Google Is Running the Android Playbook on AI and It Could Break the Token Market
Fifteen years ago, Google made a decision that reshaped mobile computing: give the operating system away for free, open-source most of it, and make the money somewhere else.
Android didn’t need to be a profit center. It needed to be a distribution funnel into Search, Gmail, Maps, the Play Store, and Google’s ad network, the businesses that actually printed money.
Now watch what’s happening in AI. Gemini is free for consumers. Gemma, Google’s open-weight model family, just moved to a fully permissive Apache 2.0 license with Gemma 4. That’s not a side project, that’s the same playbook, pointed at a new stack.
Commoditise the thing everyone else is trying to sell
OpenAI, Anthropic, and most of the frontier lab ecosystem are - at least in part - selling intelligence itself. Tokens are the product. Usage-based pricing, subscription tiers, and API margins are the business model.
Google doesn’t need that model to work. If Gemini is free for consumers and Gemma is free for developers to download, fine-tune, and self-host, the market price of “a good enough model” trends toward zero. That’s a brutal environment for anyone whose revenue depends on people paying per token. It’s a fine environment for a company whose revenue depends on something else entirely.
This is the classic “commodities your complement” move: when a product threatens to become a bottleneck that squeezes your real business, you flood the market with a free or open version of it until it stops being scarce — and therefore stops being able to charge rent.
The hooks stay closed
Here’s the part that matters most, and the part that mirrors Android almost exactly. Android itself was open source (AOSP), but Google Mobile Services — the apps and APIs that make an Android phone actually useful, like the Play Store, Gmail, and Google Play Services, stayed proprietary and tightly bundled. Anyone could fork Android. Almost no one could fork their way to Google’s distribution and data.
Expect the same split in AI:
• Open and free: the model weights, the chat assistant, the raw reasoning capability
• Closed and monetised: the integrations into YouTube, Gmail, Search, Android (as an OS-level layer), the Play Store, the ad network, and Google Workspace
The model becomes the free bait. The hooks into a decade of consumer surface area - inboxes, search history, video habits, app purchases, calendars, documents - are where the actual leverage and monetisation live.
A frontier model with no distribution is a commodity. A mediocre model wired into four billion Android devices, the world’s dominant search engine, and the ad infrastructure that monetizes attention is a moat.
Why this is a hard problem for everyone else
Labs without a consumer ecosystem to subsidise the model have one real lever: charge for intelligence. If Google keeps pushing the baseline price of “good enough intelligence” toward zero, that lever gets weaker every quarter. It doesn’t mean frontier capability stops mattering — someone will always pay for the genuine edge, the same way enterprises still pay for the best cloud infrastructure even though plenty of it is open source. But it does mean the middle of the market, the broad base of consumer and prosumer usage, gets squeezed hard, and the survivors are the ones with either a real capability edge or a distribution channel of their own.
The obvious counterpoints
This strategy isn’t a free lunch, and it’s worth naming the risks:
• Regulatory exposure. Bundling AI hooks into Search, Android, and Workspace is exactly the kind of leverage that invited antitrust scrutiny for Android in the first place. Doing it again, in a hotter regulatory climate, is not free.
• Open models are a double-edged sword. Once Gemma is genuinely Apache 2.0, competitors and open-source developers can build on it too, including building products that compete with Google’s own services.
• “Free” still needs to be funded. Serving inference to billions of users isn’t free to Google; it’s a bet that the downstream monetisation (ads, subscriptions, cloud) outpaces the compute bill.
• Frontier capability isn’t fully commoditised yet. The very top of the capability curve, the hardest reasoning, coding, and agentic tasks, still commands a premium. Google undercutting the middle of the market doesn’t automatically win the frontier.
The bet
Android didn’t win because it was the best mobile OS in a vacuum. It won because Google didn’t need the OS to make money — it needed the OS to make everything else Google owns more valuable, and it was willing to give away the commodity layer to get there.
The same bet is now on the table for AI models. If it works, the token market as a standalone business gets a lot harder to defend
and the real fight moves to who owns the hooks.