Luminal raises $25M to reduce AI inference costs on non-Nvidia hardware

Securing new capital for growth

Luminal, a San Francisco-based AI infrastructure company, has successfully closed a $25 million Series A funding round. The investment was led by Standard Capital and saw significant participation from previous seed investors, including Y Combinator, Felicis, SagaVC, and Liquid 2. These early backers chose to double down on their initial stakes in the company.

The company plans to use this new capital to expand its engineering team and its go-to-market strategy. These efforts are intended to support a broader range of AI models and different types of chip platforms. By scaling its workforce and reach, Luminal intends to become a primary tool for companies looking to optimize their AI operations.

Since launching its public cloud in November 2025 following a $5.3 million seed round, the company has seen rapid adoption. It reached a milestone of processing 90 trillion tokens per month within just four months of its launch. This growth highlights the increasing demand for efficient ways to run large-scale AI models.

How the compiler and cloud work

Luminal provides two primary products: a search compiler and an inference cloud. The search compiler is designed to test different, mathematically equivalent ways of running an AI model on a specific chip. It then automatically selects the fastest configuration for that hardware. This process ensures that models perform at peak efficiency without requiring manual tuning by the user.

Once the compiler identifies the optimal configuration for a specific chip, that setup can be reused for repeated workloads. This makes the initial search worthwhile because it creates a permanent blueprint for efficiency. By automating this complex task, Luminal removes the need for companies to employ specialist engineers to perform extensive manual tuning on every new piece of hardware.

The inference cloud complements the compiler by allowing developers to deploy their models without having to manage the underlying infrastructure. This means that developers can focus on their applications rather than the technical difficulties of hardware management. Together, these two tools allow AI companies to switch between different hardware platforms while maintaining low costs and high speeds.

Breaking the Nvidia monopoly

A significant portion of the current AI industry relies on Nvidia’s GPU ecosystem for running models. While these chips are powerful, optimizing models for alternative hardware often requires deep technical expertise and significant time. Luminal aims to change this dynamic by making it easier for companies to adopt diverse hardware options, including application-specific integrated circuits (ASICs).

One of the company's key partnerships is with Amazon, where a large share of Luminal's tokens run on Amazon’s Trainium chips. By providing a way to run models efficiently on these chips, Luminal helps companies diversify their hardware stack. This approach allows firms to avoid being locked into a single vendor's ecosystem while still achieving high performance.

The company views this as a way to widen the hardware choices available to the broader AI market. By lowering the barrier to entry for non-Nvidia hardware, Luminal seeks to ensure that the cost of running AI does not limit who can build it. The company identifies the opportunity for this type of AI infrastructure as being worth $5 trillion.

Origins and future outlook

Luminal began its journey by building in the open on GitHub as the industry began to see a surge in AI adoption. This open-source roots approach allowed the company to understand the needs of developers early on. As demand grew for more efficient ways to run models across different types of hardware, the company pivoted to its current infrastructure focus.

Moving forward, the Series A funds will be directed toward developing both the compiler and the inference cloud. The company intends to expand its support for a wider variety of models and hardware platforms to accommodate more customers. This expansion is a key part of their strategy to become a standard tool for AI infrastructure.

The ultimate goal for Luminal is to provide a seamless experience where hardware choice is no longer a bottleneck for innovation. By providing the tools to optimize and deploy models across various chips, they hope to make AI more accessible. The company continues to work on making the infrastructure layer of AI as efficient as possible for the next wave of developers.