AI chip startup Rebellions raises $400 million at a $2.3 billion valuation ahead of IPO

The AI chip market has another contender for multibillion-dollar contracts: South Korean startup Rebellions has closed a pre-IPO round of $400 million at a company valuation of $2.3 billion. The company plans to go public in 2026 and is going after a slice of a market that Nvidia currently controls almost as a monopoly.

Rebellions was founded in 2020 by a team of engineers with experience at Samsung and major American technology companies. Unlike many competitors, the startup focused not on training neural networks but on inference, that is, running finished models. It is a pragmatic choice: inference accounts for most of the operating costs of companies that deploy AI in production. Every query to a language model and every AI recommendation in an app is inference, and its volume is growing exponentially.

The company's flagship chip is designed specifically for transformer architectures, the ones underlying GPT, Claude, Gemini and other large language models. According to the company, the solution delivers high energy efficiency on LLM queries, which is critical for data centers with limited power supply. Its customers include major South Korean banks and telecom companies interested in a local alternative to American hardware.

The $400 million pre-IPO round is one of the largest in the history of the Korean technology sector. The full list of investors has not been disclosed, but both Korean institutional funds and international venture players are known to have shown interest in the company. The $2.3 billion valuation puts Rebellions alongside AI accelerator players such as Groq, Cerebras and Tenstorrent.

Context matters: in 2025–2026 the AI chip market is in a boom accompanied by shortages and geopolitical restrictions. The US has imposed export barriers on shipments of high-performance chips to China and a number of other markets, which opens a window of opportunity for alternative manufacturers. South Korea, with its powerful semiconductor ecosystem built around Samsung and SK Hynix, is strategically well placed.

Still, the road to real competition with Nvidia remains steep. Nvidia sells not just chips but the CUDA ecosystem, which AI developers have used for years. Moving to alternative hardware requires rewriting code and accepting compatibility risks. This software advantage protects Nvidia's position more reliably than any specs of the chips themselves. Rebellions is betting on open standards and integration with popular frameworks, PyTorch, JAX and vLLM, to lower the barrier to switching.

The 2026 IPO will be an important test: a successful listing would confirm that investors are ready to pay for diversification of the AI hardware supply chain. For the industry as a whole, it is a signal that the battle for the inference market is only beginning.