1 hour ago
1
AWS just made its fifth-generation custom silicon generally available, and the numbers tell a story about where cloud computing is headed. Graviton5, which launched on June 10, 2026, delivers up to 25% better compute performance compared to its predecessor Graviton4, with a particular emphasis on AI and compute-intensive workloads.
The chip’s architecture is not subtle. We’re talking 192 cores and a L3 cache that’s five times larger than the previous generation, paired with higher networking and EBS bandwidth. For anyone running large-scale cloud operations, that translates into doing more work per dollar spent, which is the metric that actually matters.
What’s under the hood
Graviton5 powers Amazon’s new EC2 M9g and M9gd instances, with C9g (compute-optimized) and R9g (memory-optimized) instances also slated for release in 2026. The processor had been in preview since December 4, 2025, meaning enterprise customers have already been kicking the tires for months before this general availability rollout.
AWS claims Graviton5 offers 30-40% better price-performance overall. In English: you get meaningfully more compute per dollar compared to the alternatives. That’s distinct from raw performance gains, which sit at around 25%. The difference comes from the chip’s efficiency, since AWS designs these processors specifically for its own infrastructure rather than buying general-purpose silicon off the shelf.
This matters because cloud costs are the line item that keeps CTOs up at night. When your chip is purpose-built for the workloads your customers actually run, you can squeeze out efficiencies that commodity hardware simply can’t match.
The custom silicon strategy has been a long game for AWS. The company launched its first Graviton processor back in 2018, and each generation has progressively reduced the company’s dependency on third-party chip makers. Graviton processors now power the majority of new AWS CPU capacity, which is a remarkable shift for a company that once relied entirely on Intel and AMD.
Meta’s multibillion-dollar bet
Perhaps the most telling signal about Graviton5’s significance isn’t the spec sheet. It’s who’s buying.
Meta signed a multibillion-dollar agreement in April 2026 to deploy tens of millions of Graviton5 cores for its AI infrastructure. That’s not an experiment or a pilot program. That’s a company with some of the most demanding compute requirements on the planet making a strategic commitment to AWS’s custom silicon.
DeepSeek has also embraced the technology, adding another high-profile name to the Graviton customer roster. When AI companies, the organizations most obsessed with compute performance per dollar, are choosing your chip, it validates the architecture in a way that benchmarks alone cannot.
AWS says Graviton chips now serve 98% of its top 1,000 EC2 customers. That adoption curve looks less like gradual market penetration and more like an industry default setting. The general availability launch extends access beyond the initial preview users, opening the floodgates for broader enterprise adoption.
Why this matters for investors
Look, a chip launch from a cloud provider isn’t normally the kind of thing that moves markets. But Graviton5 sits at the intersection of two trends that are reshaping the tech industry: the race for AI compute and the economics of cloud infrastructure.
AWS designing its own silicon means higher margins on cloud services. Every percentage point of improved price-performance is a percentage point that either goes to Amazon’s bottom line or gets passed to customers as a competitive moat. Probably some combination of both.
The Meta deal is particularly instructive for understanding the competitive dynamics. When the largest AI companies start locking in multibillion-dollar commitments to a specific cloud provider’s custom hardware, it creates switching costs that make churn extremely unlikely. You don’t casually migrate tens of millions of cores to a different platform.
For the broader cloud computing market, Graviton5 raises the stakes for Google Cloud and Microsoft Azure to accelerate their own custom silicon efforts. Google has its Axion processors, and Microsoft has been developing Cobalt, but neither has demonstrated the kind of enterprise adoption that AWS is showing with Graviton. The gap in custom chip maturity could translate into a gap in cloud market share over time.
The timing also aligns with explosive growth in agentic AI workloads, the kind of always-on, inference-heavy tasks that benefit most from efficient CPU architectures rather than raw GPU brute force. As enterprises move from training large models to deploying AI agents at scale, the economics of CPU performance become increasingly critical.
Here’s the thing worth watching: if Graviton5 adoption follows the same trajectory as previous generations, the competitive advantage compounds. Each generation of custom silicon that performs well attracts more customers, which generates more revenue, which funds the next generation. AWS has been running this flywheel for eight years now, and the momentum shows no signs of slowing.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) · 