Nvidia owns the market for AI training chips. So why is the company in talks with a Korean startup most people have never heard of?
Because the AI chip market is bigger than training, and that is where the next fight is at.
Rebellions builds neural processing units for data centers, designed for AI inference - the lower-compute, higher-volume side of AI that runs models after they have been trained. It is one of several startups trying to carve out a slice of a market Nvidia has dominated.
That focus is deliberate. Training a model takes months and costs millions. Inference happens every time the model is used - constantly, at massive scale. It is the part of the AI market that keeps growing as adoption spreads.
Jensen Huang, Nvidia's chief executive, sat down with Sunghyun Park, who co-founded and runs Rebellions, at Nvidia's headquarters in Santa Clara, California, this week. The discussions are early and could take various forms: a technology partnership, an investment, or a full purchase. As of August 21, 2026, no deal had been done, and the talks could fall apart.
Nvidia did not comment. A Rebellions spokesperson declined to comment.
This would not be Nvidia's first step of this kind. In late 2025, it made a similar arrangement with US startup Groq. Nvidia took a nonexclusive license to Groq's technology and absorbed most of its engineering staff, while Groq kept running its cloud business independently. Nvidia did not call it an acquisition.
Rebellions is based in Bundang, South Korea, and was founded in 2020. It has raised about $850 million from backers including SK Hynix, Samsung Ventures, and Arm Holdings, plus direct investment from the Korean government.
Markets move every day, and most of it never reaches you in time. Get the free Market Briefs daily newsletter and stay ahead of it.
Its latest valuation is about $2.3 billion.
The company focuses on inference chips, which makes sense. Training AI models is expensive and slow, but running them - inference - happens constantly and at massive scale.
That level of demand is where challengers see room to grow.
A deal would face two sets of regulators. In the US, large tech purchases typically need approval from bodies like the Department of Justice, and Nvidia already faces takeover scrutiny.
In South Korea, the bar could be higher. The country treats semiconductors as critical strategic assets.
New investment projects from Samsung Electronics and SK Hynix, the major memory-chip makers, are generally handled in alignment with the South Korean government's policy priorities. A deal involving a homegrown AI chip company would almost certainly draw government review.
The talks show that the AI chip fight is no longer just about the big names. Nvidia still controls a large part of training, but the inference side is where challengers are placing their bets.
If Nvidia is looking to lock in technology and talent from a smaller rival, it means competition is real enough to matter.
For investors, that pattern is worth tracking. Nvidia has now reached for two smaller chip startups in about a year.
That suggests it sees enough competition coming that it would rather bring rivals close than fight them. If you hold AI-related stocks, or are considering betting to get in, the direction of these talks could tell you a lot about where the AI chip market is headed.
The deal may never happen. But the fact that Nvidia is having this conversation at all is a sign that AI chips are no longer a one-company story. The next chapter might not belong to the usual names.
The money story behind the headlines is usually the part that matters. Join Market Briefs free and get it every morning.
