AIBRIEF
Rebellions CEO Sunghyun Park met Jensen Huang at Nvidia HQ this week. Talks are early and may not produce a deal, but it's the second time in roughly a year Nvidia has reached for an AI inference chip challenger.
Jensen Huang sat down this week in Santa Clara with Park. The talks are Nvidia's second reach for an inference-chip challenger in roughly twelve months.
Rebellions builds neural processing units (NPUs), chips aimed at AI inference, the run-time phase where an already-trained model takes in new data and returns an answer. That phase is distinct from training, the heavier compute step that built the model in the first place, and it is now eating a growing slice of total AI compute spending. It's also the part of the stack where Nvidia's training-era dominance offers the least insulation.
The talks span technical cooperation, an investment, or a possible acquisition, Bloomberg reported, and are at an early enough stage that they may not produce a deal. Rebellions last disclosed a roughly $2.3 billion valuation and has raised more than $850 million from investors including SK Hynix, Samsung Ventures, and Arm Holdings. Rebellions has not publicly confirmed the discussions, and its newsroom shows no statement as of writing.
Last year Nvidia cut a licensing deal with Groq, another inference-chip startup, instead of treating it as a rival to swat. Two reaches in twelve months says the same thing: the next AI compute war is inference, and the company that owns training is not guaranteed to own the runtime.