AMD's Anush Elangovan explains why he thinks his company's open source ethos combined with agentic AI superpowers their leverage as a company: Because AMD publishes a lot of technical details about its hardware, when engineers use AI tools, the models already “understand” AMD’s systems and can help write code... for them, debug them, or even generate new tools. And that makes developers more productive on AMD hardware without AMD having to do all the work internally. "AMD has had this ethos of open source, which really plays to our advantage. Every frontier model that I use has already seen every bit of AMD source code." "It'll rewrite my spec for me because it's already in the training data. Which you can't get from closed ecosystems." "In fact I built a virtual GPU simulator just based off our public specs, and now I'm running it on the GPU. So now I can run cross-generational GPU simulations on existing hardware." "We have that advantage. And we've run a Dev Day contest where we generated more tokens on AMD — Triton kernels and HIP kernels — than existed on the internet at the time." "So now that's all part of the pre-training data. It's a superpower because now you're open source, and you're agentically accelerating this process."show more

Open source is the fastest way to build trust with engineers. When you publish your technical details, you're not just sharing code. You're sharing your roadmap. AI tools can read that roadmap and build on top of it. That's how you create an ecosystem. I've always believed in transparency. When I was rebuilding from zero, the only people who helped me were the ones who had been open about their own failures. Closed systems protect companies. Open systems build communities.

Anush makes a fair point, and there's logic to it, but looking at reality, CUDA still dominates by a wide margin. The amount of existing code, libraries, examples, and the developer community around NVIDIA is just on a completely different level compared to AMD. Sure, you can say "every model understands AMD," but these models understand NVIDIA way deeper and broader than that. And even if we assume models now understand AMD really well, that's not where the real battle is. At the end of the day, what a developer actually cares about is real-world performance and inference speed, not how well the AI knows the hardware. If AMD's ecosystem is slower or less mature, developers will keep choosing NVIDIA regardless of everything else. Now the "Agentic acceleration loop" idea, where models automatically generate Triton and HIP kernels and optimize performance on their own, that's genuinely interesting, because it could gradually reduce the need for manual CUDA expertise, which would chip away at one of NVIDIA's strongest advantages. But that path isn't complete yet, and we're still very much in the early stages

@AnushElangovan This is a really underrated compounding loop: public specs/code make models better at your stack, which makes builders ship more examples, which feeds the next models. For agent-heavy workflows, openness becomes distribution + developer enablement, not just philosophy.

The wrapper is not the moat. The real advantage is a model readable ecosystem: open specs, docs, source trails, tooling, governance, and control surfaces. When agents can understand the stack, they can accelerate it. Open source feeds the model. The model feeds the ecosystem. The ecosystem compounds… >

@AnushElangovan open docs help but the actual code corpus is still mostly CUDA. models default to nvidia syntax even with amd docs in context.

@AnushElangovan the open-docs-as-training-data effect is the part of amd's open source story that gets undersold. the more technical detail you publish, the better the models do at recommending your stack to engineers. it's a quiet distribution channel that compounds.

@AnushElangovan amd built the ai playground for devs, they get it

Love that AMD’s transparency gives AI models a head start — makes sense why their chips are hot for agentic work. Curious: does this openness extend to firmware docs too? (Also, just shipped a new dubbing pipeline at DangcingAI — feels like the same philosophy, but for video 🎥)

@AnushElangovan 🚀