Groq is in the process of securing $650 million in new funding from its current investors, according to sources cited by Axios. The company is doubling down on its inference neocloud business, which relies on its proprietary AI chip and systems. In December, Groq entered into a deal with Nvidia—described as something short of an outright acquisition—valued at a reported $20 billion. As part of that arrangement, several senior Groq employees moved to the chip giant, and Groq licensed its hardware technology to Nvidia.

Axios reports that the deal proved advantageous for Groq’s investors, who received cash payouts. Had it been a full acquisition, it would have marked Nvidia’s largest purchase to date. Now, those same investors are being asked to contribute additional capital to support the company’s plans to expand its inference cloud business. This service allows developers and enterprises to host applications that require heavy inference processing—the computational work that takes place after an AI prompt is given. In the current AI landscape, inference is a far more pressing need than model training.

The new direction is being led by Groq’s interim CEO, Adam Winter, and interim CFO, Matt Eng. In some respects, the $650 million funding round appears all but assured. According to Axios, existing backers Disruptive and Infinitium have committed to covering the entire round if other current investors choose not to take their pro-rata shares.