Gemini's Real Moat Is Not the Model: Distribution, Defaults, and the Economics of Being Preinstalled

01 The critique and the scoreboard problem

The most-viewed critical take of the fall asks a simple question - what happened to Google Gemini? - and answers with a familiar list: slower launches than rivals, benchmark stumbles, brand confusion after renaming the assistant ecosystem, and a sense that the company with the most AI research per square mile keeps shipping the second-best model.

The list is not wrong on the facts. It is wrong about the metric. If Gemini is judged as a competitive subscription model, it looks like an underperformer. If it is judged as a default being converted to revenue across an installed base measured in billions, the optimization target is different - and the criticism mostly measures the wrong thing.

02 Distribution is the asset benchmarks cannot price

Google ships Gemini where users already are: as the assistant on new Android devices, inside Workspace seats businesses already pay for, and at the top of search results for hundreds of millions of people who never chose an AI app at all. No competitor can buy that placement, at any price, because it is not for sale - it is the ecosystem.

Opt-in competitors must persuade one user at a time, through app stores and word of mouth, at an acquisition cost that compounds. A default pays zero acquisition cost and converts whatever fraction the experience retains. Even a modest conversion rate on enormous exposure rivals every growth-hacking funnel in the industry.

03 Default economics: conversion beats capability

Consumer behavior research and platform history agree on the core pattern: most users never change a default. Browser search, mobile operating systems, and voice assistants all show retention concentrated in whatever shipped on the device. Choice is real but thin; habit does the rest.

That means Google monetizes Gemini through attachment - subscriptions upsold to existing users, API consumption pulled through Workspace and Cloud, advertising relevance improved by assistant context. Every point of attach rate is worth more than a benchmark win, because a benchmark win still has to be marketed to users one at a time, and most are not in the market to switch.

04 Why model quality still matters - differently

Decoupling capability from distribution is not sustainable forever; it is sustainable exactly as long as the default stays good enough. Subscription buyers and enterprise procurement do check benchmarks, and the API market is ruthlessly efficient at routing developers to the best capability-per-dollar. Google cannot be two generations behind and keep the premium tiers.

So the rational strategy is what the 2026 record actually shows: keep the frontier model within striking distance, never necessarily first, while spending the distribution advantage to convert users the frontier labs still have to win over. Parity at the top plus default everywhere is a stronger position than leadership in one race.

05 The TPU wrinkle in the cost curve

The quiet structural advantage is silicon: Google designs its own inference accelerators and serves defaults on fleets it owns rather than rents. At consumer-default scale - billions of short interactions - the difference between amortized owned silicon and market-rate rental compounds into either margin or pricing headroom that rivals cannot match without the same vertical integration.

This is also why the distribution strategy is coherent rather than lazy. Serving a margin-thin default to everyone is only rational if your per-interaction cost curve sits below the industry's. Owning the accelerator layer is what makes the default economics close.

06 The counterforces that could break the moat

Defaults attract regulators: antitrust scrutiny of default placement is already the defining legal risk for Google's search business, and assistant defaults are the same pattern with newer paint. A remedies regime that forces assistant choice screens would convert the moat into an auction anyone can enter.

Users are also drifting multi-assistant - asking one AI on the phone, another in the browser, a third in a work app - which erodes single-default habit faster than any regulation. And enterprise procurement, the highest-value segment, does benchmark due diligence that distribution cannot influence.

07 What to watch

Three numbers will settle the debate better than any hot take. Workspace Gemini attach rates, showing whether business seats actually convert. Android assistant churn, showing whether default users stay when the trial ends. And revenue per assistant user versus capability rank, showing which force - distribution or benchmark position - actually predicts money.

If those numbers favor distribution, the 'what happened to Gemini' genre will age like 'what happened to Bing': a fair observation about the wrong market. If they favor capability, Google has a real problem that no amount of preinstalling can fix. The data arrives quarterly; the model race updates weekly. Confusing the two clocks is how the commentary got here.