At the end of last year following Nvidia’s non-takeover, non-aquihire of Groq we discussed the future of Groq’s technology. There was lots of speculation about the reason for what looked like a very expensive deal. Did Nvidia do the deal just to end a possible challenge to its ‘GPU supremacy’ from a promising start-up? I was sceptical, saying: What is the future of Groq’s technology? … As we’ve seen Jensen Huang has stated that Nvidia plans “ … to integrate Groq’s low-latency processors into the NVIDIA AI factory architecture …”. I don’t see any reason why he would make this up in an internal email. Austin Lyons has a great post explaining how Groq’s technology fills a gap for Nvidia: Let’s zoom out a bit and take a look at why adding an ultra-low-latency SKU to Nvidia’s portfolio is the right strategy … the portfolio coverage that a hyperspeed SKU gives Nvidia is important. So I think we need to take Huang at face value. Nvidia really does plan to ‘ramp-up’ deployment of Groq-derived processors as part of its AI factory architecture. And talking of ‘hyperspeed’ Nvidia and the Groq team aren’t hanging around. Jonathan Ross and key members of the Groq team already have their Nvidia badges: Something that isn’t mentioned in the Groq press release is Nvidia buying / reselling any chips (or anything else) from what remains of Groq. I think we can take it that it won’t be doing that, because it doesn’t make financial sense and / or because Nvidia wants to quickly develop its own upgraded version. In the last few days there have been some significant developments that help to support this thesis. Last Friday the Wall Street Journal reported that Nvidia plans to launch a new design based on Groq’s technology at its GTC conference in March: Nvidia plans to unveil a new processor specially tailored to help OpenAI and other customers build faster, more efficient tools, a major shake-up to its business that is poised to reset the AI race. The company is designing a new system...