Nvidia, a California-based graphic processing unit maker, is going to build a world-class artificial intelligence research center in Taiwan to accelerate its Omniverse project, a computing platform that supports 3D applications.
The announcement was made after the United States firm’s share price surged 24.6% to US$380.6 in a single day on May 24. On Tuesday, the valuation of Nvidia hit $1 trillion, making it the first US chipmaker to join the trillion-dollar club.
Because of this, Nvidia’s founder and chief executive Jensen Huang quickly gained the nickname “One-trillion man” in Taiwan, on top of his reputed status as the “godfather of AI.”
Nvidia’s shares had dropped by 50.3% to $146.14 at the end of last year from $291.11 a year earlier as the US kept tightening the export of high-end chips to China. Last August, the US government banned Nvidia from selling its A100 and H100 chips to China and Russia.
The shares have rebounded by 166% so far this year as the company vowed to invest in AI technology.
Nvidia will hire 1,000 people and invest up to TWD24.3 billion (US$790 million) in its new AI research center, or AI University, which will be jointly managed by the National Taiwan University. It has secured a subsidy of TWD6.7 billion from the Taiwanese government for this project.
Chip war takes a toll
Meanwhile, Huang told the Financial Times that further escalation of the chip war between China and the US would cause enormous damage to US companies .
He said China will make more semiconductors itself if it can’t buy them from US companies. He said US lawmakers should be thoughtful while regulating or they will hurt the technology sector.
Huang also told a global media roundtable at the Computex Taipei industry expo on Tuesday that existing chip makers should continue to work hard to stay competitive and not underrate China’s ability to catch up in the industry.
He said China will cultivate its own chip companies amid the US sanctions, and that is why many GPU startups have been created in the country.
Currently, key Chinese GPU and AI chip makers include MetaX, Birentech, Enflame and Horizon. Many of these companies outsource their chip production to overseas foundries such as Taiwan’s TSMC.
Last October, media reports said Washington asked TSMC not to produce high-end chips for Birentech after the Shanghai-based firm claimed that its BR100, a 7 nanometer GPU chip, is faster than the A100 in AI processing.
China’s markets
In August 2022, Nvidia’s A100 and H100 chips were added to a US export control list as the government said the products could be routed to or utilized by a “military end use” or “military end user” in China and Russia. Nvidia was also barred from shipping its DGX, an AI server, to China if a unit contained one of the two chips.
At the same time, the US government also restricted sales of AMD’s MI250 Accelerator AI chip to China.
Last November, Nvidia said it would relocated its regional warehouse from Hong Kong to Taipei. It also introduced to the Chinese markets the A800, which is similar to the A100 but works at 400 gigabytes per second while the A100 operates at 600 gigabytes per second. Due to this difference, the A800 satisfies the US government’s export requirements.
Nvidia has a 60–70% share in the global GPU market while some cloud service providers, which develop application-specific integrated circuit chips, seize over 20% of the market, according to TrendForce, a Taiwan-based industry data provider.
Nvidia will be able to maintain its market dominance with a strong demand of its A100 in the US and A800 in China, as well as the growing demand arising from the development of chatbots and AI computations, said TrendForce analysts. Globally, GPU chip shipments will grow 46% year-on-year in 2023, they added.
Currently, Nvidia can still sell or ship high-performance computing hardware to China by obtaining a special export license. However, it remains possible that Nvidia will be banned from selling its A800 to the country one day.
Anton Shilov, a columnist at Tom’s Hardware, wrote in March that Nvidia will lose hundreds of millions of dollars in revenue if it cannot get licenses to sell the A800 and other products to Huawei and Inspur. He said it will not be easy for the company to replace these two Chinese customers with others.
His comments came after Inspur, the world’s third largest server maker, was added to the US Department of Commerce’s “entity list” in early March this year. Huawei has been on the list since 2019.
Two-year preparation
Nvidia has spent two years preparing for the launch of its Taiwanese AI project, which will help Taiwan nurture AI talents and offer a platform for academic use, said the Ministry of Economic Affairs. The project will be implemented within the five years ending March 2027, it said.
Separately, the company said Monday that it will build an AI cloud supercomputer in Israel for several hundred million dollars. The facility, supported by 800 technology startups and tens of thousands of software engineers, is expected to commence operation by the end of this year.
In a two-hour speech at 2023 Computex Taipei on Monday, Huang told an audience of 3,500 people about Nvidia’s AI development plan.
He said Taiwanese electronic makers can use Nvidia’s Omniverse, Isaac Sim and Metropolis to build virtual factories, simulate robots and conduct automated inspections, respectively. He also displayed how AI technology can be applied in online games, content generation and music production.
Read: China leads US in tech that matters most: report
Follow Jeff Pao on Twitter at @jeffpao3