Remember Meta CEO Mark Zuckerberg’s infamous corporate belt-tightening of 2022 and 2023 when he preached “efficiency” and cut thousands of jobs? These days, it’s just a memory, as the company is currently on an AI spending spree. Wall Street seems just fine with that—at least for now—as Meta beat its Q2 earnings expectations with strong digital ad revenue, and the company’s stock is up about 4% today.
Meta’s eye-popping AI spending is mostly going towards the infrastructure that supports training and running AI models—such as servers, data centers, and a reported 600,000 Nvidia GPUs (which each cost roughly $30,000 each). It all adds up to the tune of $37 billion to $40 billion in expected 2024 capital expenditures, while the company said infrastructure spending “will be a significant driver of expense growth next year” as well.
Zuckerberg focused some of his earnings call comments on explaining why this AI infrastructure shopping binge will pay off with increased revenue from the company’s various apps, new AI experiences, and shaping its continued work on the metaverse.
The bottom line? Training more and more advanced AI models is expensive: Zuckerberg pointed out that while Meta’s newly released Llama 3 model is “already competitive with the most advanced models,” the company is already starting to work on Llama 4, which means planning for the computing power it will need going forward.
“The amount of compute needed to train Llama 4 will likely be almost 10x more than what we used to train Llama 3—and future models will continue to grow beyond that,” said Zuckerberg on the earnings call. “It's hard to predict how this will trend multiple generations out into the future, but at this point, I'd rather risk building capacity before it is needed, rather than too late, given the long lead times for spinning up new infra projects.”
Not surprisingly, there were no comments given or questions asked by Wall Street analysts during the earnings call about the environmental impact of Meta’s AI infrastructure efforts. But there was plenty to discuss about the potential ROI of AI investments. Zuckerberg said that improving existing Meta products with AI would drive the most results in the short-term—over the next two years. “I think there’s a lot of upside” to improving recommendations and advertising experiences, he explained.
But he warned that making money on new-gen AI products—like Meta AI, the AI chatbot available on Facebook, Instagram, and WhatsApp —will take a lot longer.
“We have a relatively long business cycle of starting a new product, scaling it to something that reaches a billion people or more, and only then really focusing on monetizing at scale,” he said. When it comes to products like Meta AI or AI Studio, the company’s new make-your-own AI chatbot offering which rolled out this week on Instagram, he added, “I don't think that anyone should be surprised that I would expect that that will be years, right?”
Of course, Wall Street has recently worried about the massive AI spending of other Big Tech firms and how long it would take for those investments to pay off, including at Microsoft, whose stock dropped after a narrow cloud earnings miss this week.
But for now, it seems like as long as its AI-driven digital ad revenues stay strong—and keep in mind, the artificial intelligence used for Meta’s ad business uses the same traditional machine learning techniques they have used for years, not generative AI models—investors will remain patient, even as they wait for the potential future profits to roll in from the promise of gen AI.
Sharon Goldman
Want to send thoughts or suggestions to Data Sheet? Drop a line here.
