Applied Compute, a San Francisco company that helps businesses train and run their own AI models, raised $350m at a $3.25bn valuation, it announced on 2 November 2026. The round was led by angel investor Elad Gil.
Kleiner Perkins, Lux Capital, Hanabi Capital, NVIDIA, Blackstone, Shopify Ventures and HubSpot Ventures took part, along with a long list of other backers named in the company's post. Neither the company nor SiliconANGLE, which covered the news, named a stage for the round, and neither said whether the valuation is pre- or post-money.
The company said its annualised revenue run rate has passed $100m, less than 18 months after it was founded. That figure is the company's own and has not been independently verified.
“Training, inference, and routing are the first pieces of a much larger stack.”Yash Patil, Rhythm Garg and Linden Li, founders, Applied Compute (company announcement)
What Applied Compute does
Applied Compute was founded by three former OpenAI researchers: Yash Patil, who is chief executive and previously worked on OpenAI's Codex programming assistant, Rhythm Garg and Linden Li. It sells a platform that lets companies build models of their own on top of open-weight foundations rather than relying only on a general-purpose frontier model.
The platform has two main parts. AC2, short for Applied Compute Agent Cloud, combines model training and inference in one system. Shift-1 is a router that sends each request to the most cost-efficient large language model and can split a complex task across several models.
The company said customers include Microsoft, Shopify, Mercor, Harvey, DoorDash, Cognition, Clay and Decagon. It gave two examples of savings. Harvey runs a custom model for its Review Table product at less than half the inference cost of a frontier alternative, and a custom router built for Clay cut production inference costs by up to 47%.
Growth since the last raise
Applied Compute said the number of enterprise customers has more than doubled since its previous fundraise and that its team has tripled since January, with 85% of staff working as researchers or engineers. It did not give a total headcount.
Usage figures in the post include more than 1,700 training runs on AC2 since July and an inference service that handles more than 1 trillion tokens a day. About 40% of customers use the company for both training and inference.
The previous round was reported by ValueAdd VC, citing TokenPost, as $80m led by Kleiner Perkins in April 2026 at a $1.3bn post-money valuation. The company's own post does not give the amount. No total raised to date was disclosed.
What the money is for
The company said it will use the new capital to expand compute capacity, grow its team and speed up deployments at large customers. SiliconANGLE summarised the aim as expanding infrastructure capacity.
In the announcement, the founders wrote: "Training, inference, and routing are the first pieces of a much larger stack." The post did not say what the rest of that stack would include or when it would ship.
The raise comes as investors back companies that sit between AI labs and corporate buyers. ValueAdd VC listed Traversal's $120m round at a $1bn valuation and Fal's $1.3bn raise at a $16bn valuation as comparable recent deals.
Company website: appliedcompute.com