Was Google a victim of the innovator's dilemma?

When ChatGPT launched, Google was a textbook victim of the innovator's dilemma. Gemini 4 Argon is the evidence that it escaped, and the escape route was a core business that kept growing, plus chips no rival makes.

What happened. Google announced Gemini 4 Argon on 30 September and made it generally available on 13 October, through the paid Gemini API, AI Studio, and the Gemini app for AI Ultra and Pro subscribers. On Artificial Analysis's index it matches OpenAI's GPT-6 Astra, putting Google back among the top three labs. At introductory pricing it costs about 60 percent of Astra per task; at full price, slightly more.

The dilemma, element by element. Clayton Christensen argued that good incumbents lose to disruption by doing the rational thing. His disruptors start at the low end of a market, so this is an analogy, but each part fits Google:

• Listen to your best customers. Search users and advertisers wanted accurate links. In December 2022 Google executives cited reputational risk for holding back their own chatbot.
• The new technology starts worse. Early chatbots made things up. OpenAI, with less reputational risk, shipped anyway.
• It threatens the margin. An answer that resolves the question replaces the page of ads that funds Google.
• The new market looks too small. In 2022 chat earned nothing next to search.
• The cure is a separate unit. Google already had one in DeepMind but kept it in research. Only in 2023, after the code red, did it merge Brain and DeepMind to build products.

Where the pattern breaks. Christensen's victims lose the core business. Google did not. Search grew 17 percent in the June 2026 quarter, and Google Cloud grew 82 percent. Cloud is the infrastructure the new technology runs on. Google also designs its own TPU chips, and even Anthropic has committed to use up to a million of them. No other frontier lab builds both the models and the chips.

Google also kept far more developers than the "fallen behind" story suggests, mostly through Flash, its cheap and fast model line. Sundar Pichai calls Flash "the sweet spot of performance and cost". On OpenRouter, the marketplace developers use to switch between models, Google has been in the weekly top three for most of the year. That is the smaller part of its traffic. Google's own APIs process about 22 billion tokens a minute, roughly 220 trillion a week, which is more than OpenRouter handles in total. Flash was built for people doing work, not consumers chatting.

What it means for business leaders. Three frontier labs is good for competition. Make your systems model provider agnostic, so each task can go to the model that does it best.

Do this Monday: ask what it would take to switch from your current model provider to a new one.

More on how we think: qubit's beliefs.