Nvidia rival Rebellions taps JPMorgan for South Korea IPO

Rebellions Inc. appointed JPMorgan Chase & Co. as global lead underwriter for its initial public offering in Seoul.

Rebellions is an AI chip startup that makes accelerators for inference workloads in data centers.

People familiar with the matter said it planned to file for the IPO in the second half of this year, targeting a listing in late 2026 or early 2027.

The company focuses on inference accelerators and emphasizes cost efficiency and performance-per-watt in a market dominated by Nvidia Corp.

In September, it raised US$250 million in a series C at a US$1.4 billion valuation; its backers include Arm Holdings Plc, Samsung Ventures, and SK Hynix Inc.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Rebellions uses a South Korea supply chain edge

The IPO prospectus may spell out how closely the company works with South Korea’s semiconductor network, a point missing from the announcement.
Backers include Samsung Ventures plus SK Hynix Inc., which supplies High Bandwidth Memory (HBM) used in AI systems 1.
Nvidia Corp. leads the market, while access to advanced manufacturing plus HBM often limits AI chip output. Those ties could help Rebellions secure capacity against rivals 2.
Local design support adds more lift. SEMIFIVE helps fabless chip startups build complex chips for Samsung’s manufacturing processes 3.

An IPO could bankroll a new front in the AI hardware Cold War

The IPO also backs a push for “sovereign AI” infrastructure, meaning countries run their own AI systems rather than relying on foreign providers.
Rebellions positions REBEL-Quad as an option for governments facing GPU shortages plus total-cost-of-ownership pressure, including sovereign AI initiatives 4.
If the listing succeeds, the company expects to raise money for higher production and wider reach, with plans that include the U.S. and Europe 5.
That expansion could deepen a split in the AI supply chain, giving more countries a path to large-scale AI without depending on one nation’s hardware 4.