Lambda wins for custom GPU training
Lambda customer at Iambic Therapeutics on GPU infrastructure choices for ML training and inference
This split shows that GPU cloud buyers are not choosing between identical servers, they are choosing between cheap, configurable compute and a polished enterprise stack. In practice, more raw means a customer gets lower per GPU pricing and a provider willing to custom build clusters, storage, networking, and support workflows, but with less of the always available, deeply integrated cloud experience that AWS and Azure sell. Lambda sits in that tradeoff and wins training workloads because many teams value price and hardware flexibility more than full platform completeness.
-
At Iambic, Lambda and CoreWeave beat AWS and Oracle not just on price, but on willingness to configure the exact InfiniBand interconnect and HGX style setup needed for large synchronous training jobs. That is what raw looks like operationally, more custom work, closer to the metal, and fewer standard boxes.
-
This market has segmented around that tradeoff. CoreWeave has gone upmarket into very large reserved clusters for enterprises, Lambda has focused on more flexible growth stage customers, and Together AI sells a higher level API and developer layer on top of rented GPU capacity for startups that do not want to manage infrastructure directly.
-
Lambda’s positioning is also visible in its customer workflow. Teams use Lambda for long reserved training clusters that can cost $500,000 to $1M per month, then keep inference on AWS where they need instant capacity, S3, EC2, Terraform, and stronger uptime guarantees. The raw provider handles bespoke compute, the hyperscaler handles production plumbing.
The next step for this part of the market is turning raw GPU access into a simpler researcher cloud without giving up the price advantage. If Lambda can keep the custom cluster flexibility that won early training customers, while packaging it into easier self serve tooling, it can move from being the cheaper alternative to being the default training cloud for a broad middle of the AI market.