Nvidia may be looking beyond GPUs for its next move in the AI inference race.
The chip giant is discussing a possible partnership with South Korean AI semiconductor startup Rebellions that could involve a technical agreement, investment or acquisition, according to Bloomberg, citing people familiar with the matter. Nvidia CEO Jensen Huang met Rebellions CEO Sunghyun Park at Nvidia’s Santa Clara headquarters last week, although the discussions remain preliminary and may not lead to a deal.
Rebellions specializes in processors built for AI inference, an increasingly important part of the AI infrastructure market as companies look for faster and more efficient ways to run trained models.
The company merged with SK Telecom's AI chip startup Sapeon Korea in 2024 and raised $400 million in March 2026 in a pre-IPO funding round. Rebellions has raised roughly $850 million overall from investors including SK Hynix, Samsung Ventures and Arm Holdings, and was last valued at about $2.3 billion, Bloomberg reported.
Nvidia's inference push
The potential deal could fit in Nvidia's broader strategy of gaining access to promising AI technology without necessarily buying the entire company.
Nvidia used a similar structure with Groq in late 2025, securing a nonexclusive technology license while taking on most of the startup's engineering staff. Nvidia said Groq was not acquired because it continued operating independently and retained its cloud business.
Nvidia has since incorporated Groq technology into its own inference strategy, unveiling the Nvidia Groq 3 language processing unit in March. The company plans to introduce new LPU architectures alongside its annual GPU releases, Data Center Dynamics reported.
Rebellions could give Nvidia another technology foothold as demand for inference grows.
The inference bet comes with tradeoffs
The talks are notable because inference is becoming a bigger battleground in AI infrastructure, but specialized inference chips are not an easy replacement for Nvidia's broader GPU platform.
Nvidia data center chief Ian Buck described the tradeoff at the company's GTC conference in March: “The LPU is optimized strictly for that extreme low-latency token generation, offering token rates in the thousands of tokens per second,” according to Data Center Dynamics.
He added that “the economics, or the tokens per second per chip, is actually quite low.”
That helps explain why Nvidia may see value in acquiring technology and talent while maintaining flexibility over how it deploys them. For Rebellions, a deal could provide capital and access to Nvidia's ecosystem, but it could also complicate the startup's reported plan to eventually go public.
Regulation could complicate an acquisition
A full acquisition would carry additional regulatory risks. Nvidia already commands a dominant position in AI training chips, making any major purchase of a competing semiconductor company potentially subject to antitrust scrutiny in the U.S.
Korean regulators could also take interest because advanced semiconductors are considered strategically important. That consideration is particularly relevant given Rebellions' ties to SK Hynix and the Korean government's investment in the company.
No agreement has been announced, and the talks could still end without a transaction. But Nvidia’s interest in Rebellions adds to signs that the company sees specialized inference technology as an increasingly important complement to the GPUs that built its dominance in AI infrastructure.
More News: LG and Nvidia are teaming up to develop a bipedal humanoid robot for a planned 2027 unveiling, combining Nvidia’s physical AI stack with LG’s hardware and manufacturing expertise.