$NVDA's first commercial Groq chip is now in production — and it's coming from Samsung, not TSMC.
Groq 3 LPX racks are in full production and rolling out to Nebius data centers this year. This is the first real product from NVIDIA's ~$20B Groq acquisition.
What's notable: while NVIDIA's flagship GPUs are built at TSMC, the Groq 3 LPX is being manufactured at Samsung Foundry. Jensen confirmed Samsung's involvement publicly earlier this year.
For Samsung, this is a meaningful expansion — moving beyond HBM and server memory into producing a key AI accelerator for NVIDIA.
For the broader market, it's another data point in the shift toward purpose-built silicon optimized for real-time inference, not just training. Different workloads, different economics, different supply chains.
Watch how the inference buildout plays out. It's not just about who makes the biggest GPU anymore.
Groq 3 LPX racks are in full production and rolling out to Nebius data centers this year. This is the first real product from NVIDIA's ~$20B Groq acquisition.
What's notable: while NVIDIA's flagship GPUs are built at TSMC, the Groq 3 LPX is being manufactured at Samsung Foundry. Jensen confirmed Samsung's involvement publicly earlier this year.
For Samsung, this is a meaningful expansion — moving beyond HBM and server memory into producing a key AI accelerator for NVIDIA.
For the broader market, it's another data point in the shift toward purpose-built silicon optimized for real-time inference, not just training. Different workloads, different economics, different supply chains.
Watch how the inference buildout plays out. It's not just about who makes the biggest GPU anymore.