Meta has agreed to use millions of AWS Graviton chips to power its AI workloads, Amazon announced in April 2026, marking a significant win for Amazon’s homegrown chip program and pulling Meta’s spending away from rival cloud providers.
The AWS Graviton is an ARM-based CPU — a central processing unit that handles general computing tasks — rather than a GPU. While GPUs remain the standard for training large AI models, the rise of AI agents is shifting demand toward CPUs capable of handling workloads like real-time reasoning, code generation, search, and the coordination of multi-step tasks. Amazon says its latest Graviton version was designed specifically to address these AI-related compute needs.
The deal represents a partial reversal of Meta’s recent cloud spending trajectory. Last August, Meta signed a six-year, $10 billion deal with Google Cloud, though Meta had previously been primarily an AWS customer that also used Microsoft Azure. Amazon timed its announcement of the Meta deal to coincide with the close of the Google Cloud Next conference.
Amazon’s own AI GPU, the Trainium — used for both model training and inference — was not the focus of this agreement. Earlier in April 2026, Anthropic committed to spending $100 billion over 10 years on AWS workloads with a particular emphasis on Trainium, while Amazon agreed to invest an additional $5 billion in Anthropic, bringing its total investment to $13 billion.
The Graviton chips compete with Nvidia’s Vera CPU, which is also ARM-based and targets AI agentic workloads. Unlike Nvidia, which sells chips and AI systems directly to enterprises and cloud providers, AWS only offers access to its chips through its cloud service. Amazon CEO Andy Jassy, in his annual shareholder letter earlier in April 2026, stated that enterprises are seeking better price-performance ratios for AI and indicated that Amazon intends to compete on that basis.
Source: TechCrunch