Meta will adopt hundreds of thousands of AWS Graviton chips in latest AI infrastructure grab

```json { "title": "Meta to Deploy Tens of Millions of AWS Graviton Cores for Agentic AI", "metaDescription": "Meta signs a major AWS Graviton deal, deploying tens of millions of cores for agentic AI — weeks after $48B commitments with CoreWeave and Nebius.", "content": "<h2>Meta Signs AWS Graviton Deal, Targeting Tens of Millions of Cores for Agentic AI Workloads</h2><p>Meta has signed an agreement with Amazon Web Services to deploy AWS Graviton processors at scale, starting with tens of millions of Graviton cores with flexibility to expand as Meta's AI capabilities grow. The deal, announced on April 24, 2026, makes Meta one of the largest Graviton customers globally and significantly deepens a long-standing partnership between the two companies. The agreement also encompasses Meta's use of Amazon Bedrock at scale to support its next generation of AI — and arrives just weeks after Meta committed approximately $48 billion to AI infrastructure deals with neocloud providers CoreWeave and Nebius.</p><h2>What the AWS Graviton Deal Actually Covers</h2><p>At the center of the announcement is Meta's adoption of AWS Graviton processors — Amazon's family of custom Arm-based chips designed for compute efficiency at hyperscale. The initial deployment begins with tens of millions of Graviton cores, with the scope of the agreement structured to expand as Meta's AI infrastructure needs evolve.</p><p>The hardware at the core of this deal is the AWS Graviton5 chip, which features 192 cores and a cache that is five times larger than the previous generation. According to Amazon, Graviton5 reduces inter-core communication delays by up to 33% — a meaningful gain for the kind of CPU-intensive, multi-step reasoning tasks that define agentic AI workloads, including real-time inference, code generation, and multi-agent orchestration.</p><p>Beyond raw compute, the agreement builds on Meta's existing use of Amazon Bedrock, Amazon's managed foundation model service, positioning the partnership as more than a chip procurement deal. The companies describe the arrangement as an integrated infrastructure play spanning silicon, cloud services, and AI tooling.</p><p>Meta's Head of Infrastructure, Santosh Janardhan, framed the deal in operational terms: <em>"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><p>From Amazon's side, Nafea Bshara, Vice President and Distinguished Engineer at Amazon, described the broader ambition: <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><h2>Meta's Broader $48 Billion AI Infrastructure Push</h2><p>The AWS agreement is the latest move in an aggressive infrastructure spending campaign that Meta has been executing throughout early 2026. Just weeks before the AWS announcement, Meta made two landmark commitments with neocloud providers.</p><p>On April 9, 2026, Meta and CoreWeave announced an expanded agreement worth approximately $21 billion through December 2032, building on a prior arrangement valued at $14.2 billion — bringing Meta's total CoreWeave commitment to more than $35 billion. Around the same time, Meta committed up to $27 billion with Nebius Group for AI infrastructure capacity. Together, those two deals alone represent approximately $48 billion in external AI infrastructure commitments.</p><p>CoreWeave CEO Mike Intrator addressed the logic of large AI companies like Meta pursuing both in-house infrastructure and external cloud partnerships simultaneously: <em>"They're going to continue to do it themselves, but they're also going to continue to do it with us. There's just too much risk not to."</em></p><p>The scale of these commitments reflects Meta's overall capital expenditure trajectory. In its most recent earnings report, Meta said it plans to spend between $115 billion and $135 billion in capital expenditures in 2026 — nearly twice the amount it spent on capex in 2025. That figure encompasses both external cloud partnerships and Meta's own infrastructure buildout, which includes a $10 billion data center in Texas announced in March 2026.</p><p>The neocloud providers benefiting from Meta's spending have themselves grown rapidly. CoreWeave's 2025 revenues soared 168% to $5.1 billion. Nebius grew 2025 sales 479% to $530 million. Meta's commitments represent a significant share of both companies' forward revenue visibility.</p><h2>AWS Graviton Demand Is Outpacing Supply Across the Industry</h2><p>Meta's deal lands against a backdrop of intense, industry-wide demand for AWS custom silicon. In his 2026 annual shareholder letter, Amazon CEO Andy Jassy revealed that two large AWS customers had already asked if they could purchase all of AWS's Graviton instance capacity for 2026 — requests that AWS declined. Jassy did not name the customers.</p><p>That anecdote underscores the degree to which purpose-built cloud silicon has become a strategic resource, not merely a cost-optimization tool. Amazon's combined chips business — spanning Graviton, Trainium, and Nitro — now generates over $20 billion in annual revenue run rate, with triple-digit year-over-year growth. AWS AI revenue as a whole reached a run rate of more than $15 billion in Q1 2026.</p><p>To meet that demand, Amazon is investing heavily. The company plans approximately $200 billion in capital expenditures in 2026, with the majority targeting AWS infrastructure including data centers, power, chips, servers, and networking.</p><p>Meta is not alone among major technology companies deepening their AWS chip relationships. Uber has expanded its AWS deal to adopt both Graviton and Trainium chips. Anthropic, OpenAI, and Apple have also expanded their usage of AWS custom chips, according to reporting from The AI Insider.</p><p>Nafea Bshara highlighted the significance of Meta's scale in a separate statement: <em>"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."</em></p><h2>Why This Deal Matters Beyond the Chip Count</h2><p>The Meta-AWS Graviton deal is notable not just for its scale but for what it signals about the shifting architecture of AI infrastructure. For years, the dominant narrative in AI compute centered on GPU clusters — the dense, parallel processing hardware used to train and run large language models. The emphasis on Graviton, a CPU-class processor, reflects a different set of requirements: the orchestration, reasoning, and decision-making loops that characterize agentic AI applications.</p><p>Agentic AI systems — those capable of planning multi-step tasks, calling external tools, and adapting to dynamic inputs — place heavy demands on CPU performance, memory bandwidth, and low-latency inter-core communication. The Graviton5's architectural improvements, including its 33% reduction in inter-core communication delays and its five-times-larger cache, are directly relevant to these workloads.</p><p>For Meta, which operates AI systems at a scale that spans billions of users across Facebook, Instagram, WhatsApp, and its growing suite of AI-powered products, the efficiency gains from purpose-built silicon translate into meaningful cost and performance advantages at the margin.</p><p>The deal also reflects Meta's broader strategy of diversifying its infrastructure supply chain. Rather than concentrating AI capacity with a single provider or relying entirely on proprietary hardware, Meta is distributing its bets across AWS, CoreWeave, Nebius, and its own data centers — a portfolio approach that limits single-point dependencies in a market where compute availability has become a competitive constraint.</p><h2>What Comes Next</h2><p>The initial deployment of tens of millions of Graviton cores is structured with explicit flexibility to expand, suggesting that the current agreement represents a floor rather than a ceiling. As Meta's agentic AI products evolve and its infrastructure demands grow, the scope of the AWS relationship could scale accordingly.</p><p>Amazon's broader capex commitment of approximately $200 billion in 2026 indicates that supply-side constraints, while real, are being actively addressed. Whether that investment keeps pace with the level of demand illustrated by customers seeking to purchase entire tranches of Graviton capacity remains an open question.</p><p>For the neocloud sector, Meta's diversified spending across AWS, CoreWeave, and Nebius reinforces that the AI infrastructure market is not a winner-take-all environment — at least not at this stage. Both CoreWeave and Nebius now carry substantial forward revenue commitments from Meta, providing a degree of visibility even as they compete for the same underlying infrastructure contracts.</p><p>The convergence of Meta's $115–$135 billion 2026 capex budget, Amazon's $200 billion infrastructure investment, and the industry-wide scramble for Graviton capacity points to an AI infrastructure buildout that is accelerating, not plateauing — with purpose-built silicon increasingly at the center of it.</p><p>For more tech news, visit our <a href=\"/news\">news section</a>.</p>", "excerpt": "Meta has signed an agreement with AWS to deploy tens of millions of Graviton cores for agentic AI workloads, making it one of the largest Graviton customers globally. The deal deepens a long-standing partnership and also covers Meta's use of Amazon Bedrock — arriving weeks after Meta committed approximately $48 billion to AI infrastructure deals with CoreWeave and Nebius. Meta plans to spend between $115 billion and $135 billion in capital expenditures in 2026, nearly twice what it spent in 2025.", "keywords": ["AWS Graviton", "Meta AI infrastructure", "agentic AI", "Amazon Web Services", "AI cloud computing"], "slug": "meta-aws-graviton-cores-agentic-ai-deal-2026" } ```