Meta Adds Tens of Millions of AWS Graviton Cores as Agentic AI Becomes a Major CPU Story
Meta is expanding its AI infrastructure through a major new partnership with Amazon Web Services, bringing tens of millions of AWS Graviton CPU cores into its compute portfolio as agentic AI workloads begin reshaping the balance between CPUs and GPUs.
The partnership, announced through Meta, focuses on AWS Graviton processors, which Meta will use to support the growing compute requirements of agentic AI. While modern AI infrastructure has been heavily associated with GPUs, especially for large scale model training and inference, Meta’s move highlights a clear industry shift: agentic AI requires far more CPU resources than traditional AI deployment models.
According to Meta, the company is partnering with AWS to bring tens of millions of Graviton cores into its infrastructure. The latest AWS Graviton5 processor features 192 Arm Neoverse cores, which means Meta’s deployment represents a massive CPU investment at global scale. The announcement also reinforces Meta’s broader infrastructure strategy, where no single chip architecture is expected to efficiently serve every workload.
Agentic AI is changing how compute is allocated. Unlike traditional AI workloads that often rely heavily on GPU clusters for training or model inference, agentic systems require orchestration, planning, tool use, memory handling, task routing, data processing, and continuous interaction across multiple services. These layers depend heavily on CPUs, especially when AI agents need to manage workflows, coordinate actions, and operate across large scale user facing systems.
This is why Meta’s partnership with AWS is strategically important. The company is not simply adding more chips. It is expanding its infrastructure foundation to support a future where AI systems are more interactive, autonomous, and distributed. As Meta continues building AI products that must scale to billions of users worldwide, CPU efficiency becomes just as important as raw accelerator performance.
“This isn’t just about chips; it’s about giving customers the infrastructure foundation, as well as data and inference services, to build AI that understands, anticipates, and scales efficiently to billions of people worldwide. Meta’s expanded partnership, deploying tens of millions of Graviton cores, shows what happens when you combine purpose built silicon with the full AWS AI stack to power the next generation of agentic AI.
- Nafea Bshara”
With this agreement, Meta is set to become one of Amazon’s largest Graviton customers globally. That is a major win for AWS and a strong signal that Arm based CPUs are becoming increasingly important in AI infrastructure. Graviton processors have already been positioned as efficient cloud scale CPUs, and Meta’s adoption gives AWS a major validation point in the race to support next generation AI workloads.
The move also reflects a larger market trend. As AI companies build increasingly complex systems, CPU demand is rising alongside GPU demand. Intel, AMD, NVIDIA, Amazon, and other CPU makers are all becoming more relevant to the AI expansion cycle. GPUs remain essential for many AI workloads, but CPUs are becoming critical for the surrounding infrastructure that makes agentic systems work at scale.
Meta is also continuing its own custom silicon strategy. The company has reportedly partnered with Broadcom to develop custom AI chips for a multi gigawatt AI ecosystem, while also advancing its internal MTIA accelerator family. However, with advanced semiconductor capacity constrained across TSMC, Samsung, and other manufacturing partners, companies are being forced to diversify their compute strategies. That means combining custom silicon, cloud infrastructure, CPUs, GPUs, and specialized accelerators across a broader portfolio.
Amazon CEO Andy Jassy also highlighted the partnership through X, further emphasizing the scale and importance of AWS Graviton in the next phase of AI infrastructure. For AWS, the Meta deal is not only about selling CPU capacity. It strengthens Amazon’s position as a major AI infrastructure provider at a time when hyperscalers are competing aggressively to host, power, and optimize next generation AI workloads.
Another exciting development in our chips business as Meta has decided to bet big on Graviton, our leading CPU chip—committing to tens of millions of Graviton cores.
— Andy Jassy (@ajassy) April 24, 2026
Agentic AI is becoming almost as big a CPU story as a GPU story. Complex multi-step orchestration, real-time… pic.twitter.com/nccGQ6Gw0n
The phrase “tens of millions of cores” is also important because this appears to be only the first deployment. Meta plans to keep expanding its AI resources in the coming years, and as agentic AI systems grow more complex, the company will likely continue adding more compute capacity across multiple chip types. This could include more Graviton CPUs, more GPUs, custom accelerators, and other purpose built silicon depending on workload requirements.
The broader lesson is clear: AI is no longer only a GPU story. Training large models still demands enormous accelerator clusters, but the next phase of AI will require massive CPU scale as well. Agentic AI needs coordination, responsiveness, scheduling, context management, workflow execution, and integration across services. Those functions make CPUs a central part of AI infrastructure, not just a supporting component.
For Meta, the AWS Graviton partnership gives the company a more flexible infrastructure base as it pushes deeper into AI assistants, automation, social AI features, business tools, and future agentic systems. For Amazon, it proves that Graviton is becoming a serious force in large scale AI deployments. For the wider industry, it is another sign that the AI hardware race is expanding beyond GPUs into a full stack compute battle.
As AI workloads continue evolving, the winners may not be the companies with only the fastest accelerators, but the ones that can combine CPUs, GPUs, custom silicon, networking, storage, power efficiency, and cloud services into a scalable compute portfolio. Meta’s Graviton deployment is a major step in that direction.
Will agentic AI make CPUs just as strategically important as GPUs in the next generation of AI infrastructure?
