In another wild turn for AI chips, Meta signs deal for millions of Amazon AI CPUs

```json { "title": "Meta Signs Multi-Billion AWS Graviton5 CPU Deal for Agentic AI", "metaDescription": "Meta and AWS announce a multi-year, multi-billion-dollar deal deploying tens of millions of Graviton5 CPU cores to power agentic AI workloads at scale.", "content": "<h2>Meta and AWS Strike Multi-Billion-Dollar Graviton5 CPU Deal to Power Agentic AI</h2><p>In one of the most significant chip agreements of 2026, Meta Platforms and Amazon Web Services announced on April 24 a multi-year, multi-billion-dollar deal that will see Meta deploy tens of millions of AWS Graviton5 CPU cores across its infrastructure. The partnership marks a notable pivot in how hyperscalers are thinking about AI compute — shifting meaningful workloads away from GPUs and toward purpose-built CPUs engineered for the demands of agentic AI.</p><p>The deal, confirmed by Nafea Bshara, vice president and distinguished engineer at Amazon Web Services, underscores a broader industry reckoning: as AI systems evolve from static models into dynamic agents capable of real-time reasoning, code generation, search, and orchestrating multi-step tasks, the chip requirements are changing fast.</p><h2>What the Deal Actually Involves</h2><p>Under the agreement, Meta will begin deploying tens of millions of Graviton5 cores, with the flexibility to expand that footprint as its AI capabilities grow, according to the official Amazon press release. The Graviton5 chip contains 192 cores per chip — each assignable to different tasks — and features a cache five times larger than the previous generation. Amazon says those architectural improvements reduce inter-core communication delays by up to 33%, a meaningful gain for the kind of latency-sensitive workloads agentic AI demands.</p><p>AWS has been developing its proprietary Graviton CPU line since 2018 and is now on its fifth generation. The chip is manufactured by Taiwan Semiconductor Manufacturing Co., which continues to sit at the center of the global AI hardware supply chain.</p><p>The partnership also builds on an existing relationship between the two companies, including Meta's use of Amazon Bedrock at scale to support its next generation of AI applications, according to the Amazon press release.</p><p>Santosh Janardhan, head of infrastructure at Meta, explained the strategic rationale directly: <em>"As we scale the infrastructure behind Meta's AI ambitions, diversifying our compute sources is a strategic imperative. AWS has been a trusted cloud partner for years, and expanding to Graviton allows us to run the CPU-intensive workloads behind agentic AI with the performance and efficiency we need at our scale."</em></p><h2>Why CPUs — and Why Now</h2><p>For years, the AI infrastructure conversation has been dominated almost entirely by GPUs, with Nvidia sitting at the center of that narrative. But the rise of agentic AI is reshaping the compute landscape in ways that extend well beyond GPU clusters.</p><p>According to Amazon, agentic AI workloads — which involve real-time reasoning, continuous search, code generation, and coordinating chains of AI-driven tasks — are inherently CPU-intensive. Graviton5 was purpose-built for exactly these workloads, and the Meta deal is, in part, a validation of that design thesis.</p><p>The broader CPU market is also experiencing what Reuters has described as an AI-driven renaissance. Intel has publicly noted that CPU prices are rising as demand from AI inference workloads soars — a signal that the market is repricing compute assets that were, until recently, considered commodity infrastructure.</p><p>Nafea Bshara framed AWS's position in the market in terms of cost efficiency: <em>"We pass that savings on to the customers."</em> That value proposition — performance gains combined with cost advantages over Nvidia GPUs — appears to be a key factor driving hyperscalers toward homegrown silicon.</p><p>The demand pressure is striking. Amazon CEO Andy Jassy revealed in his 2025 annual shareholder letter that two large customers had attempted to purchase all of AWS's available Graviton capacity for 2026, a request AWS declined. Jassy also signaled the commercial potential of AWS's chip ambitions more broadly: <em>"There's so much demand for our chips that it's quite possible we'll sell racks of them to third parties in the future."</em></p><p>Matt Kimball, VP and principal analyst at Moor Insights & Strategy, offered a pointed read on what that demand signal means: <em>"Two large customers asking to buy all of AWS's Graviton capacity for 2026 says everything we need to know about where the market is."</em></p><h2>Meta's Expanding Chip Portfolio</h2><p>The AWS Graviton5 agreement adds another major supplier to what has become an unusually diverse chip roster for Meta. The company has previously signed large deals with Nvidia and Advanced Micro Devices, and has also worked closely with Arm Holdings on Arm's new CPU architecture.</p><p>In February 2026, Meta struck an up-to-$100 billion multi-year chip deal with AMD, tied to a 160 million-share AMD warrant, according to TechCrunch. That deal was framed around Meta's pursuit of what it described as "personal superintelligence." The Graviton5 CPU agreement now sits alongside that AMD relationship, covering a different category of workload — CPU-intensive agentic inference rather than large-scale model training or GPU-accelerated inference.</p><p>The scale of Meta's infrastructure investment makes the diversification strategy easier to understand. The company has committed to up to $135 billion in capital expenditures for 2026 alone, and has pledged to invest at least $600 billion in U.S. data centers and AI infrastructure over the next several years, according to TechCrunch reporting from February 2026. At that level of spend, reliance on any single chip supplier represents both a supply-chain risk and a missed opportunity to optimize costs across different workload types.</p><p>According to Axios, the move also fits into a wider industry pattern: leading cloud providers are actively pushing adoption of their homegrown chips as an Nvidia GPU shortage continues to constrain capacity across the industry.</p><h2>AWS's Homegrown Silicon Moment</h2><p>The Meta deal is the most prominent public validation yet of AWS's years-long bet on proprietary silicon. Beyond Graviton, AWS's Trainium AI accelerator chips are also seeing extraordinary demand. As of March 2026, there are 1.4 million Trainium chips deployed across all three generations, with Anthropic's Claude running on over one million Trainium2 chips alone, according to TechCrunch.</p><p>AWS's Bshara described the company's ambitions as extending well beyond chip supply: <em>"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."</em></p><p>That framing positions AWS not merely as a chip vendor but as a full-stack AI infrastructure partner — a distinction that matters as hyperscalers compete for the largest and most strategically important AI deployments.</p><h2>What Comes Next</h2><p>The Meta-AWS Graviton5 deal is multi-year and explicitly structured with room to expand, suggesting both parties expect agentic AI workloads to grow substantially. The flexibility built into the agreement — starting with tens of millions of Graviton5 cores with no stated ceiling on expansion — reflects the uncertainty inherent in forecasting AI infrastructure needs at this pace of development.</p><p>For the broader chip market, the deal adds to mounting evidence that the AI compute story is no longer a single-chip narrative. GPU capacity remains constrained and strategically critical for model training and certain inference tasks, but CPU-based infrastructure is claiming a growing share of the AI workload conversation — particularly as agentic systems require always-on, low-latency compute rather than the burst capacity that GPUs are optimized to deliver.</p><p>Intel's observation that CPU prices are rising as AI demand grows, combined with AWS's reported difficulty in keeping Graviton capacity available to all customers, points to a market where even traditionally commoditized compute is becoming a contested strategic resource.</p><p>The five largest hyperscalers — Amazon, Alphabet, Microsoft, Meta, and Oracle — are collectively on track to spend between $660 billion and $690 billion on capital expenditure in 2026, according to industry estimates cited by humai.blog. How that capital is allocated across GPU, CPU, and custom accelerator infrastructure will shape the AI competitive landscape for years to come.</p><p>For now, Meta's decision to commit billions to AWS Graviton5 CPUs — while simultaneously maintaining GPU relationships with Nvidia and a major CPU deal with AMD — signals that the company is betting on a heterogeneous compute future, one in which different chip architectures handle different layers of an increasingly complex AI stack.</p><p>For more tech news, visit our <a href="/news">news section</a>.</p>", "excerpt": "Meta Platforms and Amazon Web Services have announced a multi-year, multi-billion-dollar deal deploying tens of millions of AWS Graviton5 CPU cores to power Meta's agentic AI workloads. The agreement signals a strategic shift in AI infrastructure, with purpose-built CPUs claiming a growing share of workloads previously dominated by GPUs. The deal adds AWS to Meta's expanding roster of chip suppliers alongside Nvidia, AMD, and Arm Holdings.", "keywords": ["Meta AWS Graviton5 deal", "agentic AI chips", "AWS Graviton5 CPU", "Meta AI infrastructure", "AI chip market 2026"], "slug": "meta-aws-graviton5-cpu-deal-agentic-ai" } ```