Luminal Raises $25M Series A Led by Standard Capital

By Daniel Okafor

Luminal has raised a $25 million Series A led by Standard Capital, the investor announced on November 12, 2026. The startup builds a compiler and a hosted inference service that run AI models on Nvidia, AMD and specialized chips. Its pitch is simple: extract more useful work from hardware customers already own.

Co-founders Joe Fioti, Jake Stevens and Matthew Gunton have built two businesses around inference. Luminal hosts models on its own cloud. It also licenses its compiler to customers that already operate hardware. One product serves buyers who want deployment handled. The other serves buyers who want to improve their own infrastructure.

What does the compiler actually do?

A compiler translates a model written in a framework such as PyTorch into low-level code that executes its mathematical operations. Luminal's technical overview describes a pipeline that lowers a model into a dataflow graph, applies hardware-aware fusion, tiling, memory planning and scheduling, then emits GPU kernels or instructions for application-specific integrated circuits. Compiling ahead of execution is designed to cut the runtime overhead of interpreting a model dynamically.

The company spends substantial computation during compilation to search possible implementations rather than relying only on handwritten heuristics, the founders said. A technique called equality saturation compactly represents a search space that can contain trillions of options on a given chip, letting the compiler hunt for an optimized version. The approach targets increasingly specialized processors, which demand more complex memory and scheduling decisions in software.

A January 9, 2026 technical post explains how automatically generated megakernels fit in. Fusing a model's forward pass into one kernel, or GPU program, removes repeated kernel-launch delays and lets later operations start as soon as their input data is ready. The design also attacks idle periods caused by uneven work across processor cores and waits for initial weights to load, using fine-grained synchronization to preserve data dependencies.

How fast is it?

Luminal reports 36,000 tokens per second for GPT-OSS 120B on eight Nvidia H100 SXM GPUs. Its website's comparison credits TensorRT-LLM with 28,000 and vLLM with 26,000. Those numbers anchor the throughput claim at the center of its commercial pitch.

Luminal Cloud adds managed serverless endpoints, automatic batching and scale-to-zero capacity, with customers paying for what they use. Licensed deployments can run in a customer's cloud or on-premises. Above the compiler, its inference layer schedules and balances work across CPUs, GPUs and ASICs, redistributes tasks as utilization changes and starts or stops nodes with demand.

The production traffic is real. Real-estate media company AutoHDR processes about 5 million images a month, and Luminal handles almost all of that workload while moving a growing share onto AMD chips, the founders said. Another customer, Freebuff, sends Luminal trillions of tokens each week.

How does the licensing work?

For some compiler licenses, Fioti described pricing as a fixed fraction of the measured improvement over a customer's existing baseline. His illustrative example: doubling throughput on a GPU costing $1.50 saves 75 cents; Luminal charges 25 cents and leaves the customer with a 50-cent saving. The structure ties license revenue to value delivered, not to hosted usage alone.

Freebuff's coding requests can contain on the order of 100,000 input tokens, making cache-hit rates and request scaling important costs, the founders said. Reusing cached computation avoids running it again. Moving relevant caches through the memory hierarchy preserves that saving. A fast kernel alone does not address those expenses.

Where did the company come from?

Luminal began as a side project Fioti developed outside his job to run models on a laptop. The project later added an Nvidia backend and an open-source community. The founders settled on keeping the core compiler open while charging for hosting, a model they compared with Vercel's pairing of Next.js and its cloud.

The Series A follows a $5.3 million seed round announced on November 17, 2025, led by Felicis Ventures, with angel investors including Paul Graham and Guillermo Rauch.

Luminal is now recruiting compiler engineers and its first non-engineering employee, a head of growth. The team works in person in downtown San Francisco and has so far been engineering-only. The founders expect inference workloads to split increasingly between chips specialized for prompt processing and chips specialized for token generation, making coordination across hardware types a larger software challenge.

The immediate plan points to deeper compiler expertise and a first commercial function. The broader test is whether Luminal can preserve its reported cost advantages as inference spreads across chips built for different stages of a request, where scheduling and cache reuse join raw compiler performance in the savings calculation.