Meta Platforms has locked in a multibillion-dollar pact with Amazon Web Services, committing to deploy hundreds of thousands of Graviton processors across its 32 data centers over the next three years. The deal, announced Friday, hands Meta access to tens of millions of Graviton cores—each chip packing 192 of them—for the demanding workloads of agentic AI. CPUs, not GPUs. That’s the twist here. Business Wire broke the news first, calling it a major expansion of their partnership.
Santosh Janardhan, Meta’s head of infrastructure, put it bluntly: “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.” CNET captured that quote amid details on Meta’s $135 billion AI spend this year alone.
Graviton5 packs a punch. Five times the cache of its predecessor. 33% faster core communication. 25% better overall performance. And it’s ARM-based, sipping power compared to x86 rivals. Nafea Bshara, AWS vice president and Annapurna Labs co-founder, told Reuters the agreement spans years and runs into billions. “We pass that savings on to the customers,” he said. Meta now ranks among AWS’s biggest Graviton users worldwide.
But why CPUs now? AI training guzzles GPUs from Nvidia and AMD. Inference—the running of models—shifts gears. Agentic systems, those autonomous agents writing code, planning tasks, or orchestrating multi-step processes, lean hard on CPUs. Wall Street Journal noted how this deal revives CPUs in an Nvidia-dominated world. Meta’s not alone. Startups scramble for Nvidia GPUs amid months-long waits, per X discussions from industry watchers like @the_vc_intern. Compute stacks bifurcating. Training on GPUs. Agents on CPUs.
Amazon’s chip push pays off. Graviton and Trainium together hit over $10 billion in annual run rate, CEO Andy Jassy said earlier this year. Two top customers wanted all of 2026’s Graviton supply—AWS couldn’t oblige. Now Meta joins Uber, Anthropic, and others betting on AWS silicon. Investor’s Business Daily watched Amazon stock hit a record Friday on the news, up over 3%.
Meta’s capex explodes to $115-135 billion in 2026, from $72 billion last year. Deals stack up: billions with Nvidia, AMD, Google’s TPUs in a $10 billion six-year Google Cloud pact, plus in-house MTIA chips via Broadcom. Yet it slashed 14,000 jobs recently. Spend big. Cut headcount. Chase AI dominance through Meta Superintelligence Labs.
And AWS? Trainium GPUs went to Anthropic in a $100 billion, 10-year commitment. Graviton fills the gap. TechCrunch called it a wild turn, millions of ARM CPUs for AI needs. Bloomberg pegged it as hundreds of thousands of chips rented via AWS. Bloomberg.
Markets reacted fast. Amazon closed up 3.45% at $263.88, after-hours ticking higher. Meta gained 2.39% to $674.89. Intel surged 26% on its own CPU resurgence narrative, tying into agentic inference trends. X buzz from @HedgieMarkets highlighted the shift: inference costs now 80% of AI spend, spilling beyond Nvidia.
This isn’t isolated. OpenAI ported code to ARM for Graviton, per earlier reports. Uber expanded on Graviton4 for rides. AWS claims 98% of top 1,000 EC2 customers use it expansively, 40% better price-performance than x86. GeekWire framed Meta as AWS’s latest big win.
Bigger picture. Hyperscalers design chips to cut Nvidia dependence. Google pushes TPUs. Microsoft Maia. Amazon leads with volume. But returns? Productivity gains lag, as 90% of CEOs admit. Anthropic loses on every token yet commits billions. Meta monetizes later—ads, maybe agents in apps. Circular bets: partners today, rivals tomorrow.
Agentic AI demands it. Real-time reasoning. Task coordination. CPUs excel where GPUs overkill. Graviton5’s efficiency scales to billions of users. Meta’s 32 data centers get the upgrade. AWS locks in revenue, margins from custom silicon.
Short term: AWS earnings Tuesday loom large. Long term: Who cracks inference first wins the next wave. Meta diversifies. Amazon supplies. Nvidia watches. CPUs back in play.
WebProNews is an iEntry Publication