Applications for government-funded graphics processing units for artificial intelligence research exceeded the planned supply by more than three times, according to data obtained by a lawmaker's office. The surge reflects soaring demand for high-performance computing from smaller companies, universities and research institutes that cannot secure GPUs on their own as generative AI spreads.
In the first round of the government's GPU support program for industry, academia and research institutes, 514 projects applied for a combined 13,712 GPUs, according to data submitted by the National IT Industry Promotion Agency to the office of Rep. Kim Sun-min of the Rebuilding Korea Party on the 5th. Only 4,224 units were allocated, or 30.8% of the amount requested. The government had initially planned to provide about 4,000 units, meaning applications topped the planned volume by more than threefold.
On a project basis, only 160 of the 514 applications were selected, a selection rate of 31.1%. About seven out of every 10 applications failed to win support. Competition for GPUs intensified as applications far outstripped the planned supply.
The program is designed to support AI research and service development at small and medium-sized firms, venture companies and startups, as well as universities and research institutes, that cannot afford to build expensive GPU infrastructure themselves. The government is distributing 10,000 GPUs secured through a supplementary budget last year across the industrial sector, industry-academia-research projects and national AI initiatives. After allocating about 4,000 units this year primarily to industry, academia and research institutes, it plans to distribute 3,000 more for the same group and another 3,000 for the development of a homegrown AI foundation model.
With applications in the first round alone exceeding the planned volume by more than three times, how far the government can expand the program to meet surging demand has emerged as a key task. The need for broader support is expected to grow particularly among smaller firms, universities and research institutes with limited capacity to buy their own GPUs. An NIPA official said GPUs that go unused by selected projects will not be automatically handed to applicants who were not chosen in this round, but will instead go through a separate recruitment and evaluation process.
"The fact that demand from industry, academia and research institutes has exceeded the government's support by more than three times means high-performance GPUs are desperately needed in AI research and in the field," Kim said. "The government should not stop at one-off support but should boldly expand the scale of GPU support in line with surging demand to underpin AI competitiveness."
