Luminal raises $29.2M to speed up AI inference with a compiler-first approach
What's the deal? AI infrastructure startup LuminalDealroom has a profile for this one. Try Dealroom → has filed to raise up to $29,163,244 in new funding, according to a filing with the US Securities and Exchange Commission. The exempt offering was disclosed in October 2026.
What does it do? Luminal compiles AI models into optimised native code to deliver fast, high-throughput inference. Its compiler converts models from platforms such as PyTorchDealroom has a profile for this one. Try Dealroom → and Hugging Face into efficient programs for GPUs and application-specific integrated circuits (ASICs).
What's the endgame? The company also runs a hyperscale Inference OS that schedules and load-balances workloads across mixed CPU, GPU, and ASIC clusters to cut latency and raise utilisation. It offers both managed serverless inference through Luminal Cloud and licensed cloud or on-premises deployments.
By the numbers: Luminal says its compiler-first approach beats existing inference engines by two to three times on standard benchmarks. It cites 36,000 tokens per second for GPT-OSS 120B on eight H100 SXM GPUs, against 28,000 for TensorRT-LLM and 26,000 for vLLMDealroom has a profile for this one. Try Dealroom →.
The signal: As demand for AI compute surges, the cost and speed of running models — not just training them — has become a key battleground. Luminal is betting that software optimisation, rather than raw hardware, can squeeze more performance from the chips companies already have.
Read more: intelligence360.news
