While Meta Cut 8,000 Jobs, It Ordered Hundreds of Thousands of Amazon Chips
This is not a contradiction. It is a business model.
While Meta Cut 8,000 Jobs, It Ordered Hundreds of Thousands of Amazon Chips
Meta is laying off roughly 8,000 workers. It is also spending billions on new AI infrastructure from Amazon. Both things happened this week, and they are connected.
On Thursday, Meta announced it would cut about 10 percent of its workforce — roughly 8,000 people — to offset rising costs tied to AI infrastructure investment. The same day, Amazon announced Meta had signed a deal to deploy hundreds of thousands of AWS Graviton chips for agentic AI workloads, making Meta one of the five largest Graviton customers in the world. CNBC
The two announcements do not require a conspiracy theory. They illustrate a structural reality of building AI at scale: the cost of human workers and the cost of compute are both real, and they are substitutes in ways they have never been before.
"Expanding to Graviton allows us to run the CPU-intensive workloads behind agentic AI with the performance and efficiency we need at our scale," said Santosh Janardhan, Meta's head of infrastructure, in a statement. The deployment starts with tens of millions of Graviton cores and is expandable. CNBC
The Graviton deal marks a reversal. Meta signed a six-year, $10 billion partnership with Google Cloud in August 2025, a move that made Google its primary cloud provider after years of running primarily on AWS. The new Graviton agreement brings business back to Amazon. Amazon declined to disclose the value of the deal. TechCrunch
Why Graviton? Because agents need CPUs differently than training does.
GPUs are the standard answer for AI compute — they train large language models at enormous scale. But once a model is trained and deployed, the workloads that keep agents running are different. They involve orchestrating multi-step tasks, real-time reasoning over context, code generation, and search — all of which AWS describes as "CPU-intensive." The Graviton5 chip, Amazon's latest, delivers 25 percent better performance than its predecessor while using 60 percent less energy. It has 192 cores and a cache five times larger than the prior generation, which Amazon says reduces core-to-core communication latency by up to 33 percent. Amazon corporate blog
Intel CEO Lip-Bu Tan put a point on it during an earnings call Thursday — the same day as the Meta layoffs. "In recent months, we have seen clear signs that the CPU is reinserting itself as the indispensable foundation of the AI era," Tan said. Intel reported Q1 2026 revenue of $13.6 billion, beating consensus estimates by 9.4 percent, with demand for Xeon server chips outpacing supply. The Motley Fool
The chip economics matter at Meta's scale. The company serves roughly 3.6 billion daily active users across its family of applications and is building 32 data centers to handle the load. Compute efficiency is not a nice-to-have — it is a line item that determines whether the company hits its margin targets. CNBC
Meta is not alone in repositioning its infrastructure. Anthropic signed a $100 billion, 10-year deal with AWS earlier in April, committing to Trainium chips — Amazon's AI accelerator — while also using Graviton processors. That deal effectively locked up a significant portion of Amazon's custom silicon supply before Meta could access it, a dynamic that may have accelerated the need for Meta to formalize its own Graviton commitment. Anthropic
The broader signal is to the chip industry. Nvidia has dominated AI infrastructure conversations for two years on the strength of its GPUs. But the agentic wave — autonomous systems that reason, plan, and execute complex tasks — is creating demand for a different kind of compute. ARM-based chips like Graviton and Nvidia's own Vera CPU are positioned for exactly this orchestration work, not because they replace GPUs but because they handle the coordination layer that makes agents practical at scale.
This is not a story about one company switching vendors. It is a story about what AI infrastructure actually costs when you build for agents rather than training, and what that cost forces companies to do at scale. Meta's 8,000 layoffs are a headline. The hundreds of thousands of Graviton cores are the balance sheet entry that explains why.
