Meta won’t slow AI spending despite DeepSeek’s breakthrough

Meta CEO Mark Zuckerberg on Wednesday said it’s too early to tell what impact DeepSeek’s breakthrough AI model will have on his company and the broader tech industry. 

Still, Zuckerberg poured cold water on the notion that Meta’s overall AI spending will soon drop now that DeepSeek has allegedly proven that cutting-edge AI requires far less money and computing resources than previously thought.

“It’s probably too early to really have a strong opinion on what this means for the trajectory around infrastructure and CapEx,” Zuckerberg said. “There are a bunch of trends that are happening here all at once.”

The technology markets were reeling this week over an AI model created by DeepSeek, which is a Hangzhou-based AI lab linked to the Chinese hedge fund High-Flyer. DeepSeek claims it only took two months and less than $6 million to develop its R1 large language model. That news caused Nvidia shares to tank 17% on Monday because of the implications that companies don’t need as many of the chip-maker’s graphics processing units for their own artificial-intelligence projects.

Zuckerberg discussed DeepSeek during a call with analysts Wednesday as part of Meta’s fourth-quarter earnings. The company reported that sales rose 21% year over year to $48.39 billion, ahead of Wall Street expectations.

Meta’s shares rose nearly 2% on Monday, indicating some investors believe DeepSeek’s cost-efficiency could mean the social media company may be able to create powerful AI systems for less money. Meta announced last week plans to invest between $60 billion and $65 billion this year as part of its AI push. The company on Wednesday said its 2025 total expenses would come in between $114 billion to $119 billion.

Zuckerberg said Meta is still digesting some of DeepSeek’s feats and his team hopes to eventually implement some of those advancements for their own AI projects.

“That’s part of the nature of how this works, whether it’s a Chinese competitor or not,” Zuckerberg said, referring to how one company’s advances end up benefiting others.

While it’s possible that companies will eventually require less computing resources to train these powerful models, Zuckerberg said that having a ton of server power may be necessary when the software is actually running and performing actions, resulting in “a higher level of intelligence and a higher quality of service.”

“It’s going to be expensive for us to serve all of these people because we are serving a lot of people,” said Zuckerberg, whose company said it had 3.35 billion daily active people in the fourth quarter.

Meta will continue to learn from technologies like DeepSeek’s and release its own open source and free Llama AI models that can also push the industry, Zuckerberg said.

“I continue to think that investing very heavily in CapEx and infra is going to be a strategic advantage over time,” he said. “It’s possible that we’ll learn otherwise at some point, but I just think it’s way too early to call that, and at this point, I would bet that the ability to build out that kind of infrastructure is going to be a major advantage.”

Zuckerberg also said that DeepSeek’s emergence validates Meta’s commitment to an open-source approach to AI, Zuckerberg said.

“There’s going to be an open-source standard globally,” he said. “For our own national advantage, it’s important that it’s an American standard.”

Watch: Meta beats on top and bottom lines.