic Therapeutics buyer evaluating H100 HGX clusters found Lambda and CoreWeave highly comparable on spec and interconnect quality, with the final decision largely coming down to price. That is the clearest evidence that Lambda had entered CoreWeave’s old lane of bespoke large training clusters.3
• This is a real shift from the earlier market split, where CoreWeave skewed toward customers reserving 1,000s of GPUs on yearly contracts, while Lambda was framed as the more flexible option for growth stage companies. Adding HGX H100 and B200 clusters compresses that segmentation and pushes the two clouds into more direct head to head selling.2 4
• The economics also change with the product mix. Large HGX clusters are sold less like self serve cloud and more like infrastructure leases, with 18 month to multi year commitments, reserved capacity, and custom storage, security, and Kubernetes setups. That creates stickier revenue and makes Lambda look more like a pure training supercluster provider than a developer utility.1 3

From here, the fight shifts from who has some H100s to who can secure the next wave of Nvidia supply, power, and anchor contracts. As Lambda concentrates on large reserved clusters and wins customers that care about reference architecture and price per GPU hour, the market should look less like CoreWeave serving the top end alone and more like a two player race for frontier training demand.1 2 3 4