Many companies vying to develop South Korea's free public artificial intelligence service have yet to finalize their consortiums, even as the application deadline for the "AI for All" project draws near. Uncertainty over the project's scope and future budget is prolonging startups' deliberations over whether to join.
According to reporting by Seoul Economic Daily, the lead bidders — SK Telecom, KT and LG Uplus, along with Kakao and ESTsoft — had still not completed recruiting participating firms ahead of the application deadline on the 18th. Companies in the AI, cloud and platform sectors are weighing offers to join from multiple lead bidders and are holding out on their decisions until the last minute.
One startup official said the company had received offers from nearly every consortium since the application period opened but had yet to commit. "There are even consortiums scrambling to recruit additional members after some participants dropped out at the last moment," the official said. The head of another startup said the company had decided this week not to take part after more than three weeks of discussions. "Forming a consortium seems to be taking far longer than it did for the standalone foundation model project," the executive added. As a result, several consortiums will confirm their members only at the end of this week and then prepare their execution plans and submission documents in a three-day cram over the weekend.
The slow pace of consortium formation stems from a lack of clarity over each participant's roles and responsibilities. Unlike the standalone foundation model project, which had the clear goal of developing a homegrown foundation model, "AI for All" must carry out a range of tasks, including launching a free general-purpose AI chatbot, providing public-sector AI agents and identifying specialized AI services. The required capabilities are broad — spanning platform and cloud construction, agent and service development, and security — and each lead bidder demands a different division of roles and responsibilities. An official at one AI firm said the company had declined to participate. "We specialize in data building and refinement, but they are proposing roles with a completely different focus, such as platform construction," the official said. "The government answered most of our questions by saying it would leave the design to each company's discretion, so responsibilities and roles felt ambiguous."
Cost and staffing burdens are another reason companies are hesitant to join. This year, the government will supply up to 512 Nvidia B200 graphics processing units (GPUs) for service development, but it offers no other support such as labor costs. Another variable is that the service operating funds to be provided from next year will fluctuate depending on the government budget. Because substantial costs are required for AI inference, servers and power, security and monitoring, customer support and model upgrades, companies say it is difficult to fix each firm's roles and cost-sharing levels while the scale of support remains undetermined. One company head said small firms and institutions already taking part in the standalone foundation model project lack the capacity to commit their limited staff to a new project. "There are several cases where they gave up on joining 'AI for All' even though they wanted to," the executive said.
