Applied Compute, a startup that helps companies build AI of their own, said on Monday, Nov. 2, that it has raised $350 million. Investors valued the whole company at $3.25 billion. Its founders say that less than 18 months after they started it, its sales have reached a run rate of $100 million: what a full year would bring in at its current pace.
The company's value has more than doubled since April, when it raised $80 million at a valuation of $1.3 billion. The new round was led by the investor Elad Gil. Seventeen other investors took part, including Kleiner Perkins, Lux Capital, NVIDIA, Blackstone, Shopify Ventures and HubSpot Ventures.
Its founders say they started the company because they believed businesses would want to own and improve the AI they use. The company wrote in April that the most capable models aren't built with a company's own data and don't understand how that company's business works. Open-weight models are ones whose trained files are published, so anyone can download them and run them on their own computers. The founders say these models have improved quickly, giving companies a strong base to build on with their own data and expertise.
Applied Compute says Microsoft, Shopify, Mercor, Harvey, DoorDash, Cognition, Clay and Decagon use it, and that it serves more than twice as many companies as at its last round. For Harvey, one of those customers, it says it trained a model for Harvey's Review Table product that improved the quality of its answers and citations. Applied Compute says that model cost less than half as much to run, meaning to answer requests, as one of the most capable models available. For Clay, Applied Compute says it built a custom system that sends each request to a suitable model, cutting running costs by up to 47% without losing quality.
Applied Compute's platform, which the company says it introduced three months ago, puts training a model, running it and improving it over time in one place. The company says customers have used it to train models more than 1,700 times since July. The company says it will spend the new money on more computing power, more staff and faster rollouts at large companies.
The revenue figure, the customer results and the cost savings all come from Applied Compute itself. A run rate is a projection, not a year of actual sales. Applied Compute has not said whether it makes a profit.