The AI industry’s biggest vulnerability may be that its most expensive product is becoming easier to replace.
OpenAI charges up to $100 a month for its premium ChatGPT plan. Meta’s Muse offers free access with usage limits. Claude competes for coding work. Google’s Gemini comes bundled into an ecosystem spanning Gmail, Android and Google Workspace.
But the threat goes beyond one chatbot undercutting another. If businesses can redirect workloads between AI models without rebuilding their applications, model developers may struggle to retain customers even when their technology is excellent.
Microsoft has already made that choice easier. Its Copilot Studio lets developers select models from OpenAI and Anthropic for different tasks. Customers can keep Microsoft’s platform while changing the intelligence underneath it.
That arrangement gives Microsoft leverage. OpenAI and Anthropic compete to supply the model; Microsoft owns the environment in which the customer uses it.
The same problem appears in open-weight AI. Companies can download capable models, run them on their own infrastructure and avoid paying a proprietary provider for every request. The Washington Post reported in July that U.S. companies were increasingly adopting open-weight models from Chinese developers as the costs of closed systems mounted.
OpenAI’s predicament is that it helped create a market where customers can choose among competing sources of intelligence. If the models become sufficiently interchangeable, the developer’s brand may matter less than price, performance and convenience.
That makes the Yahoo comparison more relevant than a simple prediction that OpenAI will lose a technology contest. Yahoo remained a recognized internet brand, but recognition did not guarantee control over how people used the internet.
OpenAI could suffer a similar fate without disappearing. ChatGPT might remain popular while customers increasingly obtain intelligence through Microsoft, Google, Meta or their own locally hosted models.
The financial pressure is already worth watching. Reuters reported on October 8 that OpenAI’s annualized revenue for September was nearly $50 billion, below a previously suggested $70 billion figure. The comparison needs qualification because OpenAI and Anthropic calculate revenue from cloud partnerships differently. Still, the gap raises questions about the assumptions investors are using to value the companies.
Revenue alone cannot establish whether OpenAI has a moat. The more useful test is whether it can keep customers and maintain pricing as alternatives improve.
If a customer can switch models with minimal friction, OpenAI must continually justify its price. If an open-weight model becomes good enough, the customer may not need OpenAI at all. And if Microsoft or Google controls the interface through which AI is used, those companies may capture the customer relationship even when another company’s model performs the work.
OpenAI’s strongest defense would be to make ChatGPT, Codex or its agent ecosystem indispensable in ways competitors cannot easily reproduce. Its weakest position would be selling intelligence that customers treat as a commodity.
The risk is not that AI demand collapses. It is that demand grows while the companies producing the models struggle to capture the profits.
If the model becomes replaceable, the company that owns the customer wins.
Not financial advice.
Got a news tip or correction? Let us know
If you got something out of this, please chip in to keep this site running, or subscribe to go ad-free.
0 views