Google Stopped Selling Gemini. Now It's Laying Pipe.

Here's the tell: when a company announces the same product for your phone, your TV, your thermostat, your laptop, and your smart home hub in a single week, they're not announcing products anymore. They're announcing a bet on infrastructure — and they need you to install it before you realize that's what's happening.

That was Google I/O 2026. Almost wall-to-wall AI, according to Engadget's morning-after recap, which managed to sound both exhausted and unsurprised at the same time. The word "almost" is doing a lot of work in that sentence. There wasn't much else on stage.

Everywhere Is the Strategy

The specific moves tell the story better than any keynote framing could. Gemini Live on Android got a redesign — Neural Expressive, they're calling it — and with it, access to more first- and third-party Connected Apps, per 9to5Google. Meaning Gemini isn't just answering questions anymore; it's reaching into the apps you already use, threading itself into workflows you didn't design it into.

Simultaneously, Google announced it's making it easier for device manufacturers to build Gemini into smart home products — not a single product, not a partnership, but a broader toolkit so the hardware ecosystem can carry the flag wherever Google's own devices don't reach. Engadget covered this one with a headline that basically said: Google won't rest until Gemini is everywhere in your home. The coverage treated it as slightly aggressive. It is, but it's also rational.

And then there's the Mac app. A platform Google doesn't control, running software from a company that has its own competing AI assistant, getting two new Gemini features — a Spark agent and voice control — arriving this summer, according to 9to5Google. Apple's house, Google's assistant. That one's not subtle.

The Math Underneath

None of this is accidental or scattered. Running frontier AI models at scale costs an extraordinary amount of money in compute. The only business logic that survives that cost is ubiquity — you need the model embedded in enough surfaces that the aggregate usage justifies the infrastructure. A Gemini that lives only in Google Search or only in Pixel phones is a niche product with a datacenter-sized bill. A Gemini that lives in your phone, your TV remote, your thermostat firmware, and your Mac desktop app is something closer to a utility.

This is the move Google is making, and I find it more interesting than the individual feature announcements, which tend to blur together after the third demo. The company isn't trying to convince you Gemini is the best AI assistant — though they'd like that too. They're trying to make switching costs high enough that the question stops mattering. You don't switch away from infrastructure. You complain about it and keep using it.

The cynical read is that this is Google doing what Google does: acquiring surface area first, figuring out depth later. The more generous read is that they've looked at where this industry is going — AI baked into every interface, every device, every ambient interaction — and decided that getting there first, even imperfectly, is worth more than getting there right.

Both reads can be true. Usually are.

What I keep coming back to is the Mac app. That's the one that feels like a statement rather than a roadmap item. You don't build voice control and an agent feature for a competitor's platform unless you've stopped thinking of this as a product competition and started thinking of it as a platform land-grab — one where the platform you're building runs on top of everyone else's.

Google isn't asking to be your favorite. It's asking to be unavoidable.