Meta’s AI splurge lays bare its compute conundrum

Meta can simultaneously fuel its own AI ambitions and rent out its scarce computing capacity to bolster returns, CEO Mark Zuckerberg signaled on Wednesday. The problem is: investors aren’t buying it.

The Instagram owner is splurging billions to build compute — chips, servers, energy and data centers that power AI — leaving it with free cash flow of just $784 million in the second quarter to run and grow its business. That collapse, of 91% from a year ago, drove its stock down more than 9% on Thursday.

Pressed by analysts for details on Meta’s plans, Zuckerberg framed compute as a scarce strategic asset that the company should keep and build around, rather than simply sell for short-term profit.

But he acknowledged that the company had received a number of offers for its computing capacity from businesses that wanted to deploy their own AI plans “at a meaningful premium” over what it invested to build that capacity.

That tension sits at the heart of Meta’s challenge in diversifying its revenue stream. Renting out compute could ease Meta’s cash-flow squeeze, but would also divert scarce resources from its own push to build AI models and services.

Having built its fortune selling ads on Facebook and Instagram, the company is now trying to take on larger rivals including Microsoft, Alphabet, and Amazon that have deep ties to enterprises, an early and lucrative market for AI.

Microsoft on Wednesday showed how its AI bets were paying off even as its free cash flow fell 23%. The Windows maker breezed past expectations for growth in its Azure cloud unit and Copilot assistant thanks to a huge base of corporate customers and its early AI build-out, sending its stock up over 13%.

Spending like a cloud giant

“We believe that there will continue to be a significantly higher margin on selling intelligence rather than selling compute directly, but we think that there’s a big opportunity obviously to sell compute as well,” he said, as he painted a picture of what Meta hopes to build with its AI spending spree.

Zuckerberg argued that AI-powered personal assistants could become a mass-market product used by billions of consumers, while business agents could eventually help companies handle customer service, sales and marketing.

But beyond broad references to subscriptions and enterprise services, he offered few specifics on how those businesses would justify Meta’s massive AI spending.

Responding to a question from J.P.Morgan analyst Doug Anmuth, who noted that Meta was also purchasing capacity from third parties while selling compute, Zuckerberg said Meta was intentionally investing ahead of demand.

“There is a lead time where we’re investing in building out these data centers now. They come online at some point in the future. You obviously are not getting value out of them until they’re online,” he said.

“Meta is spending like a hyperscaler without a hyperscaler’s business model,” said Josh Gilbert, online investing platform eToro’s lead APAC analyst, referring to large cloud companies.

“Microsoft, Alphabet and Amazon can point their data center dollars at cloud businesses that sell compute straight back out the door, but Meta doesn’t have the same outlet, so every dollar of build-out leans on the ads business.”

Metaverse scare

Meta’s massive AI spending push has unnerved some investors, who are drawing parallels to the company’s costly pivot to the metaverse — a bet that racked up tens of billions of dollars in losses without becoming a large revenue generator.

Its free cash flow slump in the April-June quarter was the steepest since late 2022, when the metaverse bet attracted similar investor scrutiny.

Still, Meta raised the low end of its capital spending forecast by $5 billion, lifting the range to between $130 billion and $145 billion.

That echoes Alphabet, which raised its own forecast by $15 billion last week, reported its first cash burn on record and was punished with a 7% share drop.

Meta CFO Susan Li said on a post-earnings call that the spending was justified because the industry had “underbuilt historically” for AI demand, making existing capacity “extremely valuable.”

She said capacity would remain tight “for the foreseeable future,” creating opportunities for Meta to generate returns through products, enterprise services and compute sales.

But responding to a question from Wells Fargo analyst Ken Gawrelski on whether Meta planned to internally use all its compute capacity through 2027, Li ended the call much as Zuckerberg had started it: arguing Meta could find profitable uses for additional compute both inside and outside the company, without indicating where those returns would ultimately be the greatest.

“The earnings call felt a lot like a good old-fashioned brainstorming session,” said Bernstein analyst Mark Shmulik. – Rappler.com

Meta’s AI splurge lays bare its compute conundrum

Related News & Content

Amazon to boost spending on AI by $20B