Opinion
The AI battle has moved: It's no longer about who's smartest, but who owns the stage
The performance gap between AI models has nearly vanished. Now the winners are those who control where we already spend our time online.
In San Francisco, a city where artificial intelligence has become an ongoing performance, you learn to read the silence between the applause. And during 2025, something in the air changed. The audience still cheered for ChatGPT, but behind the curtain, a different play was beginning. The battle was no longer between artificial minds, but between those who hold the keys to the hall, those who decide who performs, how, and for whom.
The AI race started like a classic talent competition. In the first act, the rules were simple: who builds the best model, who shows the most impressive demo, who captures the audience. ChatGPT took the stage and grabbed all the attention with a 76.4% market share at the start of 2024. Everyone agreed: this was the leading actor, the strongest performer.
Competitors tried to replicate the magic, but the stage remained his.
Then, without anyone announcing an intermission, the lights went out. And when the curtain rose again, the second act had already been rewritten.
Act One: The Performance Competition
In the first act, everything revolved around one question: who performs better. ChatGPT led. With 800 million weekly active users and $10 billion in annual revenue, it established itself as the most trusted and leading AI assistant. The gap between it and competitors seemed unbridgeable: not just in market share, but in actual capability.
The audience voted with their feet. Developers chose to build on OpenAI's API, large companies signed contracts, and consumers returned to it daily. 77 percent of user traffic to ChatGPT comes from people typing the address directly, not from Google searches. That's a level of loyalty rare even for exceptionally successful consumer products, and it's built only through outstanding performance.
But it's hard to maintain such an advantage when everyone learns the same techniques. The data shows a clear trend: at the start of 2024, the gap between the strongest AI model and the tenth-ranked one stood at 11.9 percent. By year's end, the gap had shrunk to just 5.4 percent. Between the top two spots, only seven-tenths of a percent now separates certain performance metrics.
This was the stage nobody in the industry wanted to acknowledge: the saturation point. The models continue to improve, but most users no longer feel the difference. And when you can no longer tell who's the lead actor and who's just filling in on stage, the quality of performance stops being the deciding factor in the competition.
Intermission: The End of Performance Advantage
It happened gradually, then suddenly. Claude, Anthropic's model, started surpassing GPT-4 in code benchmarks, capturing 42 percent of the code generation market compared to just 21 percent for OpenAI. Google's Gemini matched ChatGPT in almost every task, and Meta's Llama models approached commercial quality. The technological advantage that defined the first act nearly vanished.
The signs appeared first among developers. AI-based code assistants, which until then seemed like the industry's great promise, jumped to an annual growth rate of 127 percent at their peak between March and April 2025, then crashed. Within just five months, a ten percent decline was recorded.
What actually happened? Thousands of developers tried the new tools in the initial wave of enthusiasm, but when they checked the actual impact on productivity, many simply stopped using them. Even when companies distributed free licenses, only about 30 to 35 percent managed to get most developers to actually adopt the tools in their daily work. The technology reached the stage where it's "good enough" for some tasks, but not game-changing.
This was the first warning sign: if even the most technologically savvy users lost interest when the models started looking the same, what does that mean for the rest of the market?
Act Two: The Venue Owners Take the Stage
When the curtain opened on the second act, the competition had changed. It was no longer about who puts on the most impressive show, but about who owns the stage, the box office, and the seats.
Meta, for instance, didn't build a better model. It simply inserted its AI directly into places where we already spend our time. WhatsApp, Instagram, Facebook. No download, no signup, no "join now." A third of WhatsApp users worldwide have already encountered Meta AI without intending to.
And it worked. At least in terms of numbers. Within months, Meta reached a billion monthly users. Yet the strange data point is that only 0.02 percent of chatbot users in the United States use it directly. In other words, almost nobody thinks of themselves as a "Meta AI user." The exposure is enormous, but it doesn't create emotional connection. It's a product that's just there, not a product people seek out.
Google did what Google always does: integrated its AI into places you can't avoid. The new AI feature automatically entered search results, and Gemini replaced Google Assistant as the default on Android. With 92 percent of the search market, Google managed to push its AI to almost every person on the planet.
Yet behind this success hides a similar problem. Gemini's independent market share dropped from 16.2 percent at the start of 2024 to 13.5 percent by mid-2025. Users encounter it almost involuntarily, but when they actually seek a smart assistant, they still return to ChatGPT. Distribution creates presence, not necessarily preference.
Then there's xAI, Elon Musk's company. A reminder that in artificial intelligence, even a genius with billions can lose if he doesn't have a captive audience. Despite an $80 billion valuation, public support from Musk, and integration within X (formerly Twitter), the Grok chatbot managed to reach only 27.8 million monthly users, 24 times fewer than ChatGPT. Revenue barely reaches a billion dollars a year, while expenses cross that number every month.
Without a platform of its own, xAI has to pay for every user who joins. Google and Meta, by contrast, distribute their products to hundreds of millions almost cost-free. In this act, you don't need to be the biggest star. It's enough to be the owner of the venue.
When Owning the Venue Is No Longer Enough
But even that's not a guaranteed recipe for success. The second act revealed a simple truth: if the product itself doesn't meet expectations, no distribution will save it.
Microsoft learned this the hard way. With more than a billion computers running Windows, hundreds of millions of Microsoft 365 subscribers, and almost a billion LinkedIn users, it controlled the most powerful distribution channel in the world. Two years were devoted to making Copilot, its smart assistant, the heart of all its products.
But when the moment of truth arrived, the numbers were grim. Adoption rate stood at just two percent. Eight million active users out of more than 400 million potential ones. Analysts called it "a resounding failure."
Microsoft tried to charge $30 per month for a service users found inferior to free alternatives. The exposure worked. Everyone saw Copilot, but almost nobody stayed. They tried it, compared it to ChatGPT or Claude, and moved on. Even owning the venue isn't worth much when the audience gets up and leaves mid-show.
In the enterprise world, patience is even shorter. Microsoft's market share in large models dropped from 50 percent in 2023 to 34 percent in 2025, while Anthropic doubled its share. Companies chose to forgo the convenience of "what's already included in the package" in favor of better performance. The competition returned to basics: quality beats accessibility.
The New Rules of Act Two
Even massive distribution doesn't guarantee success. Microsoft discovered this when trying to make Copilot the center of its software suite. Despite hundreds of millions of potential users, only two percent actually adopted the service. The exposure was enormous, but users tried it and moved on. In a world where quality is measured by results, even a giant like Microsoft can't afford a weak performance.
In the first act, everyone fought over who looks most impressive. In the second act, the winner is whoever holds the stage. Anthropic decided not to compete for users' hearts and instead chose to integrate deeply into Amazon's and Google's services. This way, it reaches millions with almost no effort. In a world where every AI query costs significant money, companies understand that the real battle is economic: how to distribute efficiently without losing money on every use. That's why the big partnerships of recent years aren't about "growth" but simply a way to stay on your feet.
The second act proved that the real battle is over the point of contact with the user. A smart model without distribution will remain on the shelf, and distribution without quality will drive the audience away. Whoever manages to connect the two reaps the rewards. The advantage has shifted to companies that spent years building our digital infrastructure, but even they have no immunity. An audience that wandered to another chat once will do so again. Entrepreneurs already understand that the question isn't "who's smartest," but "who's an inseparable part of the system."
Nicole Levin is an executive in a stealth AI startup in San Francisco.
