The Commoditization Trap—Why Model-Agnostic AI Products Will Win
LLM companies like Anthropic & OpenAI are really caught in a catch 22.
Their models are commoditized so quickly that they likely can’t even recover the cost of training them.
The answer would seem to be to build defensible product around those models.
The problem is their products are intrinsically linked to their models. The commoditization of the model thus devalues the product.
That leaves an opening for model agnostic products to “cross their product moat.”
— Me, Saturday on Threads
The economics of large language models don’t make sense right now. Companies are pouring billions into training these models. According to the Wall Street Journal, OpenAI is spending $500 million on each training run of GPT-5 and they’ve had to do several training runs. And yet the brutal reality is they will probably never recover those costs before GPT-5 becomes yesterday’s news, assuming it ever gets released, since the inference costs are so astronomical.
A model that’s state-of-the-art today will be matched by open-source alternatives within a year and a commercial competitor will likely meet or beat it within months. Models go from cutting-edge to commodity in 12-18 months.
The major players know this. It’s why they’re all racing to build products around their models. The thinking goes that while the models themselves might become commoditized, they can build lasting value through applications and platforms. This isn’t completely wrong: software products are generally more defensible than raw technology. They benefit from user lock-in, network effects, and integration into business processes.
But there’s a fundamental flaw in LLM companies taking this strategy. The products they’re building are locked in to their own models, preventing users from choosing a competitor’s model. OpenAI’s products can only use GPT or o-series models. Claude only works with Anthropic’s models. By tying their products to their models, they’re actually making their products vulnerable to the same commoditization pressures that affect the models themselves.
Think about it: if your product’s core value proposition depends on having the best model, what happens when your model is no longer the best? Or when a cheaper, good-enough alternative becomes available? Your product’s competitive advantage erodes along with your model’s superiority.
This is why I’m convinced of the competitive advantage of model-agnostic products. Instead of betting everything on a single model, companies should build products that can work with any capable LLM. When models become commoditized and lose their competitive edge—and they will—model-agnostic products actually benefit. They can take advantage of cheaper or more capable models as they become available.
I like to think of it as keeping the models hot-swappable in your architecture. Your product’s value isn’t tied to any particular model; it’s in how you solve specific problems for your users. You focus on building great user experiences, optimizing workflows, and adding unique value through your understanding of your users’ needs. The model layer becomes just another component you can upgrade when better options become available.
This approach lets you effectively cross the “product moat” that LLM companies are trying to build. While they’re focused on vertical integration—owning everything from the model to the end-user application—model-agnostic products can focus entirely on solving user problems. Let the market’s natural evolution handle the model layer.
The implications of this go beyond individual product strategies. The market will eventually catch on to this commoditization trap (the explosion of news around R1 certainly was a wake up call to investors) and realize that the highest returns from generative AI won’t be realized by investing in companies building the biggest models or the most computing power. It will belong to those who can build bridges between raw model capabilities and specific user needs.
As technologists and nerds, it’s easy for us to get caught up in the LLM arms race and think every shot fired is going to “change everything.” But in a world where state-of-the-art models become commoditized with increasing speed, adaptability and user focus matter more than raw model superiority. The companies that understand this—and build their products and design their architecture accordingly—will be the ones that thrive in the long run.