Meta prepares for gigawatt datacentres to power ‘superintelligence’
Meta’s latest results show a big jump in datacentre costs as it builds out personal AI capabilities
Meta, the parent company of Facebook, WhatsApp and Instagram, has reported increases in operating cost and research and development as it ramps up artificial intelligence (AI) across its businesses.
The big push is now “superintelligence”, with the company ploughing money into the newly formed Meta Superintelligence Labs to focus on developing the next generation of its AI models.
In an open letter, CEO Mark Zuckerberg discussed how advances in AI could mean that people spend less time using productivity software and more time creating and connecting. The lab will be working on personal superintelligence, which, Zuckerberg said: “will know us deeply, understands our goals, and can help us achieve them”.
Access to this AI, at least from Meta’s perspective, will be via personal devices such as AI glasses. Zuckerberg said such devices are able to “understand context”, adding: “They can see what we see, hear what we hear and interact with us throughout the day.” He predicted such devices would become people’s primary computing device.
In his prepared remarks, Zuckerberg said the company was building out several “multi-GW [gigawatt] clusters”, stating: “Our Prometheus cluster is coming online next year, and we think it’ll be the world’s first 1GW+ cluster. We’re also building out Hyperion, which will be able to scale up to 5GW over several years.”
The investment in datacentres is linked to Meta Superintelligence Labs. “We have conviction that superintelligence is going to improve every aspect of what we do,” Zuckerberg said.
Its latest quarterly earnings show that the cost of revenue increased 16%, which Meta said has been driven mostly by higher infrastructure costs and payments to partners. Chief financial officer Susan Li said this cost was partially offset by the company’s plans to extend the useful life of its servers.
However, Meta has continued to grow its server estate. Its stated capital expenditure was $17bn, driven by investments in servers, datacentres and network infrastructure. Li reported that Meta has also increased research and development by 23%, due to higher employee compensation and infrastructure costs.
In a blog post discussing formation of the superintelligence lab, Forrester vice-president and research director Mike Proulx said: “Meta’s CEO is hopeful that superintelligence will be used to empower people and not be ‘focused on replacing large swaths of society’. But human replacement is already happening, and this is just the beginning. Business leaders see AI as an efficiency play above all else. The fact is that AI can save companies time and money [is] a good thing for shareholder value, but will it be good for society?
“As with every major technology disruption, some good will come from it, but also some bad. How bad the impact of superintelligence gets depends, in part, on the ethics of the companies developing it. Meta says it ‘will need to be rigorous about mitigating these risks and careful about what we choose to open source’, but many companies are vying feverously to win the superintelligence race. At what cost are they willing to do so? Mere trust in companies to do the right thing isn’t going to cut it.”
