Meta Platforms' third-quarter earnings report this week was a good-news-bad-news affair. The stock was up over 7% in late trading following the Wednesday release, but down as much as 8% the following day. The dramatic swing speaks to both the potential in Meta's artificial-intelligence strategy, and the costs.
The first bit of good news was that Meta beat expectations on both top and bottom lines (after excluding a one-time charge for the company's multistate settlement in child safety litigation). It also provided solid revenue guidance for the fourth quarter, signaling that AI-boosted growth rates in the mid-20s would continue through the end of the year.
Even better news is that after a lot of internal turmoil and spending, Meta has taken the important first step of leapfrogging into a leading position in AI technology, especially as it relates to consumers. Its Muse family of AI models are now on par with the best from OpenAI and Anthropic, long the two leaders in that space.
And though still in an early stage, the company's Muse personal agent has a lot of momentum and could become a breakout product for Meta.
Muse is available only in the U.S. and Canada, but Meta said that more than 10 million people have set it up. For comparison, according to SensorTower, ChatGPT took four months to hit 10 million U.S. downloads, while Muse did it in 40 days.
Meta said that about a quarter of Muse agents get used during a given day, which is solid retention for a new product. At about 2.5 million daily users, that's still less than 0.1% of the 3.6 billion people who use at least one of Meta's social-media or messaging apps every day. That speaks to both the challenge and opportunity ahead.
Though 98% of Meta's revenue comes from ads, at least at first that won't be how it monetizes Muse. Instead, Meta will make money from e-commerce and premium subscriptions. Muse can take a simple command, such as, "Plan a five-day November trip to Maui before the Thanksgiving rush, and use my loyalty points to pay," and execute the whole thing including the purchase.
Meta is looking for a small piece of every transaction the agent facilitates. CEO Mark Zuckerberg said that a quarter of Muse users have used the shopping tools, though he didn't say how many were actually completing purchases. He said the average transaction was over $200.
Some retailers, like the largest e-commerce site Amazon.com, have banned the Muse agent, which may limit this opportunity. And once the product expands widely beyond the U.S., the average purchase will drop quickly. When Meta still broke out revenue by region in 2023, average revenue per user in the U.S. and Canada was $227, three times as big as in Europe, 11 times the size of Asia-Pacific, and 14 times the rest of the world. The 2023 global average was $45, up to $67 in the past 12 months.
Revenue per user is going to have to be high, and it's the rising costs behind this new muscular AI posture that's turning investors off. During the earnings call, the stock began to fall when Chief Financial Officer Susan Li said the not-so-magic words, "Muse's compute intensity means that expanding its user base will require significant investment, and we expect total serving costs to rise." When pressed later in the call to put a number on how much each Muse user cost Meta, Li declined except to reiterate that it was "pretty compute expensive."
Meta may spend up to $145 billion on capital expenditures this year, and with the first three quarters in the books, that comes out to a $49 billion to $59 billion tab in the fourth quarter alone. In the third quarter, all of the company's operating cash flow went to capex, plus another $2 billion. The free-cash-flow situation will probably get worse before it gets better.
In the past year, the company added over $60 billion in debt and lease liabilities to its balance sheet, and at the end of September it also has more than $800 billion in future lease and supply commitments on top of that. As a result of the cash flow crunch, Meta has now gone four quarters in a row without share repurchases. The simultaneous announcement of a new nonvoting class of shares indicates that Meta may add substantial equity sales to the funding mix, diluting shareholders' stakes.
So, as fast as Meta's revenue is expanding right now, expense growth is outpacing it. At the midpoint of Meta's guidance, it projects a 24% jump in fourth-quarter revenue from last year, but expenses will grow by 42%. Operating margin will drop to 33% from 41% last year, and analysts expect it to stay around there next year.
Muse is the good news and the bad news, both the promise of future growth and the enormous cost it will take to realize it. Meta's key lever will be bringing cost per user down.
"Muse is very compute-intensive to run, but we expect to improve its cost efficiency by more than 10 times in the next year, which will enable it to be profitable once the revenue lines are at scale," Zuckerberg said on the earnings call.
But on a follow-up call, Li cautioned that those gains weren't the "lowest hanging fruit" and that "10 times is still going to require more technological advances over the course of the year."
Meta has put in a ton of hard work and capital to get its AI technology to the frontier, but that's only the beginning.
Comments