AWS Graviton5 instances offer significant performance improvements. (Illustrative AI-generated image).
- The Graviton Journey: From Preview to General Availability
- What Makes Graviton5 Different?
- Real-World Results: ClickHouse, Honeycomb, and HubSpot
- Beyond the Benchmarks: Why This Matters for Cloud Workloads
- M9g vs M9gd: Choosing the Right Storage Option
The Graviton Journey: From Preview to General Availability
Amazon Web Services has been building its own custom processors for eight years now. The company started with an idea: what if you could design chips specifically for cloud workloads instead of using off-the-shelf parts from Intel or AMD? That bet turned into the Graviton line of processors.
Each new version has gotten faster and more power-efficient. And now the latest one is here. AWS just made its Graviton5-powered M9g and M9gd instances generally available to everyone.
The M9g was first shown off in preview at AWS’s re:Invent conference in December 2025. Back then, only a few customers could test it. They ran their software, measured the results, and came back with numbers that turned heads.
Now those same chips are available to any AWS customer who wants them. You can spin up a new instance in minutes. That is a big deal for companies that have been waiting to try the latest hardware.
The Graviton line is not just a side project for AWS. The company says that over 120,000 customers already use some kind of Graviton instance. There are more than 350 different instance types powered by these chips. That covers everything from simple web apps to heavy-duty databases and machine learning inference.
This latest generation is supposed to be the most powerful and most energy-efficient processor AWS has ever built. That is a strong claim. But early test results from real customers back it up.
What Makes Graviton5 Different?
Think of Graviton5 as a better engine for the same car. You do not need to change how you drive. You just get more speed and better fuel economy.
Here is the simple version: this is a custom Arm-based chip designed by AWS. It is the fifth generation of that design. Each generation has pushed performance higher while keeping power use low. Graviton5 continues that trend.
One of the biggest selling points is that you do not have to change your code to see gains. That is rare in the chip world. Normally, getting better performance means rewriting software to take advantage of new features. But AWS designed Graviton5 to be a drop-in upgrade for existing applications.
The chip is built for what AWS calls the “agentic AI era.” That is a fancy way of saying that AI workloads that take actions on their own, like chatbots that book appointments or automation tools that manage workflows, need fast, efficient compute. And according to some analysts, Graviton5 delivers up to 25% better performance for those kinds of tasks.
Energy efficiency is another big part of the story. AWS does not give out exact wattage numbers for its chips. But the company says Graviton5 is the most energy-efficient processor it has ever made. For businesses trying to cut their carbon footprint or reduce cloud bills, that matters. Lower power use means less heat, less cooling, and lower costs in data centers.
Compared to Intel and AMD offerings, Graviton5 is not trying to win on raw clock speed. It wins on price-performance. That means you get more work done for every dollar you spend. For most cloud customers, that is the metric that actually counts.
Real-World Results: ClickHouse, Honeycomb, and HubSpot
Numbers on a slide are nice. But real customer results are what matter. And the early adopters of Graviton5 have shared some impressive data.
ClickHouse is a fast open-source database for analytics. The company tested M9g instances against the previous generation M8g. They saw a 36% performance boost. And they did not change a single line of code. They just moved their workload to the new instances and got more speed. That is exactly the kind of result AWS wanted to show.
Honeycomb runs a monitoring and observability platform. That means they track how software behaves in production. They ran a six-month A/B test, comparing Graviton4 to Graviton5. The result: 36% better throughput per core. That is a big jump for a company that processes huge amounts of data every second. Honeycomb shared the results publicly, which adds credibility to the claim.
Then there is HubSpot. They make marketing and sales software. They moved their MySQL databases to M9g instances. The result was a drop in query duration of up to 60%. That means database queries that used to take a second now take less than half a second. For a company that handles millions of queries a day, that adds up fast.
These are not synthetic benchmarks run in a lab. These are real production workloads from companies with demanding applications. The consistency across very different use cases, analytics, monitoring, and databases, suggests the gains are real and repeatable.
Beyond the Benchmarks: Why This Matters for Cloud Workloads
Benchmarks are helpful. But they do not tell the whole story. The real question is: does this chip make a difference for everyday cloud workloads?
The answer seems to be yes. Graviton5 targets the kinds of tasks that make up the bulk of cloud computing. Web apps, microservices, analytics, databases, machine learning inference, electronic design automation, gaming, and video encoding. These are not exotic workloads. They are the bread and butter of the cloud.
For a company running a web application on a fleet of servers, a 36% performance improvement without code changes is huge. It means you can either handle more traffic with the same number of servers, or you can reduce your server count and save money. Either way, your cost per request goes down.
For observability platforms like Honeycomb, better throughput per core means they can monitor more services with less hardware. That keeps costs manageable as their customers grow.
And for databases like MySQL, faster queries mean better user experiences. A 60% drop in query duration can make an app feel snappier and reduce load times for customers.
There is also the energy efficiency angle. Data centers use a lot of power. Companies are under pressure to reduce their carbon emissions. Moving to more efficient hardware helps. AWS says Graviton5 is the most efficient processor it has built. That translates into lower electricity bills and a smaller environmental footprint for every customer who uses it.
M9g vs M9gd: Choosing the Right Storage Option
AWS launched two instance types at the same time: M9g and M9gd. They share the same Graviton5 processor. The difference is storage.
M9g instances use network-attached storage. That is the standard setup for most cloud workloads. Your data lives on remote storage volumes, which can be resized and moved independently of the compute instance. This is good for flexibility and durability. If an instance fails, your data is still safe on the network storage.
M9gd instances add high-speed local NVMe SSD storage directly attached to the server. This is for workloads that need extremely fast read and write speeds with very low latency. Think of it like the difference between a shared kitchen in a dormitory (network storage) and a private kitchen in your own apartment (local storage). The private kitchen is faster because you do not have to wait for anyone else.
NVMe stands for Non-Volatile Memory Express. That is a technical way of saying the storage is fast. SSDs are already faster than old spinning hard drives. NVMe makes them even faster by connecting them directly to the processor through a high-speed lane.
Who needs M9gd? Anyone running applications that need to read and write data at high speed with almost no delay. That includes databases that cache data locally, analytics that process large datasets in memory, and machine learning training that needs fast access to training data.
The trade-off is that local storage is ephemeral. If the instance stops or fails, the data on the local NVMe drives is gone. You cannot detach it and move it to another instance like you can with network storage. So M9gd is best for temporary data that can be rebuilt or for applications that handle data durability themselves.
For most general-purpose workloads, M9g is the right choice. It offers flexibility and durability. For high-performance needs, M9gd is worth the extra cost.
What’s Next? The Role of Graviton in the Agentic AI Era
AWS is positioning Graviton5 for what it calls the “agentic AI era.” That is a buzzword that means AI systems that do not just answer questions but take actions. Think of a virtual assistant that books meetings, orders supplies, or adjusts cloud configurations on its own.
These agentic workloads need a mix of compute, memory, and fast response times. They also need to run cost-effectively at scale. Graviton5 seems designed for that sweet spot. It is not a specialized AI accelerator like Nvidia’s GPUs. It is a general-purpose chip that handles the supporting tasks around AI: running the logic that decides what actions to take, managing databases that store context, and serving the user interface.
Analysts at Insider Monkey noted that Graviton5 delivers up to 25% better performance for agentic AI workloads. That is significant because these workloads are expected to grow rapidly. Every major cloud provider is racing to make their platforms better for autonomous AI agents.
Amazon also points out that Graviton5 is built for the agentic era specifically. That suggests future generations may include even more optimizations for these kinds of tasks. Maybe we will see features that speed up the decision-making logic inside AI agents. Or improvements in how the chip handles the rapid reading and writing of data that agents need.
One open question is how far AWS can push this. Intel and AMD are not standing still. They have their own new chips coming. And Nvidia dominates AI training and inference. But Graviton is carving out a clear niche: general-purpose compute that is highly efficient, cost-effective, and good enough for a wide range of workloads, including the new wave of agentic AI.
The early customer results from ClickHouse, Honeycomb, and HubSpot suggest that Graviton5 delivers real, measurable gains. For businesses running cloud workloads, the decision to try M9g or M9gd is simple: you might get 36% more performance or cut query times by 60%. And you do not have to rewrite your code to get it.
That kind of upgrade is rare. Cloud computing moves fast. But every now and then, a new chip comes along that makes a real difference. Graviton5 looks like one of those chips. The question now is how many customers will make the switch.
Frequently Asked Questions
What are AWS Graviton5 instances?
AWS Graviton5 instances are new cloud computing servers powered by Amazon Web Services' custom-designed, Arm-based Graviton5 processors. These instances are designed to offer significant performance and energy efficiency improvements for cloud workloads.
What is the main benefit of using Graviton5 instances?
The biggest advantage is that you can achieve up to 60% better performance without needing to rewrite your existing code. This makes it a straightforward upgrade for many applications, offering more speed and efficiency easily.
How much of a performance increase can I expect with Graviton5?
Real-world tests show significant gains, with companies like HubSpot seeing up to a 60% reduction in database query duration. ClickHouse reported a 36% performance boost, and Honeycomb achieved 36% better throughput per core.
Are Graviton5 processors more energy-efficient?
Yes, AWS states that Graviton5 is their most energy-efficient processor ever built. This means lower power consumption, which can lead to reduced heat, cooling needs, and overall operating costs in data centers.
Who is using Graviton processors?
Over 120,000 customers are already using some form of Graviton instance across more than 350 different instance types. These are used for a wide range of applications, from simple web apps to complex databases and machine learning tasks.
What kind of workloads are Graviton5 instances good for?
Graviton5 is designed for common cloud tasks such as web applications, microservices, analytics, databases, and machine learning inference. It's particularly suited for the 'agentic AI era' which involves AI workloads that take independent actions.
Do I need to change my software to use Graviton5?
No, a key feature of Graviton5 is that it's designed as a drop-in upgrade. You generally do not need to modify your code to benefit from the increased performance and efficiency it offers.
