AI Chip Market Grows 7-Fold in 6 Years, but Nvidia Runs Away With It
The market for AI semiconductors for data centers is set to grow roughly sevenfold over six years, with Nvidia holding the top spot at a 78.2% market share. According to a recent report by the Export-Import Bank of Korea's Overseas Economic Research Institute, the market is projected to grow from $124 billion in 2024 to $860 billion in 2030, an average annual growth of 38%. As AI spreads beyond search and recommendations into autonomous driving, robotics, healthcare, and manufacturing, demand for computing power is swelling along with it. In particular, demand is growing even faster in the 'inference' stage—the process of generating real-time responses for users—after AI has completed its training phase.
The market landscape is dominated by U.S. companies. Nvidia's share last year stood at 78.2%, followed by Google at 4.7%, AMD at 4.1%, Intel at 3.7%, and China's Huawei at 2.6%. Nvidia maintains its high market share not only because of chip performance but also because of an ecosystem spanning software, development tools, and libraries.
The niche domestic latecomers are eyeing is the inference market. As the center of gravity in the AI industry shifts from training to inference, a growing recognition is that not all computing needs to be handled by expensive GPUs. As AI services scale up en masse, data center operators now factor in not just chip prices but also power consumption and cooling costs, boosting demand for low-cost, low-power inference chips (NPUs).
FuriosaAI and Rebellions are the leading contenders. FuriosaAI, founded in 2017, began mass production of its second-generation AI chip, the 'RNGD' (Renegade), earlier this year. Rebellions, founded in 2020, plans to launch its second-generation product, the 'REBEL 100,' in the second half of this year. Both companies are pursuing a strategy of entering the high-priced GPU-dominated structure by leading with NPUs (neural processing units) that handle specific-purpose AI computations with less power and at lower cost.
The Overseas Economic Research Institute analyzed that while domestic companies have secured chip design capabilities, they lag more than three years behind leading companies in capital strength, commercialization experience, and ecosystem building. In reality, supplying products to data centers requires mass production capacity, quality control, integration with customer systems, and software compatibility verification. There is also the remaining challenge of persuading customers whose data centers are built around the Nvidia ecosystem.
Accordingly, the report emphasized the government's role. It proposed offering tax benefits or subsidies to companies that use domestic AI chips to reduce early adoption risks. It also suggested creating initial demand from the public and private sectors to open opportunities for actual data center supply and performance verification, as well as diplomatic support to expand touchpoints with overseas data center operators and AI companies.
