According to reporting by Seoul Economic Daily on Sept. 22, Ha Jung-woo, standing vice chairman of the National AI Strategy Committee, discussed domestic cooperation plans on the TPU business with Google Korea at a meeting of the technology innovation infrastructure subcommittee held on the 9th of this month. It is the first time Google has met directly with a senior South Korean government official to propose cooperation since it announced plans in the first half of this year to supply TPUs to external customers. Ha serves as the control tower for the government's AI policy.
At the meeting, Google expressed its intention to provide TPU-based AI infrastructure to domestic universities and research institutes and to build a comprehensive partnership spanning technology development using the chips to the use of physical AI data in areas such as robotics and autonomous driving. The company stressed that TPUs could be installed not only remotely via Google Cloud but also on servers built directly by users, known as on-premises deployment. Google also outlined plans to expand investment in South Korea's AI ecosystem and contribute to cultivating AI talent.
Industry observers believe Google has chosen South Korea as one of the strategic bases for its external TPU supply business. The company is expected to first secure early success cases at universities and research institutes before pushing for large-scale TPU supply to domestic AI data centers. Google's medium- to long-term strategy is to raise the TPU's share of an AI infrastructure market dominated by Nvidia GPUs. TPUs are known to cut AI infrastructure costs by up to 50% compared with GPUs.
The plan also aligns with the government's interests, as reducing dependence on a single vendor's GPUs remains a continuing challenge in building domestic AI infrastructure. The government has previously discussed ways to secure AI accelerators with Nvidia and AMD as well.
