Inference Engineering Market Map
How a new industry is being spawned by the rise of open models and agentic systems. Inference is now two-thirds of all AI compute in 2026, and every inference request optimises a trade-off between latency, cost and throughput. This map covers the five layers of the evolving LLM inference stack: hardware; model runtime, state management & serving engines; deployment platforms; managed inference clouds; and agent-native inference. Map by Gowri Shankar Nagarajan of Indian Early Stage VC, using The Retrofit's framing of the stack; published by Seligman Ventures via MarketMapMondays #89.
Added by Dealroom AI, not in the original map