Meta is all in on AI as new chip & supercomputer are in development
Firefly AIMeta, formerly Facebook, has pivoted hard into AI from the metaverse. The company now plans to develop its own chips and supercomputer for AI research.
AI is the current new hotness down in Silicon Valley and across the world. The latest to join the fray is Meta, which has begun to develop its own chip to power supercomputers and other AI research.
Writing in a blog on Meta’s press website, Santosh Janardhan, VP and Head of Infrastructure, details that the firm will be developing its “first generation custom silicon chip for running AI models”. This will also be backed by a “new AI-optimized data center”, as well as the continued development of its 16,000 GPU supercomputer.
Meta’s chip is expected to be completed in 2025 and is intended to train and run AI models.
The company has split its plans for expanding the infrastructure of AI into three components. These are:
- MTIA (Meta Training and Inference Accelerator)
- Next-gen data center
- Research SuperCluster (RSC) AI Supercomputer
The plans involve deploying the new chips, which will help Meta provide “better performance, decreased latency, and greater efficiency”. On top of this, the AI-optimized data center will feature liquid cooling amongst its server racks. Its supercomputer will be powered by 16,000 GPUs to allow it to access “2000 training systems”.
Meta pivots to AI after metaverse blunder
Meta’s pivot to AI is not unexpected, as the company has routinely lost money since it tried to make the metaverse a thing. The exponential losses caused a brief price increase for its aging Quest 2 VR headset, as well as the ensuing price drop on the Meta Quest Pro, which failed to live up to its potential.
Janardhan mentions that they expect this type of development to happen over “the next decade”, as more companies – including Meta – seek out more “customization in chip design”.