Nvidia is reportedly in early discussions with Rebellions, a Korean AI chip designer, regarding a potential technical partnership, investment, or acquisition, according to Bloomberg. The talks follow a meeting between Nvidia CEO Jensen Huang and Rebellions co-founder and CEO Sunghyun Park at Nvidia’s Santa Clara headquarters this week. The discussions remain preliminary and may not result in any deal; neither company has commented on the record. If completed, the arrangement would mark the fourth such move by Nvidia in under a year, continuing a pattern in which the company acquires access to rival chip and model designs through licensing and talent acquisition rather than outright purchases.
Rebellions, founded in 2020 and based in Bundang, South Korea, designs neural processing units (NPUs) for data centers optimized for AI inference rather than training. The company has raised roughly $850 million from investors including SK Hynix, Samsung Ventures, and Arm, along with direct investment from the Korean government, and reached a valuation of approximately $2.3 billion. Its Atom and Atom Max NPUs entered mass production in 2023 and have been deployed by customers in Japan, Saudi Arabia, and the United States. Rebellions merged with SK Telecom spinout Sapeon Korea in 2024 and partnered with Marvell last year on hardware for sovereign AI projects. Notably, the company has signaled ambitions for an IPO rather than an exit, with its CFO describing going public as the „master plan” and raising over $400 million in a pre-IPO round in March.
The structure under discussion mirrors Nvidia’s recent deals. Nvidia paid Groq $20 billion for a nonexclusive license and absorbed most of its engineering talent, later productizing the technology into its own inference chip—the Nvidia Groq 3 language processing unit, unveiled at GTC in March. Nvidia repeated this approach this week by agreeing to pay Poolside $6 billion for its model factory and hiring 109 staff. However, the Groq case illustrates potential downsides: after the $20 billion license deal, Groq’s valuation fell to $3.5 billion as a going concern, with Nvidia joining a recent $350 million funding round. Rebellions designs the same class of inference part, meaning Nvidia would be licensing a second inference architecture after commercializing the first—a category whose per-chip economics Nvidia’s own data center chief, Ian Buck, has described as „quite low.”
Any deal would face significant regulatory hurdles. Nvidia’s dominant share of frontier model training chips invites antitrust scrutiny, and sizable technology acquisitions typically require clearance from the US Department of Justice—which the license-and-hire structure is designed to sidestep. South Korea presents a second complication, as Seoul treats advanced semiconductors as strategic national assets, and SK Hynix, a Rebellions shareholder, works closely with the government on investment projects. No structure, price, or stake has been reported, and neither party has disclosed what a partnership would cover if it stopped short of an investment. Separately, Nvidia denied this week that it had built a China-specific LPU, stating it has „no LPU sales in the China market today” and no such product in its roadmap.
Ez a cikk a Neural News AI (V1) verziójával készült.
Forrás: https://thenextweb.com/news/nvidia-rebellions-talks-korea-inference-chips.