Meta, the guardian firm of Fb, Instagram and WhatsApp, has reported a 19% enhance in income to $40.6bn for the third quarter of 2024, pushed partly by what founder and CEO Mark Zuckerberg calls “good progress with AI [artificial intelligence]”.

“We had a great quarter pushed by AI progress throughout our apps and enterprise,” he mentioned. “We even have sturdy momentum with Meta AI, Llama adoption and AI-powered glasses.”

Throughout the earnings name for the quarterly outcomes, Chad Heaton, vice-president of finance for Meta Platforms, was requested concerning the return on funding on Meta’s capital expenditure (CapEx), given the corporate reported a 15% enhance in prices and bills and $9.2bn CapEx. He answered the query by discussing the corporate’s datacentre and server enlargement to assist the AI and non-AI companies Meta’s varied companies supply.

“We’re nonetheless working by means of our 2025 infrastructure plan, however at this level we count on whole infrastructure spend [on] non-Gen AI, GenAI and core AI to extend in 2025,” mentioned Heaton.

He went on to clarify that the corporate plans to take a position primarily in servers, however is anticipating spending on datacentres and community gear to develop as Meta builds out its IT infrastructure footprint. Heaton mentioned the corporate would even be investing in servers to assist better consumer engagement in “non-AI capability”, in addition to refreshing present servers.

As well as, he mentioned the corporate can be getting into the core development part of its technique to construct out smaller, higher-density datacentres. “We’re constructing higher-capacity networks to accommodate the expansion in GenAI and core AI-related visitors in 2025, together with investing in fibre to deal with future cross-region coaching visitors.”

Up to now, Meta’s chief monetary officer, Susan Li, has mentioned extending the lifetime of datacentre servers as a method to scale back prices. However when requested if this coverage would proceed, she mentioned: “We don’t at the moment have plans to increase the helpful lives of our servers. That’s often because, given the continued efficiency features from new generations of GPU-based servers, we actually need to make it possible for we’re making one of the best use of the capability accessible in our datacentres through the use of the most recent era of servers.”

Past {hardware}, Li mentioned the corporate has targeted on optimising present suggestion programs to assist drive incremental enhancements and effectivity.

Discussing current optimisations on Instagram, she mentioned: “We accomplished [infrastructure] effectivity tasks that consolidated a few of our present suggestion programs, which lets us focus our future enhancements on a smaller variety of suggestion fashions.”

In accordance with Li, this helps to drive larger engineering effectivity.

Trying particularly on the return on funding, she mentioned: “Among the nearer-term alternatives are actually round relevance features as we scale our fashions for content material suggestion.”

This entails growing mannequin complexity, which permits Meta to be taught extra from the prevailing information utilizing extra compute energy. Meta additionally plans to broaden the quantity of information units utilized in AI, which, as Li famous, permits the machine studying algorithms to be taught from extra information.