Meta, the parent company of Facebook, Instagram and WhatsApp, has reported a 19% increase in revenue to $40.6bn for the third quarter of 2024, driven in part by what founder and CEO Mark Zuckerberg calls “good progress with AI [artificial intelligence]”.
“We had a good quarter driven by AI progress across our apps and business,” he said. “We also have strong momentum with Meta AI, Llama adoption and AI-powered glasses.”
During the earnings call for the quarterly results, Chad Heaton, vice-president of finance for Meta Platforms, was asked about the return on investment on Meta’s capital expenditure (CapEx), given the company reported a 15% increase in costs and expenses and $9.2bn CapEx. He answered the question by discussing the company’s datacentre and server expansion to support the AI and non-AI services Meta’s various businesses offer.
“We’re still working through our 2025 infrastructure plan, but at this point we expect total infrastructure spend [on] non-Gen AI, GenAI and core AI to increase in 2025,” said Heaton.
He went on to explain that the company plans to invest mainly in servers, but is expecting spending on datacentres and network equipment to grow as Meta builds out its IT infrastructure footprint. Heaton said the company would also be investing in servers to support greater user engagement in “non-AI capacity”, as well as refreshing existing servers.
In addition, he said the company would be entering the core construction phase of its strategy to build out smaller, higher-density datacentres. “We’re building higher-capacity networks to accommodate the growth in GenAI and core AI-related traffic in 2025, along with investing in fibre to handle future cross-region training traffic.”
In the past, Meta’s chief financial officer, Susan Li, has discussed extending the life of datacentre servers as a way to reduce costs. But when asked if this policy would continue, she said: “We don’t currently have plans to extend the useful lives of our servers. That’s really because, given the ongoing performance gains from new generations of GPU-based servers, we really want to make sure that we are making the best use of the capacity available in our datacentres by using the latest generation of servers.”
Beyond hardware, Li said the company has focused on optimising existing recommendation systems to help drive incremental improvements and efficiency.
Discussing recent optimisations on Instagram, she said: “We completed [infrastructure] efficiency projects that consolidated some of our existing recommendation systems, which lets us concentrate our future improvements on a smaller number of recommendation models.”
According to Li, this helps to drive higher engineering efficiency.
Looking specifically at the return on investment, she said: “Some of the nearer-term opportunities are really around relevance gains as we scale our models for content recommendation.”
This involves increasing model complexity, which enables Meta to learn more from the existing data using more compute power. Meta also plans to expand the volume of data sets used in AI, which, as Li noted, enables the machine learning algorithms to learn from more data.