Nvidia paid Groq $20 billion and took its top engineers. Now Groq is raising $650 million for what's left.

Groq is raising $650 million from existing investors to support its inference cloud business, six months after Nvidia’s $20 billion not-acqui-hire. Nvidia paid Groq’s investors in cash, hired several senior engineers, and licensed Groq’s hardware technology without absorbing the company. Disruptive and Infinitium will cover the round if other investors do not take their pro-rata shares, effectively guaranteeing funding. Adam Winter serves as interim CEO and Matt Eng as CFO. Groq’s inference cloud enables developers and enterprises to host inference-heavy applications on its proprietary Language Processing Unit hardware. Inference demand is expanding because every AI prompt, response, and agent action requires inference compute. Groq’s LPU architecture targets this workload, and its tokens-per-second performance has benchmarked above Nvidia GPU inference at comparable price points. Groq has shipped chips to multiple model providers and cloud customers.

Read at TNW | Artificial-Intelligence