The Layers of Commoditization of Generative AI: Which Areas Would Accrue the Most Value?

Author(s): Jesus Rodriguez

Originally published on Towards AI.

I recently started an AI-focused educational newsletter, that already has over 160,000 subscribers. TheSequence is a no-BS (meaning no hype, no news, etc) ML-oriented newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers, and concepts. Please give it a try by subscribing below:

TheSequence

The best source to stay up-to-date with the developments in the machine learning, artificial intelligence, and data…

thesequence.substack.com

One of the most recurrent questions within venture capital circles is which areas will capture the most value in generative AI. Undoubtedly, the model infrastructure layer, in the form of GPUs and large foundation models, has driven the biggest valuations and captured the most value. But will this trend continue? I don’t have the answer to that question, but as the co-founder of two generative AI companies, I’ve spent quite a bit of time thinking about it. One approach I’ve found effective is to adopt a first principles approach and tackle the problem using the inverse question: what areas of generative AI are more likely to be commoditized?

To answer that question, it might be good to look at the different layers of the generative AI stack. A hyper-simplistic view can be found in the following diagram.

Let’s go one by one and explore which layers we think are more likely to be commoditized.

1) Chip Manufacturers

This space is categorically dominated by TSMC, and it’s likely to continue that way. Outside of a gray swan event such as a geopolitical conflict in Taiwan, this area is unlikely to be commoditized. But also, building a new chip manufacturer at scale is short of impossible given TSMC’s monopoly.

2) GPU Manufacturers

The scarcity of GPUs has skyrocketed the valuations of companies like NVIDIA and AMD. Also, a new generation of highly specialized chip providers has emerged with very interesting valuations. If history serves as a proxy, GPUs should become more commoditized little by little. The ramp-up in manufacturing, plus the fact that companies like Amazon, Google, and Microsoft are entering the space, should contribute to this.

3) GPU Providers

GPU cloud providers such as CoreWeave, Lambda Labs, AWS, Azure, and Google Cloud are seeing crazy levels of demand. This layer is also likely to be commoditized as a result of the commoditization of the GPU supply chain.

4) General Purpose LLMs and Foundation Models

This is the most controversial area to evaluate. General-purpose LLMs seem unstoppable at the moment. Taking a somewhat contrarian view, there is a non-trivial probability that these types of LLMs might become commodities, and they will accrue less and less value outside the top 2–3 players in the spaces. Factors such as the rise of open-source LLMs are acting as strong commoditization vectors.

5) Domain-Specific LLMs

A new segment of smaller, highly specialized LLMs is emerging. Coding seems to be a dominant sector, but new domains are likely to emerge. This area is likely to capture a lot of value in the next few years and could be quite resilient to commoditization.

6) Generative AI Platforms

Infrastructure platforms for building generative AI applications are popping up everywhere. From super innovative startups to incumbents such as Databricks or Amazon, there are over a dozen high-quality platforms enabling the core building blocks of generative AI apps. If we think that every company in the world is going to incorporate generative AI capabilities, this category is likely to experience a good run for a while, but, again, following historical precedents, it’s likely to be dominated by 3-to-5 players. So, very defensible in the near future but will eventually experience some commoditization.

7) Autonomous Agents Platforms

You can consider agents as part of the app layer, but I like to differentiate it into its own category because I think autonomous agents are the automation paradigm of the generative AI era. Agents remain an unsolved problem in generative AI, without a platform that has achieved mainstream adoption. The first platforms that crack the agent experience are likely to capture a tremendous amount of value and remain very defensible.

8) Generative AI Apps

Finally, the application layer of generative AI. The first generation of apps have been basic wrappers around LLM APIs, but recently, we have seen platforms such as Poe or Perplexity that are pushing the boundaries of the space. User experience is being reimagined with generative AI and is likely to capture a lot of value. I used to think that applications relying on third-party LLMs were very vulnerable to commoditization, but I am starting to realize that, like in previous trends, good UX is very defensible.

9) New Compute Platforms

A special mention to a very unique category. Generative AI is likely to unlock new computing paradigms in the form of new hardware devices. This category is too nascent to evaluate qualitatively and presents monumental challenges. The opportunity to create a new generation of AI-first devices shouldn’t be ignored as a great value creator.

This is a relatively quick view of the commoditization layers in generative AI, which could be used as a proxy to evaluate which areas in the space are likely to capture the most value. My thinking in this area is constantly evolving, so I reserve the right to change my point of view in future posts U+1F609

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI