Mark Zuckerberg thinks China’s DeepSeek AI models have some “novel” innovations he hopes to emulate. But it hasn’t given him second thoughts about his mission to plow hundreds of billions of dollars into Meta’s AI infrastructure.
“I continue to think that investing very heavily in capex [capital expenditure] and infrastructure is going to be a strategic advantage over time,” the Meta CEO and cofounder said during the company’s Q4 earnings call on Wednesday.
Meta reaffirmed its plan Wednesday to spend between $60 billion and $65 billion in capital expenditures this year, up from $39 billion in 2024, as the company races to build the expensive data centers necessary to power new types of AI services, including a “highly intelligent and personalized assistant.” Over the long term, Zuckerberg said, Meta will spend “hundreds of billions of dollars” on AI infrastructure.
The advent of generative AI, such as OpenAI’s ChatGPT, has spurred an arms race among tech companies like Meta, Microsoft, and Alphabet, which all believe that owning the computing resources to run the most advanced AI models is critical to their future business prospects. Earlier this month, OpenAI, Oracle, and SoftBank announced plans to invest $500 billion over four years into the so-called Stargate AI data centers.
That notion was recently called into question when DeepSeek, a Chinese startup, released new AI models that rivaled the most advanced AI offerings developed by U.S. companies. But, according to DeepSeek, they were trained at a fraction of the cost. While some industry observers are skeptical about DeepSeek’s claims regarding costs, the capabilities of its AI models have nonetheless earned praise.
Zuckerberg said there were a number of novel things DeepSeek did “that we’re still digesting” and that he hoped to implement some of those advances into Meta’s own AI technology. And he noted that the fact that DeepSeek’s model is open-source is proof that Meta’s decision to make its Llama AI technology open-source was the right move.
“There’s going to be an open-source standard globally, and I think that for our own national advantage it’s important that it’s an American standard,” Zuckerberg said. “The recent news has only strengthened our conviction that this is the right thing to be focused on.”
While Zuckerberg did note that changes in the way advanced AI models are trained and run might at some point change the calculus for infrastructure investment, he said that it was still “too early to really have a strong opinion” on this based simply on DeepSeek.
Meta’s overall financial results for the last three months of 2024 beat analysts’ expectations, with revenue of $48.4 billion, up 21% year over year and ahead of the $46.99 billion expected by Wall Street. The company posted net income of $20.8 billion, or $8.02 per share versus the $6.76 EPS expected by analysts.