Last year people started saying AI would become a commodity. I dont believe its going to happen.
What makes something a commodity
There are two parts: what is being sold, and how it reaches you.
The part that already looks like a commodity
Models are converging. Claude, GPT, Gemini — their capabilities are getting close. Like electricity: whether you buy from PG&E or a provider in Texas, it powers your bulb the same way. If you need AI to answer questions, read emails, or help with code, you can get similar results from any of the frontier models. We are almost there. Almost.
The part that does not look like a commodity at all
How does AI reach us? Through agents: ChatGPT, Claude, Gemini. These agents add tools to the models, which makes them useful. But you cannot move AI that these tools provide from them to another. For example you are using a todo app to manage your todos. Now you need AI to help you manage these tasks and lets say the app provides an option to bring your own AI provider. Good. But how do you bring AI from what open ai provides in chatgpt to the todo app you have. you cannot! my guess is you never will be able to. Currently we are buying electicity and the lamps from the same company. Electricity they provide can only be used in the lamp they are selling.
you may say, they do sell AI through their apis. we can use it in the todo app. but you would be paying chatgpt subscription and then you would be paying for the api separately. so eventually you would be like lets just use chatgpt alone. todo app is out. Not just that: API pricing is completely different from subscription pricing. $20 claude subscription gives you ridiculously lot more AI than $20 of api credits for the same model. I built a sample app today with both. The API credits ran out before I finished two screens. The subscription got me through the entire app, even with rate limits pausing me a few times.
This pricing gap is the distribution lock.
Subscriptions are sold at a loss to capture market share and usage data. That data trains the next model. APIs are priced to encourage you from building your own wrapper on top of them at-least for consumer use cases. May be they are making a profit when they sell it through apis atleast. So you pay twice if you want AI anywhere other than the agent: once for the subscription you already use, and again for API access elsewhere. Why this matters
AI should empower everyone. It should be a shared utility, like electricity. But the way distribution is locked right now, we are going to see fewer choices, not more. These agents will absorb every tool we use because paying twice does not make sense. We will use ChatGPT for everything, or Claude for everything, because it is cheaper than integrating AI into the tools we actually want to use.
What about open-weight models
Open models have come a long way. Qwen, Gemma, models from z.ai — they work. But they are not close to Claude or GPT-4. When you need to get real work done fast and right, you pick the frontier model. If frontier models did not exist, open weights would feel like magic. But they do exist, so by comparison, consumers will not choose them for real work that has consequences.
Is there a solution
Electricity did not become a commodity on its own. It happened through regulation. Could AI be regulated the same way so it reaches consumers through all channels at the same price? With how much money is being poured into frontier AI, companies valued at trillions even while private - chances are very low. The distribution lock is the moat. They will not give it up.