The Rise of the AI Application Layer

DeepSeek accelerated the shift in AI economics, making the application layer the biggest winner as intelligence becomes cheaper and more accessible.

People say DeepSeek is a big deal, even calling it a breakthrough and comparing its launch to the original ChatGPT debut. I agree, but for a different reason.

I believe DeepSeek is going to reshape the economics of software, making the application layer the clear winner in the AI stack. Foundational models are evolving rapidly, improving at an incredible pace, and competition between AI labs is getting even more intense. That’s great news for startups building at the top of the stack, as they rely on these foundational LLMs to power their applications.

Since the launch of ChatGPT, people and tech companies have been trying to understand where real value will settle in the AI ecosystem. Will it concentrate in the AI models themselves, or will it shift toward the application layer over time? This is a crucial question because companies need to decide where to invest in the AI stack as they plan their engineering/product roadmaps.

Let me paint two scenarios for you. In one, AI remains extremely proprietary and expensive. In the other, AI becomes nearly free and relatively open. These two worlds would lead to completely different outcomes.

If AI stays expensive and proprietary, the companies building these models will dominate, capturing most of the value while leaving little room for developers. But if AI becomes cheap and widely accessible, the real value will come from how it is used. The ability to build meaningful applications on top of these models will matter far more than the models themselves, shifting the economic advantage to the application layer.

With DeepSeek’s latest launch, it feels like this question is finally being answered, and the shift is clearly moving toward the second scenario. We have already seen consistent cost reductions and quality improvements from AI labs over the past few years, but DeepSeek takes this trend to a whole new level.

As the cost of intelligence continues to plummet, more value will flow back to the application layer. The products that successfully combine AI, UX workflows, and unique datasets (fine-tuned models) will be the ones creating the most value in the future.

Everyone loves to frame this as a battle of winners and losers, but it is not that black and white. Leading AI labs will incorporate lessons from DeepSeek into their own models, driving intelligence costs even lower and making AI more affordable across a broader range of applications.

If AI becomes ten times more efficient today, it is almost certain that we will find a hundred times more uses for it in the coming years. That surge in demand will only make GPUs and data centers even more critical than they are now.

DeepSeek is a massive win for software developers building at the application layer, and it is going to push AI labs to keep raising the bar. Incredible times, indeed. It’s time to build.