Skip to content

AI: U.S.'s Project Stargate Upstaged by Chinese DeepSeek

President Donald Trump supports investing up to $500 billion on half a dozen massive machine learning data centers across the U.S. When he made the announcement at the White House on Jan. 21, he had with him OpenAI’s Sam Altman, SoftBank’s Masayoshi Son, and Oracle’s Larry Ellison. These data centers will use enormous amounts of energy, around as much as all the currently existing data centers in the U.S.

The companies plan to spend $100 billion themselves, with the goal of drawing in a total of $500 billion invested. The goal? To provide “the physical and virtual infrastructure to power the next generation of AI,” as Trump put it.

(When it comes to investments in infrastructure, consider the rapid depreciation and obsolescence of computer equipment, compared to the more durable value of physical infrastructure such as energy, transport, and water.)

But Stargate has been upstaged by the release of a large language model developed in China: DeepSeek-R1. DeepSeek is able to offer developer access to its system at prices far lower than competitors such as ChatGPT or Grok. And the model weights have been released in an open-source form, allowing people—or at least those with significant computing ability—to run it directly themselves on their own equipment. The model was released in late December, and on Jan. 22—the day after Trump’s Stargate announcement—the developers released a paper documenting the new techniques they used to develop the model while using only a fraction of OpenAI’s computing requirements.

Looks like the chip sanctions against China haven’t stopped it from progressing—although they have cost American companies like Nvidia billions of dollars in lost revenue.