Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days

Luminal Raises $25 Million to Make AI Inference Portable Across Chips, Claiming 90 Trillion Tokens a Month

Luminal Raises $25 Million to Make AI Inference Portable Across Chips, Claiming 90 Trillion Tokens a Month

Luminal, a Y Combinator-backed startup, raised a $25 million Series A led by Standard Capital with participation from Y Combinator, Felicis, SagaVC and Liquid 2. The company pairs an inference compiler with a managed cloud, says the cloud reached about 90 trillion tokens a month within four months, and extends support to alternative processors such as Amazon Trainium.

Luminal, a San Francisco startup that builds an inference compiler and a managed cloud for running AI models across different kinds of chips, has raised a $25 million Series A led by Standard Capital, with participation from Y Combinator, Felicis, SagaVC and Liquid 2. Axios and the lead investor both put the round at $25 million; one other outlet reported $23 million, a small discrepancy worth noting in a market where round sizes are often rounded for effect.

The company was founded by Joe Fioti, who serves as CEO, together with Matthew Gunton and Jake Stevens, and came out of Y Combinator. Its core product is a compiler that takes a trained model and searches for the fastest mathematically equivalent way to execute it on a given processor, testing alternative implementations and selecting the one that runs quickest. A managed cloud service then deploys the resulting workloads for customers.

Luminal says its cloud now processes about 90 trillion tokens a month, a figure it says it reached within four months of launch. That number is company-reported and has not been independently verified, but if accurate it would put a four-month-old service in the same conversation as inference providers many times its size. The compiler approach also extends beyond Nvidia hardware, with support for alternative processors such as Amazon Trainium, which is the more strategically interesting part of the pitch.

The timing is not accidental. Nvidia has just gone through the most turbulent leadership change in its history, and inference customers are suddenly asking what happens to the software ecosystem if the incumbent wobbles. A compiler that can retarget a model from one chip to another turns hardware choice from a lock-in decision into a runtime optimization, which is exactly the kind of insurance buyers want when the supply side looks uncertain.

Luminal also enters a crowded field. The inference layer has become one of the hottest parts of the AI stack, with Baseten reportedly seeking $2 billion at a valuation of up to $30 billion and Fireworks, Together, Baseten and the hyperscalers all competing on latency and price. Most of those players optimize on fixed hardware; Luminal is betting that the compiler, not the cluster, is the durable asset.

The technical bet has real risk. Compilers that promise chip portability must chase every new accelerator architecture, and the performance gap between a hand-tuned Nvidia pipeline and a portable one is often the deciding factor for latency-sensitive products. But the round is a signal worth reading: even a $25 million check suggests investors see demand for inference that is not married to a single vendor silicon.

For AI builders, the practical takeaway is that the inference market is starting to look like the database market a decade ago: a thin portable layer on top, with cost and speed as the only differentiators that matter. If Luminal can keep its claimed throughput while adding chips, the compiler-first approach could give smaller labs and enterprises real negotiating leverage on price.

Comments (0)

Log in to join the discussion

Log In

No comments yet