NVIDIA is building Taiwan’s “first national AI supercomputer” and opening a new office in the country, CEO Jensen Huang said during his Computex 2025 keynote speech on Monday. The world’s leading supplier of AI chips will be collaborating with Foxconn, TSMC, and the Taiwanese government for the supercomputer facility, which will house 10,000 of its latest Blackwell GPUs. Foxconn will provide the AI infrastructure, while TSMC will use the system for advanced research and development.
“Taiwan doesn’t just build supercomputers for the world,” Huang said in his speech, delivered the day before Taiwan’s Computex tech expo. “ We’re also building AI for Taiwan. Having a world-class AI infrastructure in Taiwan is really important.”
“By building this AI factory with NVIDIA and TSMC,” Foxconn CEO and chairman Young Liu said in a statement, “we are laying the groundwork to connect people in Taiwan as well as government organizations and enterprises such as TSMC to accelerate innovation and empower industries.”
Huang said his company’s engineering staff had grown “beyond the limits of (its) current office” in Taiwan. As a result, NVIDIA intends to acquire a larger office in northern Taipei dubbed the “NVIDIA Constellation.”
These announcements come at a time of turbulence in the AI industry after President Donald Trump imposed sweeping tariffs on countries with which the U.S. has a trade deficit. While his stated intention was to encourage domestic manufacturing, such policies may disrupt tech supply chains in the short term. In addition, the U.S. is blocking sales of certain advanced chips to China without a license, which NVIDIA predicts could cost it up to $5.5 billion.
In response, NVIDIA has sought to diversify its strategy by expanding operations in Taiwan while also reinforcing its presence in the US. One day after Trump warned that semiconductor tariffs are coming in the “very near future,” the company said it would be building supercomputers in the U.S.
Last week, the U.S.-based chipmaker unveiled plans to sell 18,000 of its high-performance Grace Blackwell AI processors to Saudi Arabia as part of a nascent alliance that combines American tech with Saudi capital. In April, NVIDIA also confirmed that it will not hit the brakes on its plans to build more AI data centers, even after the tariffs sparked concerns that rising tech prices will reduce demand for AI.
SEE: U.S. & China Slash Tariffs in ‘Very Good Deal,’ Markets Soar
NVIDIA also unveils an open AI ecosystem, updated robot software, and a compute marketplace
The supercomputer project and new Taipei office were only part of NVIDIA’s broader slate of announcements this week. The company also introduced NVLink Fusion, a new open server platform that allows tech companies to co-design AI chips and infrastructure. The system gives companies greater flexibility to build custom AI systems using NVIDIA’s high-speed NVLink interconnects, broadening the company’s reach in data center deployments.
NVIDIA also unveiled the first update to Isaac GR00T, its foundation AI model for humanoid robots, designed to help them learn and adapt in real time. The latest version, Isaac GR00T N1.5, includes GR00T-Dreams, a simulation tool that creates virtual training environments to accelerate robotic learning.
Finally, NVIDIA introduced DGX Cloud Lepton, a compute marketplace that allows developers to rent and scale powerful AI infrastructure on demand. NVIDIA also announced a new line of DGX personal AI supercomputers, launched in partnership with Acer, Asus, Dell, and other major hardware manufacturers.
