Nvidia Negotiates Partnership or Acquisition of Korean Startup Rebellions, Valued at $2.34 Billion

Jensen Huang met this week with Sunghyun Park, co-founder and CEO of Rebellions, at Nvidia's headquarters in Santa Clara, California, for discussions covering technical partnership, capital investment, and potential acquisition of the South Korean AI inference chip startup, according to Bloomberg on August 21. The negotiations are preliminary; Nvidia and Rebellions had not publicly commented on the matter by the time this article went to press.

Rebellions and Its Inference Chips

Founded in 2020 in Pangyo, South Korea, Rebellions develops NPUs aimed at LLM inference in data centers, positioned as an energy-efficient alternative to general-purpose GPUs. The Atom and Atom Max chips have been in production since 2023; in March 2026, the company introduced the Rebel100, a chiplet platform, and the integrated RebelRack and RebelPOD systems for modular deployment in data centers.

On March 30, 2026, Rebellions completed a $400 million pre-IPO funding round with a valuation of $2.34 billion, more than doubling its previous valuation of $1.4 billion, obtained in a round led by Arm and Samsung Ventures. The total raised since inception has reached $850 million, with participation from SK Hynix, Mirae Asset Financial Group, and the Korea National Growth Fund, which contributed $166 million from the K-Nvidia program. The company plans to go public in the Korean market by the end of 2026.

Nvidia's Strategic Logic

Nvidia's dominance in training large AI models is well-established, but the inference segment, which executes pre-trained models to respond to production requests, faces increasing pressure from specialized hardware. Startups like Tenstorrent, founded by Jim Keller, and Groq have built architectures optimized for low latency and reduced energy consumption in inference workloads. For Nvidia, incorporating Rebellions' NPU architecture could accelerate its ability to compete in this segment without relying on adaptations of general-purpose GPUs.

Huang's week also includes a $6 billion licensing deal with Poolside AI, closed on August 20 according to Bloomberg, in exchange for access to the startup's Model Factory system for building open-source models. These two moves, at different layers of the AI chain, from development to inference deployment, suggest a strategy to consolidate intellectual property before competitors verticalize the same capabilities.

The Regulatory Hurdle in South Korea

South Korea classifies advanced semiconductors as national strategic assets. The government actively funds Rebellions through the Korea National Growth Fund, part of the K-Nvidia program, designed to cultivate domestic NPU manufacturers and reduce technological reliance on foreign suppliers. Any active acquisition arrangement invites political scrutiny that goes beyond conventional antitrust analysis. The closest precedent is the limitation imposed on Arm when it attempted to consolidate Korean semiconductor assets before its IPO in 2023.

According to analysts cited by Dataconomy on August 21, the most viable format is a technical licensing agreement or minority investment. This structure would grant Nvidia access to Rebellions' inference architecture without triggering regulatory hurdles from Seoul. A full acquisition would require approval from the Korean government, which tends to treat the matter as industrial policy, not merely a commercial transaction.

Two Markets on Opposite Sides of the Table

In the United States, the CHIPS Act of 2022 and its successors earmarked over $52 billion for domestic semiconductor manufacturing, but leave companies like Nvidia free to acquire capacity abroad. In South Korea, the public capital invested in Rebellions exists precisely to prevent strategic technologies from migrating to foreign control. The two systems are now in direct negotiation, mediated by Huang and Park.

MUFG, Mizuho, and SMBC have invested in local inference infrastructure with chips from startups like Tenstorrent and Groq, as part of a Japanese strategy to diversify hardware suppliers and reduce reliance on American manufacturers. An agreement between Nvidia and Rebellions would revise that planning: Rebellions' integration into Nvidia's supply chain would alter the risk profile of supplier concentration for these financial institutions.

The planned IPO for late 2026 adds pressure to the negotiation timeline. A minority investment from Nvidia ahead of the offering elevates the valuation benchmark for the public market; a full acquisition concludes the process before it occurs. Park and Huang have opposing incentives regarding the timeline, and this asymmetry explains both parties' discretion regarding the details of the discussions.