world

Chinese firms rent computing power abroad to train AI models

The United States has tightened rules on advanced chips and remote computing access, a move that could hit Chinese firms training AI models abroad.

china, south east asia, china the, chinese, nvidia, restrictions, chip, chip restrictions

Key facts

  • Firms tied to those centers will now need a US government license to buy advanced chips from Nvidia or other American chipmakers.
  • The step matters because Chinese firms currently rent computing power abroad to train artificial intelligence models, a route that grew after earlier…
  • Some of the firms behind the centers have been accused of helping China evade US export controls on advanced chips.

Firms tied to those centers will now need a US government license to buy advanced chips from Nvidia or other American chipmakers.

The step matters because Chinese firms currently rent computing power abroad to train artificial intelligence models, a route that grew after earlier US export controls.

The bureau said the measures were aimed at ensuring sensitive US technology"cannot be circumvented through remote computing arrangements involving data centers in third countries.".

Some of the firms behind the centers have been accused of helping China evade US export controls on advanced chips.

Beijing has made self-sufficiency in technology a national priority and needs the best chips to build powerful AI models.

Experts told the BBC that China still needs advanced US chips if its best AI models are to stay competitive.

The curbs pose a fresh challenge for Beijing, which has become deeply dependent on computing power from abroad in the wake of US export restrictions.

Tighter restrictions could also push Chinese firms toward domestic semiconductors and further drive advancement at home.

Singapore said it is"engaging relevant US authorities"to understand the new rules better, and is"exploring how to facilitate trusted access"for local firms.

Dig deeper

What is Nvidia?

Nvidia is a US company that designs advanced chips widely used to train large AI models, and Washington restricts sales of its most powerful processors to China.

Why does Beijing treat chip self-sufficiency as a priority?

China wants the ability to build competitive AI at home without depending on US technology that American authorities can cut off.

What does renting computing power abroad mean in this story?

Chinese firms pay to use data centers outside China that already hold advanced American chips, so they can train models without importing those chips themselves.

Why do US export controls matter here?

Earlier limits on selling top chips to China pushed firms toward overseas computing, and new licensing rules aim to stop that remote route from bypassing the curbs.

What is at stake for Chinese AI?

If access to the best US chips tightens further, including through foreign data centers, China's leading models may struggle to stay competitive unless domestic semiconductors catch up.

Lire en français