The Only AI Moat is Hardware. And Compute is the Upper Bound for… | by Murat Onen | Medium

I have lost count of how many times I have been asked about DeepSeek over the past week — specifically, whether it signals the obsolescence of high-performance AI compute or, by extension, the beginning of the end for NVIDIA. The answer is “No.” — but if you still need more than one word, here is why. Over the past decade, there have been three major approaches (often used in combination) to develop frontier AI models: Performing R&D to develop new architectures that redefine AI capabilities, such as transformers, diffusion models, state-space models, and mixture-of-experts (MoE) architectures, or methods such as chain-of-thought (CoT) prompting. Exploiting the predictable relationship between scale and performance by increasing model size and training data. Adapting pre-trained foundational models to specific tasks or domains by training them on curated (often proprietary) datasets, improving efficacy and deployment viability. Only one of these approaches can provide a sustainable moa

I have lost count of how many times I have been asked about DeepSeek over the past week — specifically, whether it signals the obsolescence of high-performance AI compute or, by extension, the beginning of the end for NVIDIA. The answer is “No.” — but if you still need more than one word, here is why. Over the past decade, there have been three major approaches (often used in combination) to develop frontier AI models: Performing R&D to develop new architectures that redefine AI capabilities, such as transformers, diffusion models, state-space models, and mixture-of-experts (MoE) architectures