Meta has bet big on artificial intelligence. On Wednesday, it said that bet would not let up.
The Silicon Valley company, which owns Facebook, Instagram and WhatsApp, increased the lower end of its capital expenditure forecast for the year to at least $130 billion, up from $125 billion that it had projected in April. Much of that investment will go to building data centers, the computing facilities that power A.I.
Meta also said that its costs and expenses rose faster than its revenue growth for the second quarter. Revenue was up 28 percent to $60.8 billion from a year earlier, while costs jumped 55 percent to $42 billion. Profit was $18.3 billion, down 14 percent from a year ago.
Mark Zuckerberg, Meta’s chief executive, focused on how A.I. was aiding his company’s businesses, including its digital advertising.
“We are now at a point where our investments in A.I. are accelerating every major part of our core business,” he said. He added that “potentially selling compute directly,” which refers to computing power, could be a new business opportunity.
But investors appeared to be spooked by Meta’s rising expenses. The company’s stock fell more than 9 percent in after-hours trading.
Meta’s continued spending follows Google’s announcement last week that it would raise its capital expenditures. Big tech companies are expected to pay $1.5 trillion on building data centers this year and next year. The enormous sums have raised questions about whether such spending is justified, with the stock market gyrating over concerns about the outlays in recent weeks.
Mr. Zuckerberg has previously said that Meta was considering selling computing power from its data centers to other companies. In June, Anthropic offered to buy computing power from Meta in a deal that could be worth up to $10 billion; the talks are ongoing.
Unlike Google, Amazon and Microsoft, which are also spending on A.I. data centers, Meta is without a business that lets companies rent computing power and A.I. tools.
People think A.I. spending “is supposed to slow down, but who’s going to be the first company to blink?” said David Wagner, the head of equity at Aptus Capital Advisors.
Meta’s shift from a social media company to an A.I. firm has not been smooth. This month, it removed a feature that let people make A.I. images of each other on Instagram just days after its release because of a backlash over privacy concerns, among other snafus.
But the company has made progress developing its own A.I. models. This month, Meta released the latest version of Muse Spark, its most advanced A.I. model developed under Alexandr Wang, the company’s chief A.I. officer. It also introduced an A.I. image generator called Muse Image and plans to release a video generator in the coming months.
Muse Spark still trails other A.I. models in benchmarks that measure coding, reasoning and writing. Meta plans to release a more powerful model, code-named Watermelon, this fall.
In recent days, Mr. Zuckerberg has also gone on the attack against companies like Anthropic and OpenAI in defense of open-source A.I. models, which are freely available for others to build upon. In an interview with The New York Times on Tuesday, he said that tightly controlling A.I. development would be “abandoning our values” and would stifle innovation.
Meta also faces ongoing litigation over claims that its social media products are addictive and said it spent $2.4 billion on legal fees in the second quarter. It lost the first of nine bellwether addiction trials in March but gained a reprieve this month after the plaintiff in one lawsuit, a 15-year-old teenager from Florida, dropped his suit.
Meta’s A.I. smart glasses continue to be a bright spot. Its Reality Labs division, which develops those glasses, generated $431 million in revenue in the quarter, up 16 percent from a year earlier. The division lost $4.6 billion, about the same as a year earlier.
Meta’s family of apps grew to 3.6 billion users, up 3 percent from a year earlier.
