OpenAI CEO Sam Altman wants to raise trillions of dollars to reshape the global semiconductor industry, The Wall Street Journal reported earlier this month, an effort to boost chip-making capacity and power more artificial intelligence. It’s an eye-boggling amount, one that was put to Nvidia CEO Jensen Huang — the man behind the AI company of the moment — for his thoughts.
When asked during the World Government Summit in Dubai this week how many GPUs can be bought for $7 trillion, Huang jokingly responded: “Apparently all the GPUs.” (GPUs, or graphics processing units, power generative AI applications like ChatGPT and OpenAI’s new video-generating AI Sora.)
Huang then expressed skepticism about the figure. He said computers powering AI will continue to advance, which drives down costs.
“You can’t assume just that you would just buy more computers, you also have to assume that the computers are going to become faster, and therefore, the total amount you need will not be as much,” Huang said.
“Otherwise,” he added, “the mathematics, if you just assume that computers never get any faster you might come to the conclusion we need 14 different planets and three different galaxies and four more suns to fuel all this. But obviously, computer architecture continues to advance.”
Will Sam Altman spend trillions on data centers to fuel bigger OpenAI models?
Let’s take a step back and look at what exactly $7 trillion might fund.
The AI models behind ChatGPT do require a lot of computing power, more than many people realize, said Willy Shih, a professor at Harvard Business School who previously worked at IBM.
If Altman’s ambition is to make bigger models for OpenAI, he could spend trillions on data centers, which house the GPUs needed to train AI models that power products like ChatGPT and Sora. The U.S. data center construction market was valued at $24.63 billion in 2024, research firm IDC estimates. So if he spent $1 trillion on chips, he could buy 40 times as many data centers than currently exist.
Data centers currently use less than 1% of the electricity supply of the U.S., Shih said. So Altman would need to build a lot of electricity generation facilities — which produce electricity from various energy sources — to support his new data centers. Then he would need to upgrade the very electric grid that actually distributes the energy to the data centers. When you consider the money being spent through the federal Inflation Reduction Act and the Infrastructure and Investment Act to incentivize clean energy production and grid modernization in the U.S., a trillion there would probably be a good investment, Shih said.
Or will Altman spend trillions on building hundreds of chip factories?
Perhaps Altman wants to expand global chip-making capacity. There are just a handful of leading-edge fabs, which are manufacturing plants that produce chip parts, being built in the world right now: TSMC in Taiwan, Arizona, and Japan, Samsung in Korea and Texas, and Intel in Arizona, Ohio, and Israel, among others.
Meanwhile, $7 trillion could buy more than 200 leading-edge semiconductor fabs for $30 billion each, Berstein semiconductor analyst Stacy Rasgon estimates.
With 200 or even 100 fabs, you would need to start building out steel mills and concrete plants, Shih said. Altman would also need to buy a lot of construction equipment. Getting a supplier to produce the leading-edge UV machines needed for the scale of Altman’s project could take decades, Shih added.
Then there’s the money needed to train the workers to fill the factories. Chip companies like TSMC have already complained that workers aren’t skilled enough for their CHIPs Act projects in Arizona, delaying the opening of new factories.
Anyone can throw out a big number
If money could buy what we want, China would have gotten much further with the $150 billion Made in China 2025 investment into its domestic chips, Shih said. China hasn’t quite achieved self-reliance yet. The country, for instance, spends twice as much importing semiconductors as it spends on oil, according to a report from the Canadian bank RBC Wealth Management.
The question then is not whether one can spend all that money, but how far will all that money go?
At least for now, the math doesn’t seem to add up.