Amazon (AMZN) has spent years establishing one of the most formidable cloud businesses in the world, but the artificial intelligence explosion is requiring every big cloud provider to face a tougher question.
No longer is it enough to rent out computational power.
Customers are now looking for infrastructure that supports faster inference, greater workloads, real-time reasoning and increasingly complex AI agents capable of handling multi-step tasks without continuous human direction.
That change has made chips far more critical to the cloud story.
AMZN, Amazon.com, already derives much of its profit from its cloud computing unit, Amazon Web Services. Now, AWS is digging deeper into custom silicon as it aims to deliver clients more performance, better efficiency and cheaper prices at a time when demand for AI infrastructure remains fierce.
The company’s recent step might not garner the same buzz as a new Nvidia GPU, but it could matter a lot for Amazon’s AI ambitions. AWS has made its latest Graviton5-powered EC2 instances broadly available, providing users access to the most powerful CPU Amazon has ever manufactured.
AWS said Graviton5 is “purpose-built for the demands of agentic AI,” including real-time reasoning, code generation and multi-step task orchestration.
Amazon Graviton5 targets the agentic AI boom
The company is also making the Amazon EC2 M9g and M9gd instances powered by AWS Graviton5 available to all customers, extending the chip beyond preview.
Timing is key because AI infrastructure is fast becoming one of the most significant battlegrounds in innovation. Cloud vendors no longer compete just on storage, scale and software services. They are also competing on the underlying hardware.
Graviton5 is Amazon’s latest proprietary CPU designed to accelerate performance for workloads that require high core density, quicker memory, and efficient communication across working units.
The chip contains 192 cores per CPU, a substantially larger cache and and lower inter-core latency than the previous iteration. AWS says the new M9g instances give up to 25 percent better computing performance than the previous generation.
Related: Amazon’s bestselling storage shed with lockable doors is just $110
That kind of improvement matters because agentic AI might put different demands on infrastructure than traditional model training. AI agents generally need to do continuous reasoning, reply fast, write code, coordinate steps and manage various environments at the same time.
That makes the CPU more relevant.
GPUs are still important for training and intensive AI acceleration, but CPUs still do a lot of orchestration, data transport and application logic. In essence, AWS is saying that accelerators alone won’t drive the age of agentic AI.
The business also claims that M9g instances powered by Graviton5 can deliver 35% quicker web apps, 35% faster machine learning inference, and 30% faster databases.
M9gd instances are intended at workloads that need high-speed local storage and offer up to 11.4 terabytes of NVMe SSD storage with 30 percent more input/output operations per second than the previous generation.
That gives AWS another tool as enterprise clients hunt for ways to execute AI workloads without letting infrastructure costs spiral out.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
Amazon has the advantage of more control over the stack than many corporations that buy chips from outside providers. AWS can create chips for its own cloud workloads, optimize them for EC2 instances and offer them through conventional customer adoption pathways.
That’s not to say Graviton5 is a replacement for either Nvidia (NVDA) GPUs or Amazon’s own Trainium CPUs. That does mean AWS has yet another method to make its cloud platform more appealing as enterprises reconsider the cost of AI computing.
picture alliance / Getty Images
AWS custom chips sharpen Amazon’s AI strategy
The arrival of Graviton5 is a timely event for Amazon.
AWS continues to be one of the company’s most important earnings engines, and artificial intelligence has become one of the major development prospects inside that sector. Amazon said AWS’s AI revenue run rate is now over $15 billion. In its Q1 2026 results, AWS’s operating income increased to $14.2 billion from $11.5 billion a year ago.
Those stats explain why Amazon continues to pour money on bespoke chips.
The startup hopes to provide consumers more options throughout the AI stack. Nvidia chips are still at the heart of many AI applications. Amazon’s Trainium processors are designed for AI training and inference. Graviton chips, however, are focused on improving price-performance for general purpose and increasingly AI-adjacent applications.
That blend could become increasingly crucial as buyers seek beyond raw performance.
AI infrastructure is expensive. Companies are trying to run more models, deploy more agents and support more real-time apps without letting cloud expenses swamp the business case. If AWS can deliver major speed improvements while boosting efficiency, it gives Amazon a stronger argument against Microsoft (MSFT), Alphabet’s (GOOGL)Google Cloud and other cloud competitors.
Key takeaways from Amazon’s Graviton5 launch
- AWS Graviton5-powered EC2 M9g and M9gd instances are now generally available.
- AWS says M9g instances deliver up to 25% better compute performance than the prior generation.
- The chip is designed for agentic AI workloads, including reasoning, code generation and multi-step task orchestration.
- Meta, Uber and Snowflake are among the companies deploying Graviton for agentic workloads, according to Amazon.
- The launch gives AWS another custom-chip tool as cloud providers compete for AI infrastructure spending.
The consumer names are important too.
Amazon stated Meta is fully committed to Graviton at scale, including ambitions for tens of millions of cores for agentic AI activities. Uber and Snowflake are also using Graviton for agentic workloads.
That helps AWS market Graviton5 as more than just a chip upgrade. It’s a customer retention and customer acquisition tool at a time when the companies are determining where to build their AI systems.
The greater question for investors is whether Amazon’s custom-chip plan will help lift the AWS growth story without weighing too heavily on expenses.
Amazon is already spending heavily on AI infrastructure. That can fuel long-term revenue growth, but it also means Wall Street will continue to focus on capital spending, margins and how soon AI demand transforms into sustained profit.
But Graviton5 doesn’t solve all those questions itself.
But it does offer a glimpse of how Amazon wants to compete on AI infrastructure. AWS isn’t just depending on third-party processors or splashy product launches. Instead, it’s attempting to speed up the cloud itself, from the silicon up, making it cheaper and more efficient.
This may be an even bigger plus if agentic AI advances from demos into daily enterprise use.
Related: Amazon’s stock buybacks explained
#Amazons #chip #push #rivals #problem