“Mark Zuckerberg showed me a plant where he put it over a map of Manhattan, and it was basically the size of Manhattan,” Trump told the crowd of global elites. “I said, ‘You’ve got to be kidding.’ It was miles long, miles wide, and very high. It literally covered most of the island.”

Meta’s stock jumped 1.6% after Trump’s statement, paring gains to 0.9% as of press time.

While Trump referred to the project as a “big plant,” he could have been referring to Meta’s Richland Parish “Hyperion” AI data center campus, which is currently underway in Louisiana and is set to open in 2030. The massive industrial undertaking comprises 2,250 acres—a landmass equal to roughly 1,700 football fields; the campus will house 4 million square feet of data center space, making it one of the largest silicon clusters on earth. However, that 4 million square feet is much smaller than Manhattan’s 22 square miles; raising questions about whether Trump was referring to a much larger project being built by Zuckerberg. 

Trump also marveled at the project’s $50 billion price tag, contrasting the staggering investment with traditional real estate, harkening back to his past life as a developer. That $50 billion matches the projected cost of building the Hyperion data center campus.

“If you spend $500 million, you can build a good shopping center,” Trump remarked. “But how do you spend $50 billion? When I looked at this thing, I understood why.”

Meta is currently transitioning from a social media company to an AI infrastructure giant, raising its capital expenditure guidance for the 2025 fiscal year to nearly $72 billion, 70% more than the previous year. Looking into 2026, Zuckerberg has warned of “notably larger” spending, with market analysts projecting that annual outlays could surpass $100 billion.

The president said that AI is “massive” and driving a rapid demand for energy that requires a shift in policy. Recognizing that the aging U.S. electrical grid cannot support the two-to-five gigawatts of power that these “titan clusters” require, Trump ordered the government to step aside during his speech: He told the Davos audience that he has authorized AI companies to act as their own private utilities.

So far it seems that under this new framework, tech giants will build their own on-site power plants using natural gas, coal, or oil. Trump promised to slash the bureaucratic red tape that usually bogs down such projects, making them drag on for four or five years. Instead, he pledged to deliver federal approvals for these private energy plants in just two weeks, as long as the companies build their “own electric generating plants.”

“You people are brilliant. You have a lot of money,” Trump said, addressing the tech executives in the room, egging them on to start building their own plants. 

This industrial blitz serves as the president’s primary leverage ahead of his planned state visit to China in April. Trump has argued that American leadership in AI depends on physical infrastructure rather than just software, and said Chinese President Xi Jinping respects this rapid industrial execution.

“I’ve always had a very good relationship with President Xi,” Trump said. “But we’re leading the world in AI by a lot because I’ve allowed these big companies to build their own electric capacity. We’re creating more energy than any country anywhere in the world.”

China and the U.S. are widely seen to be neck and neck in the AI race, with each country emphasizing their own strengths to best the other. The U.S. leads China in AI talent, investment, and access to semiconductors, while China has the U.S. beat on AI-related infrastructure, according to a recent Morgan Stanley report. China is just “months” behind the U.S. in terms of model quality, Demis Hassabis, CEO of Google DeepMind, told CNBC.

The same day Trump was speaking at Davos, Nvidia CEO Jensen Huang dismissed fears of an AI bubble while discussing China’s progress. The release of DeepSeek in 2023, in particular, he highlighted as a “major breakthrough.” Huang called China’s advance in this regard “a huge event for most of the industries, most of the companies around the world, because it’s the world’s first open reasoning model.” Since then, he explained, many open reasoning models have emerged, enabling researchers to create something that’s domain specific or specialized for their needs. In the U.S., Mark Zuckerberg has stood out for his own commitment to open-source AI, citing DeepSeek as a model.