The person who invested $13 billion in AI is warning everyone to be wary of AI vendors.
In July, Microsoft CEO Satya Nadella brought up the one subject Silicon Valley tends to discuss quietly, if at all. His point: the big companies selling proprietary AI models are building structures that lock enterprise customers deep into their ecosystems, and companies that adopt AI without recognizing this may find their options gone later.
The remark carries extra weight because Nadella himself runs the company that has served as OpenAI's largest outside investor. His warning isn't aimed only at enterprise procurement teams. The same question applies to Korean solo entrepreneurs, freelancers, content directors, and professionals who pay a monthly subscription to use AI in their work: could you switch to a different tool right now, if you had to?
Why Adopting an AI Tool Is the Start of Dependence
Adopting an AI tool tends to follow the same arc. You start with a free plan or a cheap subscription. It turns out to be convenient, so you open the same tool to draft an email, clean up meeting notes, or write a client proposal. Six months in, your work visibly slows down whenever you don't have it open. A year in, you suddenly realize you can no longer produce the same output without it.
What happens if the vendor raises its prices at that point? What if it moves a core feature behind a paywall, slashes the free tier, or quietly changes its API policy? In theory, you switch to a competitor. In practice, that means rebuilding your prompt structures from scratch, redesigning workflows built around the old tool, and retraining your team's habits.
This switching cost is exactly what Nadella pointed to. At the adoption stage, AI vendors only show you the API rate or the subscription fee. But the point where an organization actually gets locked in sits much deeper: data structures, the new work habits employees have picked up, processes optimized around a specific tool. None of that migrates overnight just because you're willing to pay for it.
This isn't a new pattern. The same thing played out in the ERP software market of the 2000s and the cloud infrastructure market of the 2010s. Moving from AWS to Azure was never just a matter of swapping servers — the entire codebase, security policy, and operating model had to move with it, a process that took months or years. The AI model market is walking the same path, just faster.
As of 2024, AWS, Azure, and Google Cloud together account for more than 65% of the global cloud infrastructure market. The AI model market is consolidating in a similar direction. As a given AI provider's enterprise customer base grows, so does the cost for those customers to switch to a rival's model. Nadella's warning arrived at a point where that consolidation is already well under way — and the timing matters, because if you only recognize a lock-in structure after you've adopted it, the cost has already been incurred.
The Company Issuing the Warning Is Building the Same Structure
Microsoft itself is one of the biggest beneficiaries of exactly this kind of AI dependency.
Microsoft 365 Copilot builds AI directly into Teams, Outlook, Word, and Excel. Once a company comes to depend on those features, the cost of leaving the Microsoft ecosystem isn't just the software license — it's the working habits of hundreds of employees that would have to move with it. Microsoft's AI integration strategy also happens to be one of the most effective tools for keeping enterprise customers locked to its platform.
That's why some in Silicon Valley read Nadella's comments less as a neutral warning and more as positioning against the proprietary AI strategies of rivals like Google, Amazon, and Meta. Microsoft has, in fact, recently started offering open-source models such as Llama and Mistral on Azure — a "we support an open ecosystem" message. But those models still run inside Azure. The dependence on Microsoft's cloud infrastructure doesn't go anywhere.
Seen this way, Nadella's statement carries both a fact and a subtext. The fact: organizations that depend heavily on a proprietary AI model become vulnerable to that vendor's policy shifts. The subtext: the very company issuing this warning is building the same kind of structure. Miss the subtext, and you've only read half the warning.
No AI vendor is going to volunteer "we're locking you in" — not Nadella, not OpenAI, not Google. Just as investors who actually read a company's disclosures make better calls than those who only skim its marketing materials, professionals adopting an AI tool need to start by actually reading its terms of service and data policy. Is your input used to train the model? What happens to your data after you cancel? How much advance notice do you get before the API policy changes? None of this is complicated. Most people simply skip the step.
What a $20-a-Month Subscriber Should Be Asking Right Now
All of this can sound like a story about big corporations — companies with six-figure AI contracts. But the same structure operates just as much on a Korean solo entrepreneur paying $20 a month to use AI in their work. Only the shape of the dependence is different.
For a large company, the dependence shows up in contract terms and infrastructure switching costs. For a solo operator or an individual professional, it shows up as an erosion of capability — reaching a point where you can no longer produce the same output without a specific AI tool. It's no longer being able to write your own meeting notes once AI has been organizing them for you, no longer being able to structure your own content once AI has been doing it, no longer trusting your own judgment on a proposal once AI has been reviewing it. That kind of loss is harder to reverse than any monetary cost.
Imagine a Korean freelance copywriter who has spent a year drafting with ChatGPT, editing with Claude, and pulling keywords with Gemini. What happens if one of those three tools doubles its price, restricts a feature, or changes its regional availability? The other two can pick up the slack. But if all three start moving in the same direction, and that copywriter can no longer work at the same pace without them, the options really have narrowed.
The point isn't to stop using AI. It's that holding onto the ability to make the same judgment calls regardless of which tool you're using matters just as much as using AI in the first place. A professional who can lose access to ChatGPT and keep working with Claude, or lose Claude and switch to something else without missing a beat, isn't being pulled around by their tools. They understand the underlying logic well enough to reach the same conclusions no matter what changes around them. The faster the business environment moves, the wider the gap grows between people locked deep into one platform and people who've kept that underlying judgment intact.
There's really just one thing worth checking: of the tasks your AI tool currently handles, how many could you still judge and handle yourself without it? If that question makes you uncomfortable, now is the time to find out.
Nadella didn't offer a specific solution. He didn't say which AI tool to use or which vendor to avoid. The warning itself was the message: recognize what structure your current AI tools are quietly building, and understand that the responsibility for checking that structure sits with you, not the vendor. The gap between using AI and being used by AI is invisible at first. You can only start to control it once you see it.
