Three years ago, "add AI to this" almost always meant "add ChatGPT." Not because anyone had done a rigorous vendor comparison. Because ChatGPT was AI, in the way that Google was search. The default assumption was structural. It showed up in vendor contracts, integration architectures, employee training programmes, and executive briefings. For the three years since its launch, the assumption held.
It stopped holding in May 2026.
I can't say I'm surprised — in fact it's overdue, given the operational risks of relying on just one vendor's model to drive business outcomes.
What the numbers say
For the first time since ChatGPT launched in November 2022, it holds less than half the global AI assistant market. Sensor Tower puts its share at 46.4% as of May 2026. That is still the largest single share in the market. But it is no longer a majority.
Google Gemini is at 27.7%, up from a smaller base just months ago. Claude is at 10.3%, with an absolute user count of 245 million — roughly four times what it was in December 2025. To put that growth rate in context: Claude added more users in five months than most enterprise software companies accumulate in their entire history.
The mechanics behind the shift are specific. Gemini's growth is structural, not a capability win. Google made Gemini the default AI on Android, replacing Google Assistant at the operating system level, on the world's most widely deployed mobile platform. When a model ships as the default experience on a few billion devices, the user numbers follow.
Claude's trajectory is different. Part of the surge tracks to February 2026, when OpenAI announced a partnership with the US Department of Defense. Analytics firms recorded a 295% day-over-day surge in ChatGPT uninstalls in the United States immediately after the announcement, and a corresponding 51% spike in Claude downloads in the same window. Claude reached number one on the US App Store within days. Whatever people thought of the deal, a segment of users voted with their feet.
What is striking in the Claude numbers is not just the growth rate but the conversion rate. Thirteen percent of Claude's users pay for a subscription — the highest paid conversion rate among the major AI assistants. The users it is attracting are not casual dabblers.
What this means if you run or advise a business
Three things have shifted in how AI vendor strategy needs to work.
The single-vendor assumption needs explicit review.Most enterprise AI deployments were built at a moment when ChatGPT was the obvious default. That was a reasonable shortcut in 2023 and 2024. It is a less reasonable shortcut in 2026, when two substantial alternatives have materially different strengths, pricing structures, and policy stances. Vendor selection should be a deliberate decision, not a default. The question to ask is not just which model you use today, but what your business would have to rebuild if that model changed behaviour, raised its prices, or retired the version your processes depend on.
Distribution and platform strategy matter more than anyone expected.Gemini's growth is not primarily a capability story. If your business is building on top of an AI platform, the distribution channel that platform controls is now a material strategic variable. A model embedded at the OS level has a structural advantage that benchmarks do not capture. For businesses evaluating AI platforms to build on, this is worth examining explicitly — platform staying power and ecosystem reach are now legitimate procurement criteria, not secondary considerations.
AI Operating System
Dynome's AI Operating System is designed from the start to be model-agnostic. Your business's AI capabilities are built as an architecture — not as a dependency on any single provider. When the market shifts, you shift with it, without rebuilding from scratch.
Learn more about the AI Operating SystemMulti-model architectures are now practical.When ChatGPT was the only credible option at scale, routing different task types to the right model was theoretical. It is not theoretical now. Gemini and Claude are real options with enterprise tiers, API access, and operational track records. Building a deployment architecture that uses the right model for the right task is a sensible default, not an advanced strategy. This does not mean using every model available — it means knowing which model best suits which task, and building your internal systems so that the routing decision is yours to make, not locked to whichever provider you happened to start with.
The bigger picture
A market where one provider holds less than half the share is a different market from one where a single provider holds most of it. This is true whether the market is cloud computing, search advertising, or AI assistants.
What it means practically: the leverage dynamic between AI providers and their enterprise customers shifts. When ChatGPT was the default, switching costs were high and alternatives were thin. When Gemini is at 27.7% and Claude is at 10.3% and both are growing, enterprise buyers have real alternatives. That changes procurement conversations, contract structures, and the willingness of providers to compete on price and capability.
The businesses that will get the most from AI over the next few years are not necessarily those using the most impressive model. They are the ones who have set up AI in a way they can adjust as things change — without starting over every time the market shifts.
The first-vendor problem is not that ChatGPT is a bad choice. It is that any single-vendor assumption, made by default rather than by design, becomes a liability the moment the market moves. That moment is now.
