Korean chipmakers with core design capabilities confront a widening commercial gap in the booming data centre AI semiconductor market, with critical strides needed in the next three to five years to secure a foothold amidst fierce global competition.
Korean chipmakers may have the core design capabilities to compete in AI semiconductors, but they still face a wider commercial gap that could decide who survives the next phase of the market, according to a report by the Export-Import Bank of Korea’s Overseas Economic Research Institute. The institute argues that the decisive window is the next three to five years, as the global contest over data centre AI chips accelerates and the market shifts from research training to real-world inference workloads.
The scale of the opportunity is large. The institute estimates that the data centre AI semiconductor market will expand from about $124 billion in 2024 to $860 billion by 2030, a rise that would make it one of the fastest-growing segments in the technology industry. Its forecast implies an average annual growth rate of 38%, underscoring how quickly the economics of compute infrastructure are changing as generative AI becomes more widely deployed.
At present, Nvidia remains far ahead of rivals. The institute said Nvidia accounted for 78.2% of the data centre AI semiconductor market last year, with Google on 4.7%, AMD on 4.1%, Intel on 3.7% and Huawei on 2.6%. Even so, the report sees room for late entrants as demand moves towards inference, the process of generating responses from trained models, where cost, energy use and deployment flexibility become more important than raw training performance.
That shift helps explain why neural processing units are drawing increasing interest. NPUs are designed specifically for AI inference and typically consume less power and cost less than graphics processing units. In South Korea, companies including FuriosaAI and Rebellions are positioning themselves in this segment. FuriosaAI, founded in 2017, began mass production of its second-generation chip, Renegade, earlier this year, while Rebellions plans to launch its second-generation REBEL 100 in the second half of this year.
Yet the institute says technical promise alone will not be enough. It concludes that Korean firms are still at least three years behind global leaders in areas such as funding, customer reach, software support and practical experience in data centre deployment. That matters because the AI chip market depends not only on chip specifications, but also on software compatibility, steady supply and proof that customers can run the products reliably at scale.
For that reason, the report argues that government policy should move beyond research support and help create actual demand. It suggests tax incentives or subsidies for companies that adopt Korean-made AI semiconductors, reducing the risk for early customers. It also calls for closer government-to-government cooperation to help domestic chipmakers enter foreign supply chains, especially in the Middle East, where countries are increasing investment in AI infrastructure and looking for alternatives amid US export controls, high Nvidia prices and limits on power and available sites.
The report’s broader warning is that the structure of the market could harden quickly. If the next three to five years determine which suppliers secure real customers, software ecosystems and long-term contracts, then Korean companies may need to move from development to commercial execution far faster than they have so far. In a market increasingly shaped by energy costs, infrastructure constraints and access to deployment partners, timing may be as important as technology.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
