Meta is going bigger on AI infrastructure, and this time it is not just about GPUs. The company has struck a deal with Amazon Web Services to deploy Graviton processors at a scale that is honestly hard to wrap your head around. We are talking tens of millions of cores right out of the gate, with room to grow from there. That is not a test run. That is a serious commitment.
The move says a lot about where AI is heading. Training large models still leans heavily on GPUs, sure, but running them in the real world is a different story. Agentic AI, which is the buzzword of the moment, is all about systems that can think through steps, generate code, search, and juggle tasks on the fly. That kind of work tends to hammer CPUs, and Meta seems to be embracing that reality.
That is where Graviton comes in. AWS has been pushing its custom ARM-based chips for years, pitching better performance per watt and lower costs compared to traditional options. The latest version, Graviton5, ups the ante with 192 cores and a much larger cache. That bigger cache matters more than it sounds, since it cuts down on delays between cores. When you are coordinating massive AI workloads across thousands or millions of threads, those small improvements add up quickly.
Meta plans to use these chips across a range of workloads tied to its AI efforts. Think billions of interactions, all happening in real time, with systems that need to reason through multi-step processes without falling over. That is not light work, and it is exactly the kind of thing CPUs can quietly excel at when designed for it.
There is also the underlying AWS infrastructure doing a lot of heavy lifting here. The Nitro system gives Meta near bare-metal access while still keeping the flexibility of the cloud. Then you have things like Elastic Fabric Adapter, which helps different compute instances talk to each other with low latency. Again, not flashy, but critical when everything needs to stay in sync.
None of this comes out of nowhere. Meta has been working with AWS for years and already leans on its services for AI. What is new is the scale and the strategy. Instead of going all in on one type of hardware, Meta is spreading the load. GPUs for training, CPUs for everything else that keeps these systems actually running.
There is also a practical angle here. AI is getting expensive to run, and energy use is becoming a real concern. Graviton5 is built on a 3-nanometer process, which should help with efficiency. AWS claims a decent performance bump over the previous generation too, which could translate into lower costs over time. When you are operating at Meta’s scale, even small gains matter.
Step back for a second and this feels like a shift. Not a dramatic overnight change, but a clear adjustment in how big tech is thinking about AI infrastructure. It is not just about throwing more GPUs at the problem anymore. It is about using the right hardware for the right job.
Meta betting big on Graviton is a pretty strong signal that CPUs still have a major role to play in the AI future.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz