The next phase of AI spending will be decided in inference, and that is where Korea’s chip industry sees an opening. Economists at a state export bank in Seoul have now sized the prize for the rest of the decade.
Their projection puts AI chips used in data centers at $860B by 2030, lifted by the shift away from training and toward neural processing units, which are leaner and cheaper than the graphics accelerators that dominate today.
Today’s market is close to a one-company story. Nvidia took more than three quarters of it last year. Google, AMD and Intel each held less than 5 percent, and Huawei sat just above 2 percent.
Korea fields two challengers, FuriosaAI and Rebellions. Both are preparing second-generation NPUs, one of which entered volume production earlier this year while the other is slated to launch within months.
The institute’s assessment is blunt: domestic designers have the engineering, but trail leading vendors by more than three years on capital, commercial experience and software ecosystem.
Its recommendations point to the state. Tax relief or subsidies for buyers who pick Korean chips would give late entrants a home market to learn in. Abroad, the report singles out the Middle East, where export control risk, the cost of Nvidia hardware and limits on power and land are pushing governments to look elsewhere.
