In a rare instance of AI chips, Meta symbols work with millions of Amazon AI CPUs

Amazon just got a big prize with Meta thanks, again, to Amazon’s home chips. Meta has signed an agreement to use millions of AWS Graviton chips to power its growing AI needs, Amazon announced Friday.

Note that AWS Graviton is an ARM-based CPU, (central processing unit, a chip that handles general computing tasks) and not a GPU (graphics processing unit).

While GPUs remain the chip of choice for training large models, once those models are trained, the AI ​​agents built on top of them change the type of chip needed. Agents create complex computing workloads such as real-time reasoning, writing code, searching, and communication involved in managing agents with multi-step tasks. AWS’ latest version of Graviton is specifically designed to handle AI-related computing needs, the company said.

The deal returns most of Meta’s revenue to AWS instead of competitors like Google Cloud. Last August, Meta signed a six-year, $10 billion deal with Google Cloud, although Meta was, until then, an AWS customer that used Microsoft Azure.

We couldn’t help but notice that AWS had timed the announcement of this deal just as the Google Cloud Next conference was wrapping up, as a visible smile to its cloud rival. Google, of course, is customizing its AI chips and announcing new versions of them at the show.

It’s true, Amazon makes its own AI GPU as well: Trainium, which, despite its name, is used for both training and expression – the phase that occurs after the training of the model, when it processes commands.

But Anthropic had already stepped in with a deal announced earlier this month that licenses most of those chips for years to come. The manufacturer of Claude agreed to spend $ 100 billion over 10 years to run his work on AWS – focusing mainly on Trainium – while Amazon agreed to invest another five billion dollars (bringing its value to $ 13 billion of investment) in Anthropic in return.

Ultimately, the Meta deal allows Amazon to showcase its biggest AI customer as a benchmark for home CPUs. These are chips that compete with Nvidia’s new Vera CPU, which is also based on ARM and designed to handle AI workloads. The difference, of course, is that Nvidia sells its AI chips and systems to businesses and cloud providers (including AWS). AWS only sells access to its chips through its cloud service.

Earlier this month Amazon CEO Andy Jassy took aim at Nvidia and Intel in his annual letter to shareholders, saying that businesses are looking for better AI price performance metrics, and that he intends to win deals that way. This also means that the pressure wouldn’t be as high on Amazon’s internal chip design team to deliver, a team we visited last month on a special tour of their lab.