Meta’s AI investments in Q2 made a significant difference to its advertising business — it’s just those models aren’t the ones that have everyone, including The Social Network, pouring tens or even hundreds of billions into datacenters every year.
The Zuckercorp’s massive spending on infrastructure and AI visionaries will take a few years to pay off, CFO Susan Li said on Wednesday’s earnings call. She said
“We don’t expect that our genAI work is going to be a meaningful driver of revenue this year or next year,” . For now, improvements to the more conventional machine-learning models that power Meta’s recommender systems pay the bills.
The company’s new AI powered recommendation models were widely deployed, and this led to a five percent increase in Instagram ad conversions and a three-percent gain on Facebook. This was the claim made by CEO Mark Zuckerberg in his prepared remarks.
These models, as their name suggests are designed to boost user engagement across the company’s various platforms by connecting them to friends, posts, or other content. Zuck claimed that improvements to these systems resulted in a five- to six-percent increase in the time spent scrolling through Facebook and Instagram in the last quarter.
These recommender models are the way Meta makes money. They sort through tens of thousands or hundreds of thousand of ads to find those that are most relevant to each user. Meta has created customized accelerators for this process to ensure it runs as quickly as possible.
Li and Zuckerberg claim that genAI will play a larger role in the advertising business. Facebook’s parent company already uses genAI, for example, to help users create marketing collateral. Li said.
Li revealed that Meta had begun incorporating large-language models (LLMs), into the recommender system used by Meta’s X competitor Threads. She said “LLMs are now driving a meaningful share of the ranking-related time spent gains on Threads,” . Meta is also experimenting with LLMs, genAI tools and other areas. Zuck teased earlier a AI engineer that would speed up the development of his Llama herd.
Zuck stated on Wednesday’s call that teams have begun using Llama 4 in order to build autonomous AI agents. “improve the Facebook algorithm to increase quality and engagement.”
Zuck spoke of the open weights model, saying work is ongoing. All of this “happening in low volume right now, so, I’m not sure that result, by itself, was a major contributor to this quarter’s earnings… but I think the trajectory on this stuff is very optimistic,” Zuckerberg said.
- Zuck justifies AI spending with talk of “superintelligence” for all
- Europe is cracking down on AI this week, and Big Tech is not happy
- A new Google AI model maps the world in 10-meter-squares for machines to understand
- Altman embraces his inner Viking and raids Europe with a 100K GPU supercluster located in Norway
Meta is focusing its efforts to build the infrastructure needed to
The firm is currently building out a series AI compute clusters. One of these is called Hyperion, which Zuck claims will be the size and capacity of Manhattan. Meta is also spending lavishly to build its AI superintelligence team to build new systems which not only match human intelligence, but surpass it.
It’s a good job that Meta’s old-school ML is still paying for the bills. Li revealed that compensation packages tied to strategic hires will account for the second largest growth in expenses in 2026. Li said.
Zuck’s AI expenditure and superintelligence pipedream aren’t likely to put Meta into the poor house for the time being. In Q2, Meta’s profit jumped by 36 percent on revenues of $47.5billion. (r)
