Alexander - stock.adobe.com

Meta ramps up GPUs to get ready for general intelligence

Meta CEO Mark Zuckerberg anticipates that training and running AI systems requires 10x computer capacity each year

Meta, the parent company of Facebook, WhatsApp and Instagram, has posted full year revenue of  $134.90bn, 16% more than in 2023.

The Q4 2023 and full-year earnings show the company’s commitments to investing in artificial intelligence (AI) and datacentre capacity. It expects to use 10 times as much compute capacity each year to build out its AI strategy, following the under capacity it initially faced when building the AI in its Reels video sharing platform.

In a transcript of the earnings call, Zuckerberg said: “We initially under-built our GPU clusters for Reels, and when we were going through that I decided that we should build enough capacity to support both Reels and another Reels-sized AI service that we expected to emerge so we wouldn’t be in that situation again. At the time, the decision was somewhat controversial and we faced a lot of questions about capex [capital expenditure] spending, but I’m really glad that we did this.”

In terms of training AI models and running AI inference engines, he anticipated that going forward these would require even more intensive compute power, adding: “We don’t have a clear expectation for exactly how much this will be yet, but the trend has been that state-of-the-art large language models [LLMs] have been trained on roughly 10x the amount of compute each year.”

Zuckerberg described 2023 as “our year of efficiency”, which, he said, focused on making Meta a stronger technology company and improving the business., which are needed to give the company the the stability to deliver its long-term vision for AI and the Metaverse.

Looking ahead at 2024, he said: “A major goal will be building the most popular and most advanced AI products and services. If we succeed, everyone who uses our services will have a world-class AI assistant to help get things done, every creator will have an AI that their community can engage with, every business will have an AI that their customers can interact with to buy goods and get support, and every developer will have a state-of-the-art open source model to build with.”

Zuckerberg believes that the next generation of services will require building full general intelligence. “Previously I thought that because many of the tools were social, commerce, or maybe media-oriented that it might be possible to deliver these products by solving only a subset of AI’s challenges. But now it’s clear that we’re going to need our models to be able to reason, plan, code, remember, and many other cognitive abilities to provide the best versions of the services that we envision.”

He said Meta would be working on general intelligence in its products and services going forward.

These ambitions are likely to have a material impact on how much the company spends on building out compute capacity in its datacentres. When asked for guidance on capital expenditure, CFO Susan Li, said: “This continues to be a pretty dynamic planning process for us. There are also certainly other factors that drive uncertainty: how quickly can we execute on the new datacentre architecture, how the supply chain turns out to unfold over the course of the year.

“But our expectation is, generally, that we will need to invest more to support our AI work in the years ahead, and we’re seeing some of that reflected in 2024.”

Zuckerberg also used the earnings call to discuss the benefits of using open source in the development of AI, adding: “Open sourcing improves our models, and because there’s still significant work to turn our models into products and there will be other open source models available anyway, we find that there are mostly advantages to being the open source leader and it doesn’t remove differentiation from our products much.”

Read more about AI investments

  • Micosoft used its annual Ignite conference to showcase the work it is doing to optimise AI and make more energy-efficient hardware.
  • Four common AI chips -- CPU, GPU, FPGA and ASIC -- are advancing with the current market for AI chip design. Read on to see what the future holds for AI chips.

Read more on IT architecture

CIO
Security
Networking
Data Center
Data Management
Close