Luminal Lands $25M Series A as Standard Capital Leads the Round

Luminal has raised a $25 million Series A led by Standard Capital, the investor announced on November 12, 2026. The AI infrastructure startup is expanding its compiler and hosted inference business across Nvidia, AMD and specialized chips, with a focus on extracting more useful work from hardware customers already own.

Co-founders Joe Fioti, Jake Stevens and Matthew Gunton have built two businesses around inference, the computation that produces an AI model’s output. Luminal hosts models on its own cloud and licenses its compiler to customers that already operate hardware. That gives the company a managed service for customers seeking deployment and a separate product for those trying to improve their own infrastructure.

A Compiler Built to Search for Faster Implementations

A compiler translates a model written in a framework such as PyTorch into the 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, and then emits GPU kernels or instructions for application-specific integrated circuits. Compiling ahead of execution is designed to reduce the runtime overhead of interpreting a model dynamically.

Luminal spends substantial computation during compilation to search possible implementations rather than relying only on handwritten heuristics, the founders said. Equality saturation compactly represents a search space that can contain trillions of options on a given chip, allowing the compiler to search for an optimized version. The approach is designed for increasingly specialized processors, which require more complex memory and scheduling decisions in software.

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

Luminal reports 36,000 tokens per second for GPT-OSS 120B on eight Nvidia H100 SXM GPUs, compared with 28,000 for TensorRT-LLM and 26,000 for vLLM in its website’s comparison. Those results illustrate the throughput improvement at the center of its commercial pitch, while the hosted service adds infrastructure work beyond the compiler itself.

Hosted Inference and Value-Based Licensing

Luminal Cloud provides 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.

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. Those workloads give Luminal production traffic spanning image processing and large language models.

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

For some compiler licenses, Fioti described pricing as a fixed fraction of the measured improvement over a customer’s existing baseline. In 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 the value delivered rather than to hosted usage alone.

Open-Source Origins and Expansion

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 developed 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 compiler license adds a second route for organizations whose hardware is already in place.

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 be divided increasingly between chips specialized for prompt processing and those specialized for token generation, making coordination across hardware types a larger software challenge.

The immediate hiring 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 designed for different stages of a request, where scheduling and cache reuse become part of the savings calculation alongside raw compiler performance.