Metra on Thursday announced plans to cut about 8,000 employees, about 10% of its workforce. As of Friday, it had signed a multi-year chip deal with Amazon. The contrast reflects exactly where CEO Mark Zuckerberg has decided to place his bets: fewer people, far more computing power.
Amazon Web Services confirmed Friday that Meta has agreed to use its Graviton processors in an arrangement lasting at least three years. Financial terms were not disclosed, but the scale is significant given that, according to Nafia Bashara, AWS vice president and distinguished engineer, Meta will deploy hundreds of thousands of Graviton chips, making it one of AWS’s top five Graviton customers globally.
Graviton chips are general-purpose Arm-based processors that are closer to Intel or AMD CPUs than Nvidia’s GPUs. They are not designed for the massively parallel processing that training large AI models demands, but they play a different and increasingly important role: the CPU-intensive workloads that power agentic AI, where models take sequences of actions, make decisions, and interact with other software in real time.
Meta has rented Nvidia GPUs from AWS since 2017 and hasn’t moved away from them. The Graviton deal is an extension, not a replacement, a sign that the infrastructure requirements to run AI at meta scale now exceed those that can be efficiently covered by any single chip architecture.
In recent weeks alone, Meta has contributed a combined $48 billion to CoreWave and Nebius, both GPU rental platforms built on Nvidia hardware. Graviton fills a different gap. AWS says the chips provide the best price-performance ratio in its EC2 computing service, while consuming 60% less energy than comparable alternatives, which is a meaningful figure when operating 32 data centers serving 3.6 billion daily users.
Intel CEO Lip-Bu Tan told analysts this week that demand for its Xeon server chips is exceeding supply, and “CPUs are re-establishing themselves as the indispensable foundation of the AI age.”
