Despite beating expectations and generating more than $40 billion in revenue during the third quarter – a 19% YOY increase – Meta got punished on Wednesday after its Q3 earnings report.
The company’s stock dipped by more than 15% in after-hours trading on the news that its spending will ramp up this year and into the next.
Meta’s capex in 2024 will clock in somewhere between $38 billion and $40 billion, roughly $1 billion more than previously anticipated. CFO Susan Li told investors that Meta expects “significant capital expenditure growth” to continue in 2025.
Where’s most of that money going?
You guessed it: AI.
It’s not like AI-powered Ray Bans – and the servers that support them – develop and produce themselves.
“Our AI investments continue to require serious infrastructure,” said CEO Mark Zuckerberg, “and I expect to continue investing significantly there, too.”
Investing in efficiency
Meanwhile, engagement is growing with Meta AI, which is Meta’s consumer-facing answer to ChatGPT. So is usage of its generative AI tools for advertisers.
According to Zuckerberg, more than 1 million advertisers have used Meta’s generative AI tools so far this year, and they have generated more than 15 million ads over the past month alone. Meta estimates that businesses using its image-generation technology see a roughly 7% increase in conversions.
But there’s more to do. Meta is investing in AI to improve its monetization efficiency and overall marketing performance.
For example, Meta is running tests to see when and where is optimal to show ads within a single user session. “This is enabling us to drive revenue and conversion growth without increasing the number of ads,” Li said.
Another part of its plan is to adopt new approaches to modeling, she said.
Meta recently deployed new learning and modeling techniques that allow its ad systems to consider the sequence of actions a person takes before and after seeing an ad.
“Previously, our ad system could only aggregate those actions together without mapping the sequences,” Li said. “This new approach allows our systems to better anticipate how audiences will respond to specific ads.”
Since implementing the new models during the first half of the year, Meta has seen a 2% to 4% increase in conversions for its advertisers within certain segments.
The latest models
Another top priority – and another use case for AI – is Meta’s longer-term investment in initiatives to improve the user experience across its family of apps, including improving content recommendations.
In the past, Meta operated separate ranking and recommendation systems for each of its products, Li said, “because we found that performance did not scale if we expanded the model size and compute power beyond a certain point.”
But, more recently, Meta began to experiment with the scaling laws of large-language models, which dictate that a model’s quality can evolve over time as it’s trained on larger and larger volumes of data.
Last year, Meta developed new ranking model architectures capable of learning more effectively from much larger data sets, Li said. When applied to its Facebook video ranking models, she said, the better recommendations “unlock meaningful gains in watch time.”
Now, Meta is exploring whether the new models can create similar improvements for recommendations on other surfaces. And, eventually, Meta even plans to feed its models by combining data across different surfaces.
“We’ll look to introduce cross-surface data to these models so our systems can learn from what is interesting to someone on one surface of our apps and use it to improve their recommendations on another,” Li said. “This will take time to execute.”
