AI's Platform War Has Started

Making models isn't enough anymore, the winners will be the ones who distribute them

In AI, the best model doesn’t win anymore. The most-used model wins. OpenAI and Anthropic have started to change the rules of this game. And they are doing it not quietly, but with billion-dollar moves.

Anthropic announced a $1.5 billion joint venture with Blackstone and Goldman Sachs a few weeks ago. This company will sell AI tools to medium-sized businesses and manage the integrations itself, like a consulting team. Anthropic’s existing Claude Partner Network, built with giants like Accenture and Deloitte, focuses on large corporate customers. But this new venture targets companies that don’t have enough engineers inside, but still need to change.

On the OpenAI side, a company called “The Deployment Company” is being built, with a value of $10 billion. Backed by TPG and SoftBank, this venture aims to speed up the adoption of OpenAI software by giving direct access to more than 2,000 companies in the partners’ portfolio. Sam Altman calls this an “AI cloud.” It’s not like AWS selling infrastructure without making software. It’s a structure that deploys its own product and is present in the field.

Unfair Competition

These moves bring up questions about unfair competition. Model labs are now playing on the same field as the startups that build apps on top of their models. Anthropic’s Claude Code competes directly with independent companies that build tools using the same model. OpenAI’s push into the “operator” layer threatens the business models of its own API customers.

The advantage here is three-sided: they know the model best from the inside, they have the lowest inference costs, and they can improve the model fastest using the data from user interactions. On top of $10 billion in funding and huge data centers, this becomes a very strong barrier to entry.

Still, Altman says, “Our goal is not to compete with developers.” It is reported that these companies still spend more than 70 percent of their resources on core research. The energy going into the application layer is a small slice of total capacity. But whether this is real comfort or a real promise, only time will show.

The Competition Door Is Not Closing, It’s Changing

Looking at all this, it makes sense to ask: what will independent developers and startups do?

The answer is not to run away, but to understand the rules of the game. Model labs focus on improving general abilities. Successful startups, however, first solve a specific user problem and then shape the model around that problem. Cursor did this by forking VS Code. First, it understood what developers needed; only then did it optimize the model for those needs. Product quality came before model quality.

This is why vertical expertise is still a strong card. A company with deep knowledge in a niche like health, law, or finance can beat a giant that offers a general solution. At the end of the day, Anthropic and OpenAI cannot focus on every sector at the same time. There are still truly wide and empty areas.

Model independence is also a critical advantage. Big labs are stuck with their own models. Independent startups, however, can use all models together, including open-source ones, and choose the best combination for each task. As open-source models become more powerful, a fine-tuned model in a specific area can perform better than closed alternatives. Or it can offer a huge cost advantage.

Human-focused control is another point of difference that is often missed. Model labs focus on autonomous systems and on the idea that the next update will solve everything. Independent developers, however, can build tools that push current limits by prioritizing user control, speed, and interaction, instead of waiting for model updates.

Finally, pricing. A company that doesn’t have to earn money per token can charge a high price per result. If an AI tool really takes over part of a person’s work, pricing based on the value of that work is a much more profitable model than selling infrastructure.

Final Word

History has played this game before. The best technology didn’t win — the technology with the widest distribution network won. VHS was technically weaker than Betamax. Windows was not more elegant than its alternatives. OpenAI and Anthropic know this lesson well. So the question is: where will independent developers and startups stand in this distribution war?

The answer is actually simple. Don’t fight the giants, you can’t win that war. Instead, go deep: pick a niche and be unmatched in it. And more importantly, don’t depend on a single model. OpenAI or Anthropic could raise prices tomorrow, or change their policy. Open-source models are really very important. A fine-tuned open-source model in a specific area can easily beat closed alternatives. In short, while there is still time, build your own infrastructure and own your own data. In the long term, independence is the best competition strategy.