The Future of ROCm: AMD’s Developer-First Bet for the AI Era

At TensorWave’s Beyond CUDA 2025 Summit, Anush Elangovan, VP of AI Software at AMD, laid out AMD’s vision for ROCm — and it’s refreshingly clear: developers come first.

Forget the fluff. This wasn’t a marketing pitch — it was a straight-up challenge to the AI industry to rethink what ROCm is and where it’s going.

The $64,000 Question: What’s Next for ROCm?

The question Elangovan says he was asked most at the summit:

“What the heck are we going to do about ROCm?”

His answer? “Developers, developers, developers.”

ROCm’s future is being built for and with the open-source community. AMD knows it hasn’t always nailed the developer experience, but they’re owning it — and fixing it.

From Internal Tools to Open Ecosystem

Historically, ROCm’s tooling and documentation lagged behind. That’s changing.

• AMD is building a robust CI/CD and open-source build system
• They’re making internal tools available to the community
• And they’re actively listening to developer feedback — not just pretending to

“We’re investing heavily in the developer ecosystem — so that external devs can do what internal AMD engineers do.”

ROCm Performance: MI300 vs. H200

Elangovan also shared new ROCm performance data comparing the MI300 vs. NVIDIA’s H200 on DeepSeek R1 workloads — across latency, inference, and throughput.

Full benchmarks are coming soon, but the sneak peek was clear: AMD’s MI300 is holding its own.

AMD’s New Mindset: Unlocking Your Potential

This was the biggest shift in tone: AMD isn’t positioning ROCm as a walled garden. They’re framing it as a toolkit to unlock your workloads — wherever you are, whatever you’re building.

“It’s not about what we can do on AMD. It’s about what we can unlock for you.”

It’s a subtle but powerful reframe. AMD doesn’t want to just compete with CUDA — it wants to move beyond it by offering freedom, flexibility, and performance.

Want to Build on ROCm? Just Ask.

Elangovan’s closing message was a direct call to action:

“If you want to run models, optimize performance, or deploy at scale — let us know. I’ll personally help.”

No layers of abstraction. Just support.

TL;DR

• ROCm is shifting to a developer-first mindset
• AMD is investing in open tooling, documentation, and ecosystem support
• New performance data shows MI300 competing head-to-head with H200
• AMD wants to unlock workloads, not just run them
• Need help building on ROCm? Just reach out

About TensorWave

TensorWave is the AI and HPC cloud purpose-built for performance. Powered exclusively by AMD Instinct™ Series GPUs, we deliver high-bandwidth, memory-optimized infrastructure that scales with your most demanding models—training or inference.

Ready to get started? Connect with a Sales Engineer.