AI Coding Is Standardizing the Software Stack, and I Think It Pushes Servers to ARM

AI is picking the stack

When people vibecode, the prompt is usually about the outcome and less about the architecture that gets you there. There's nothing wrong with the architecture the AI chooses, but it leads to fewer projects in other fantastic languages like Go and Rust, because those aren't the first ones the models target.

GitHub's Octoverse 2025 report found that TypeScript became the most used language on GitHub in August 2025, with more than 2.6 million developers contributing each month, up 66.6 percent from a year earlier. GitHub credits part of that rise to AI coding tools, which tend to work better with typed languages, and points out that most popular web frameworks now start new projects in TypeScript by default, though it stops short of saying AI caused the shift.

That stack runs on single-core speed

For hardware, what matters is single-core performance specifically. I saw it in large SQL operations on one of our own x86 builds.

Node.js is designed to serve many users from a small number of threads, and PostgreSQL gives each connection its own single process, so with this stack one large request generally runs on one core at a time.

Why I think ARM wins the next refresh

ARM has higher performance per dollar in single-core speed, and especially at high core counts. I use a Mac for development, and the Macs are incredible at performance per watt. On the server side, you can load an 80-core Ampere CPU all the way up with nearly no performance hit, provided the hardware has adequate cooling.

According to Ampere's documentation, they offer up to 128 cores at an all-core speed of up to 3.0 GHz. Amazon makes a similar case for its own ARM chips, Graviton, which it says give the best price performance on its cloud.

I think the next big server refresh cycle will swing heavily toward ARM. I expect ARM to gain 7 to 10 percent of the market a year and plateau around 50 percent. TrendForce is already forecasting a jump from just over 10 percent in 2025 to about 30 percent in 2027, which is at the top of that range.

Where x86 still makes sense

There are good reasons to stay on x86, and I agree with all of them. We use it ourselves simply because we already own it. Some software won't run on ARM at all, such as SQL Server on Linux, which Microsoft only supports on x86 processors. A lot of older Windows and business software was built for x86 too. Switching also costs developer time to move the database and retest the application, even if hosting is cheaper afterward. And for some companies, raw throughput isn't needed in the first place.

The question to ask your developer or IT vendor

I want business owners to think about whether they're using the right choice for their deployments and their development. Could you see an increase in performance by leaving MSSQL on x86 and going to Postgres on ARM? Postgres runs on ARM and SQL Server on Linux doesn't, so making that move means changing your database as well as your hardware.

See our AI consulting services

See our software development services

Related: Why AI spending is not paying off yet