At Apple’s developer conference on June 8 ET, the company announced something quiet but consequential: Siri, its 15-year-old voice assistant, will now be powered by Google’s Gemini AI model instead of Apple’s own technology. Apple will pay roughly $1 billion per year for the privilege.
Alphabet’s stock jumped about 120% in the aftermath. Microsoft — which bet heavily on OpenAI — fell around 7%.
That kind of divergence is worth unpacking.
The AI economy currently has two distinct types of players. The first type builds foundation models* — the underlying AI “brains” that process language, images, and reasoning. Google’s Gemini, OpenAI’s GPT series, Anthropic’s Claude. The second type takes those brains and deploys them inside real-world systems: corporations, hospitals, governments, military operations.
Apple just moved from trying to build its own brain to renting Google’s. That signals something important: competing in foundation model development is becoming prohibitively expensive, and the gap between leaders and followers appears to be widening.
For investors, this raises a practical clarifying question. If the model layer is consolidating toward a few winners, what happens to the companies that sit between the models and real-world use?
Companies in the deployment layer — those that specialize in taking whatever AI wins and making it function inside complex, regulated environments — have a different kind of value than the model builders. Their moat* (a durable competitive advantage that’s hard to replicate) doesn’t come from building the brain. It comes from the trust, workflows, and integrations they’ve already built inside government agencies and enterprises. Those don’t disappear just because a better model comes along.
There’s a genuine counter-risk worth tracking, though. If foundation models eventually become capable enough to handle their own deployment and integration — essentially absorbing the middle layer — then deployment infrastructure becomes less valuable. That scenario is not imminent, but it’s the right question to keep asking as models get more capable.
My take: The Apple-Gemini deal is a useful lens for sorting AI investment theses*. “Who builds the brain” and “who makes the brain work inside real operations” are two separate questions — and they don’t behave the same way as investments. Apple just made Google’s answer to the first question more powerful. The second question is still wide open, and for now, harder to replicate.
Disclaimer: This is not investment advice.