Performance increased by 25%! Amazon's new self-developed server chip Graviton5 is coming out, challenging the dominance of x86.

Amazon Web Services (AWS), a cloud computing service platform under Amazon, has launched a new generation of self-developed server CPU - Graviton5. AWS claims that this is the most powerful and energy-efficient central processing unit (CPU) in data center servers to date. The latest self-developed CPU and the self-developed AI ASIC chip Trainium3 launched at the cloud technology conference this week highlight the fact that this cloud computing and e-commerce giant is continuously expanding its chip product lineup and increasingly leaning towards self-development. With cloud computing giants such as Amazon, Microsoft, and Google launching their own self-developed server CPUs and AI ASIC series AI accelerators, their future demand for server CPUs from Intel and AMD, as well as overall demand for NVIDIA AI GPU computing clusters, may significantly decrease. Although AWS currently uses high-performance chip products from NVIDIA, Intel, and AMD in its large-scale data centers worldwide, its management has always emphasized the importance of providing customers with a diversified hardware configuration choice for cloud computing. Statistics show that for the third consecutive year, over half of the large-scale CPU-level computing clusters added by Amazon AWS are provided by Amazon's self-developed Graviton series CPUs, rather than predominantly from AMD and Intel in previous years. In a statement, AWS said, "Traditional methods often force you to compromise between speed and efficiency. To meet this demand, today we are launching the Graviton5 central processor - the most advanced self-developed server CPU from AWS to date, designed for a wide range of cloud workloads. Compared to the previous generation, Graviton5 can provide up to a 25% increase in computing performance while maintaining industry-leading energy efficiency, allowing customers to run applications faster, significantly reduce computing costs, and achieve sustainable development goals." Some of the large AWS customers who have chosen to use Amazon's self-developed Graviton series chips in large-scale application scenarios include Adobe, Airbnb, SAP, Snowflake, Synopsys, Atlassian, Siemens, Pinterest, and Epic Games. AWS stated, "High-performance and computationally intensive workloads such as real-time high-load games, high-performance databases, big data analytics, application servers, and electronic design automation (EDA) can now be scaled through faster data exchange between cores." Each Graviton5 core has 2.6 times the L3 cache capacity of Graviton4, and its network and storage bandwidth has also increased by 15% to 20%. The chip is manufactured using a 3nm advanced process, which is also used for AWS's newly released self-developed AI ASIC chip Trainium3. Amazon stated that the M9g instances based on Graviton5 for general workloads are currently in open preview; C9g instances for compute-intensive workloads and R9g instances for memory-intensive workloads are planned to be launched in 2026. The above release coincides with Amazon AWS hosting the 2025 re:Invent conference this week in Las Vegas. Amazon AWS has achieved an astonishing annual revenue run rate (ARRR) of $132 billion. What is Graviton5 architecture? Why are cloud computing giants developing their own server CPUs? Graviton5 is an ARM architecture server CPU developed by AWS, a customized server chip designed for Amazon cloud workloads. It is based on the Arm Neoverse series, offering higher performance at the same power consumption level and more power efficiency at the same performance level than equivalent options, aiming to significantly reduce Amazon's overall TCO. Arm, a leader in chip design, is also one of the biggest winners in the global artificial intelligence frenzy. NVIDIA's self-developed Grace CPU is based on the ARM architecture, and Amazon's self-developed Graviton server processor also uses ARM architecture. Similarly, Google's Axion Processors, the first self-developed ARM architecture data center CPU from Google, and Microsoft's Azure Cobalt 100, a self-developed ARM architecture data center CPU, are also based on the ARM Neoverse series. ARM architecture is evolving from the "king of smartphones" to a cornerstone of AI cloud computing infrastructure in the AI era. Why are cloud giants embracing self-developed server CPUs? When a cloud computing company reaches the scale of Google and Amazon, factors such as "performance per core/watt consumption/yearly electricity cost" become massive considerations. Self-developed CPUs can be optimized for the most common cloud workloads (microservices, databases, cache, cloud AI inference) on their own cloud computing platform, deeply integrated with their own compilers, operating systems, network stacks, storage, AI accelerators, resulting in at least a 10% improvement in efficiency and operating costs, multiplied by millions of chips, leading to significant savings. In addition, having self-developed alternatives gives them more bargaining power with Intel and AMD in terms of price and roadmap, and avoids being completely passive in case a supplier's technology or capacity falls short in the future. The rise of cloud giants' self-developed ARM architecture server CPUs may indeed "squeeze the growth space of x86 architecture server CPUs" in the long term, but it is not yet marginalizing Intel/AMD immediately, and more so limiting the ceiling for x86 in the cloud. For the "entire global server market," x86 architecture server CPUs (i.e., Intel + AMD) remain the absolute mainstream central processing unit in global data centers, especially in traditional enterprise cloud infrastructure, self-built data centers, and many general instances in the cloud. The share of ARM's self-developed server CPUs is mainly concentrated in the in-house data centers of a few super-large cloud factories, and is not yet shipping on a large scale globally.