AI coding erodes Nvidia's CUDA moat, opening AMD's 2 GW door

AI coding agents are dismantling the software barrier that has kept Nvidia's CUDA platform untouchable for two decades.

AI coding agents are dismantling the software barrier that has protected Nvidia's CUDA platform for two decades, giving AMD a real opening to capture share of the AI accelerator market. Major AI cloud providers — OpenAI, Meta, and Anthropic — have all announced significant AMD GPU purchases since the second half of last year, a shift that would have been unthinkable when Nvidia GPUs were in their most severe shortage.

"Nvidia's CUDA moat is being dismantled rapidly because AI can write code — building the software layer around chips is much easier than before," Liang Wenfeng, founder of DeepSeek, said in a leaked four-hour conversation that circulated widely in the industry.

The most telling moment came at AMD's Advancing AI 2026 conference, where Anthropic co-founder Tom Brown announced the company would deploy 2 GW of Helios racks. Asked by AMD CEO Lisa Su why Anthropic had only now adopted AMD hardware, Brown said his team tested a single AMD rack by giving Anthropic's Claude model a simple instruction: "Let's bring up this machine." Over the weekend, the model's performance improved on its own. The audience broke into applause.

If AI-written software erodes CUDA's lock-in, AMD stands to benefit most as the world's second-largest GPU maker. Semiconductor research firm SemiAnalysis published an analysis titled "Can AMD Break CUDA's Moat?" arguing that as AI agents take over work previously performed by human software engineers, the importance of the CUDA moat is diminishing, allowing AMD to offset Nvidia's engineering headcount advantage.

The 10-Hour CUDA Rebuild

Jeremy Nixon, a former Google Brain researcher and founder of AI software startup Infinity, told Business Insider his startup used AI coding agents to recreate CUDA-like software for chip startup D-Matrix in 10 hours — evidence that one of Nvidia's biggest competitive barriers is being crossed. Cloud giants Google, Amazon, and Microsoft have spent years building software around their own AI chips, while Amazon's internal documents identified CUDA as a major roadblock to adoption of its Trainium and Inferentia AI chips.

The shift toward AI inference — where models answer requests rather than train — also weakens CUDA's grip. Companies care less about maximizing performance with the most powerful chips and more about running AI profitably, said Marshall Choy, chief business officer of Korean AI chip startup Rebellions. "That's where the CUDA moat from Nvidia gets broken because CUDA is no longer a factor in the inference side," Choy said. "It's an open source play."

Verification Becomes the Next Battleground

Not everyone agrees the moat is cracking. Bing Xu, founder of AI software startup INT21 and a former Nvidia engineer, said AI-generated code still has to be verified and optimized. He believes CUDA's verification tool suite could become its next competitive advantage. "Agents can generate a lot of code in a short time, but verification is the biggest bottleneck," Xu said.

Chris Lattner, cofounder and CEO of Qualcomm-owned AI software startup Modular, called the hype "very overblown," noting that writing code is only a small part of building software compared to optimizing it for production. Chip software is also a niche field with fewer examples for AI to learn from than app development.

Nvidia is not standing still. The company said it uses AI coding agents to develop CUDA faster and validate at greater scale, according to Ankit Patel, Nvidia's vice president of developer ecosystem. Nvidia also recently signed the largest memory deal in history with SK Hynix, securing HBM supply for its Vera Rubin platform while SK Telecom builds a 2 GW AI cloud data center in Korea.

For investors, the question is whether competitors can catch Nvidia faster than it can extend its lead. Nvidia's stock has been stagnant over the past year, trading at a forward P/E of just over 15 times analyst estimates for fiscal 2028. AMD, meanwhile, has the hardware specifications to compete — the question is whether AI-written software finally removes the barrier that has kept customers locked in.

This article is for informational purposes only and does not constitute investment advice.