AILSA CHANG, HOST:
More people are using generative artificial intelligence in their day-to-day lives, which requires more computing power, a lot more. And that has sent companies and governments all around the world scrambling to build new computing clusters. The computing race has taken off in Taiwan, where NPR's Emily Feng reports.
EMILY FENG, BYLINE: I wanted to see what the artificial intelligence buildup looked like in person, so I came to the Asus headquarters in Taipei. It is swarming with engineers.
UNIDENTIFIED PERSON #1: (Non-English language spoken).
FENG: Asus, a Taiwanese electronics company, signed plans to build multiple supercomputers with the Taiwanese government as far back as 2017, well before ChatGPT entered mainstream awareness. Here's the CEO of Asus Cloud, Peter Wu.
PETER WU: It is a very interesting time here because Jensen Huang, the CEO of Nvidia, came to Taiwan to visit our minister to discuss the next generation of AI infrastructure.
FENG: Nvidia, which designs and sells the special type of chips that are crucial to AI computing all over the world. Nvidia's Huang, who was born in Taiwan, wanted Taiwan to prepare for artificial intelligence. And Taiwan was receptive because it wants what Steve Chen, an Asus executive, calls computing sovereignty.
STEVE CHEN: Sometimes called sovereign AI. You need to own your own AI. Some sensitive data you cannot just upload to the cloud.
FENG: As in building and training your own AI models on your own data. But that requires a lot of physical infrastructure. An AI model has to live somewhere, so to speak. It requires huge data centers, and it needs to be trained using immense computing clusters. And so Taiwan is hoping to build its own computing power and not just rely on the U.S. This year, Wu and Chen are among the engineers who built a supercomputer called Forerunner 1, a network of powerful computers linked together in southern Taiwan.
CHEN: Every country wants to own that technology because that is the future.
FENG: And Wu says in the next four years, Taiwan wants to expand the computing cluster to about.
WU: About 300 petaflops.
FENG: Meaning performing 300 quadrillion calculations a second.
WU: So you can say that Taiwan want to be somebody. We don't want us just to serve others to have supercomputer. We want to be somebody to use servers.
FENG: Taiwan is not the only one investing big in computing. The U.S.'s OpenAI and Microsoft reportedly plan to spend $100 billion to build a supercomputer by 2025. China's government put out a plan saying it wants 300 exaflops of computing power by next year. That's 1,000 times more powerful than the computer Asus is building in Taiwan. At a recent forum on AI in Shanghai, Chinese companies explained just how much resources they're putting into AI computing. China Telecom said they just bought 300 acres outside Shanghai to build a computing center.
UNIDENTIFIED PERSON #2: (Non-English language spoken).
FENG: The company says they have 12 buildings planned just to house all the computing racks they will need. And they will have their own power station and backup generators. Taiwan is trying to keep up in part for national security reasons. They do not want China, their competitor, to have access to their sensitive data because China wants to one day take over, even potentially invade Taiwan. But Taiwan has one advantage, says Asus' Wu. A lot of the suppliers making the components needed for AI computers are in Taiwan or run by Taiwanese people.
WU: If Taiwan cannot leverage its own servers, computing powers to do AI, yeah, soon there will be others to replace us.
FENG: To replace Taiwan in the digital world. That's if Taiwan does not build the physical computing power needed to stay relevant.
Emily Feng, NPR News, Taipei, Taiwan.
(SOUNDBITE OF HI-TEK SONG, "ALL I NEED IS YOU") Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
