Iambic's training-heavy GPU strategy
Lambda customer at Iambic Therapeutics on GPU infrastructure choices for ML training and inference
This split shows that many vertical AI companies are training heavy but inference light. Iambic spends roughly $500,000 to $1M a month on training at Lambda, versus about $50,000 to $100,000 on inference at AWS, because it serves a small number of high value scientific users rather than a mass consumer or developer API audience. That makes cheap reserved GPU clusters more important than squeezing every last cent out of inference.
-
OpenAI and Anthropic live in the opposite regime. Their products sit behind chat apps and APIs used by huge numbers of people and software products, so inference becomes the main economic bottleneck. That is why throughput, latency, and token serving efficiency matter much more for frontier labs than for a drug discovery company running a smaller number of high stakes jobs.
-
Iambic can afford to pay the AWS tax on inference because reliability matters more than unit cost. The team describes AWS as the place for always available A10G and L40S instances, S3 storage, Kubernetes, and infrastructure as code, which together make it easier to run secure production services that do not go down.
-
This also maps onto how the GPU cloud market is segmenting. NeoClouds like Lambda and CoreWeave win reserved training clusters with lower per GPU pricing and more willingness to customize hardware and interconnect, while hyperscalers win production inference when companies need mature storage, security, orchestration, and uptime.
Going forward, more vertical AI companies in law, biotech, engineering, and finance are likely to look like Iambic rather than OpenAI. That points to a two tier market, with NeoClouds optimized for long running training reservations, and hyperscalers staying strongest where inference must plug into durable, enterprise grade software infrastructure.