Seoul-based AI chip startup FuriosaAI has closed a $620 million Series D funding round at a $2.9 billion valuation, roughly four times its $735 million valuation in early 2025. The round was led by South Korea’s National Growth Fund with $280 million, alongside Infinitum Partners, Jericho Capital, Worldline, QUDA Capital, Abu Dhabi’s NeoFort, and existing backers Naver, Korea Development Bank, and Kakao Investment.
FuriosaAI began mass-producing its RNGD inference chip in January and is scaling production with TSMC from 20,000 chips this year to more than 40,000 in 2027. Inference chips — the processors that run trained AI models rather than training them — have become one of the hottest corners of the semiconductor market as companies look for alternatives to the dominant suppliers.
The distinction between training and inference matters for understanding this deal. Training a large AI model is a one-time, enormously compute-heavy job, but inference — actually answering user prompts — happens millions of times a day, every day, and it is where most AI computing demand now sits. Chips designed specifically for inference can be cheaper and more power-efficient for that job than general-purpose hardware, which is why a wave of startups has entered this space even though the market is dominated by one very large incumbent.
For buyers, more suppliers means more choice and, eventually, more negotiating power. For the industry, it means the AI hardware story is no longer a single-company story. FuriosaAI’s challenge now is execution: turning funding into shipped chips, and shipped chips into repeat customers. The production ramp it has announced is the number to watch over the next year.
Why it matters
The size of the round and the speed of the valuation jump show how hungry investors still are for credible challengers in AI hardware. A fourfold valuation increase in under two years is rare even in AI, and the participation of a national growth fund signals that South Korea sees AI chips as strategic infrastructure, not just a startup bet. If FuriosaAI hits its production targets, it would give cloud providers and enterprises another supplier at a time when AI computing capacity is a bottleneck for the whole industry.
For readers following the chip market, the interesting question is not whether AI hardware stays hot — demand for inference capacity shows no sign of slowing — but which suppliers survive the transition from funded startup to reliable vendor. FuriosaAI’s new war chest buys it time and factory capacity; converting those into long-term contracts is the harder part, and the next year of production figures will tell that story.
Source: ChainCatcher, reported October 29, 2026.