Luminal
Category: AI Infrastructure
San Francisco AI infrastructure startup building a search-based ML compiler and inference cloud that runs models fast on GPUs, ASICs and CPUs. Luminal was founded in 2025. The company is led by Joe Fioti. Based in San Francisco, CA, United States. Team size: 1-10. Total funding raised: $30.3M. Latest round: Series A. Key investors include Standard Capital, Felicis Ventures, Y Combinator, SagaVC, Liquid, Paul Graham, Guillermo Rauch, Ben Porterfield.
- Founded
- 2025
- Headquarters
- San Francisco, CA, United States
- Team size
- 1-10
- Total funding
- $30.3M
Value proposition
Cut the cost and raise the throughput of AI inference, and reduce dependence on NVIDIA, by compiling models ahead of time into optimized code for any hardware.
Products and solutions
Open-source Luminal ML compiler (GitHub), Luminal Cloud (managed serverless inference with scale-to-zero and automatic batching), On-prem/licensed deployment, Inference OS for scheduling across heterogeneous compute.
Unique value
Treats kernel optimization as a search problem: tests mathematically equivalent ways of executing a model on a chip and selects the fastest, with no hand-engineering. Claims 2-3x faster than existing inference engines (3.2x vs vLLM on GPT-OSS 120B, 8xH100 SXM, per the company's own benchmarks).
Target customer
AI developers and companies running high-volume, continuous inference, such as AI coding agents, voice agents and image-editing tools; enterprises needing on-prem deployment.
Industries served
AI infrastructure, AI coding agents, voice AI, image/generative media
Technology advantage
Search-based compiler with graph-level IR and rewrite rules; hardware-aware fusion, tiling and scheduling; heterogeneous compute across GPU, ASIC and CPU; optimized implementations are reusable per model-hardware pair; can re-run on new chips to get fresh kernels.
How they differentiate
Compiler-first, ahead-of-time compilation rather than runtime interpretation, combined with its own inference cloud. Targets hardware-agnostic performance across GPUs and ASICs (including Trainium) rather than NVIDIA-only optimization.
Main competitors
Standard Kernel, RadixArk (SGLang), vLLM / TensorRT-LLM, Fireworks AI, Baseten
Key partnerships
Amazon (expanding partnership, a large share of workloads committed to AWS Trainium chips), Y Combinator (S25 batch)
Notable customers
Positron AI
Major milestones
Founded 2025, YC Summer 2025 batch, open-source compiler launched on GitHub, $5.3M seed in Nov 2025 led by Felicis, public cloud launched after the seed, $25M Series A led by Standard Capital on Nov 12, 2026. CEO Joe Fioti confirmed as current CEO via Axios Pro (Nov 12, 2026) and his Nov 2026 LinkedIn post announcing the Series A.
Growth metrics
About 90 trillion tokens processed per month on Luminal Cloud within four months of its public launch (company-reported, Nov 2026).
Market positioning
Challenger in AI inference optimization, positioned as a hardware-agnostic compiler plus cloud alternative to NVIDIA-centric inference stacks.
Geographic focus
United States / global
About Joe Fioti
Ex-Intel chip designer; worked on CPU microcode and ML accelerators, including the AI accelerator shipped in Intel chips. Co-founder and CEO of Luminal (YC S25).
Latest news about Luminal
More AI Infrastructure companies
Official website: https://luminal.com