OpenAI Is the Only AI Giant That Rents Its Distribution
In 2022, Google paid Apple an estimated $20 billion to remain the default search engine on iPhones.1 Google. A company that won search outright, became a verb, and still had to pay rent to the company that owned the screen in front of the user.
Now compare that with the standard story about OpenAI in mid-2026: roughly 900 million weekly users, a giant IPO coming, a hardware project with Jony Ive, a rumored browser, a commerce link with Visa. People see an empire. The structure says something else.
OpenAI is the only AI giant building a global consumer business entirely on land it does not own, while its landlords quietly assemble the models to evict it.
Among the big AI firms, OpenAI is the most exposed: it has no operating system or hardware base of its own. Apple has 2.2 billion active devices.2 Microsoft has Windows. Google has Android and Chrome. OpenAI has an app in someone else’s store and a website inside someone else’s browser. Call it the tenant lab: a frontier model company whose user access runs through surfaces owned by rivals that are also building the models that could replace it. What looks like expansion is also a rush to find a way out.
Why the Benchmark Scoreboard Measures the Wrong Game
The moat is moving away from the model layer and toward distribution. Most benchmark coverage still watches the wrong layer. Frontier models are bunching together on the LMSYS leaderboard, and open-weight systems keep closing the gap with proprietary ones.3 Performance gaps are shrinking while inference costs keep falling.4
Rich Sutton saw the pattern years ago. In The Bitter Lesson, he wrote that "the biggest lesson that can be read from 70 years of AI research is that general methods that use computation are ultimately the most effective."5 Clever tricks fade. Scale keeps winning. For OpenAI, that reads less like theory than a timer counting down the value of proprietary model advantage.
Sam Altman has bet on the other side. "Compute is going to be the currency of the future," he told a Stanford audience. "It will be the most precious commodity in the world."6 Compute is precious, but precious inputs do not create durable margins by themselves. In mobile, Apple captured more value than the suppliers underneath it because Apple owned the device, the OS, and the customer relationship. The same logic applies here. The money goes to whoever owns the last mile.
The Tenant Lab Pays a Distribution Tax, and Its Landlords Are Now Armed
A tenant lab pays for distribution in two ways: with revenue share or with subordination to someone else’s brand. OpenAI has no third option because it owns no surface where users show up by default. Google’s $20 billion payment to Apple is the benchmark for the first kind of tax.1 Google’s traffic acquisition costs hit $50.9 billion in 2023, up from $48.9 billion in 2022 and $45.6 billion in 2021.7 That line tells you what rented distribution looks like in the income statement. The more successful the tenant gets, the more it owes the landlord. The DOJ ruling explained why: "Default status is highly valuable because many users simply stick with the default."1
Today the bill is hidden: OpenAI reaches users through the App Store and web browsers it does not control. Those are rented channels too. If OpenAI wants to move from an app people seek out to a built-in default where people already spend their time, the DOJ case gives a rough starting price. If Google pays $20 billion a year for default search placement, default AI placement will not come cheap. The tax has not disappeared. It is waiting.
Then the setup gets worse. The landlord is building the rival product too. When Google paid Apple for search placement, Apple did not have its own search engine. AI blows up that arrangement. Apple ships Apple Intelligence "integrated deeply into iOS 18, iPadOS 18, and macOS Sequoia," and sends requests to ChatGPT only as a fallback with user permission: "Siri can tap into ChatGPT's expertise when helpful... users are asked before any questions are sent to ChatGPT."8 Apple keeps the personal context and sends out only the generic query. Ben Thompson nailed the result: "Apple has effectively commoditized OpenAI by making them a plug-in. The user relationship remains entirely with Apple."10 Users learn that Siri is the agent and ChatGPT is the helper behind the curtain. Apple can swap it out later.
"Poor distribution, not product, is the number one cause of failure." Peter Thiel, Zero to One
Microsoft is doing the same thing from the enterprise side. Satya Nadella has been blunt: "We are the ones building the products, we are the ones taking the risk, we are the ones who have the customer relationship."11 Copilot is being pushed as "the default interface for how people interact with computing" across Windows.12 And Microsoft is building its hedge in plain view, bringing in DeepMind co-founder Mustafa Suleyman to run a consumer AI group focused on internal model work.13 OpenAI’s biggest distribution partner is preparing for a world where OpenAI is optional.
The Double Vulnerability: Renting Compute and Distribution at Once
Microsoft controls both the supply line and a major demand channel for OpenAI. Azure sits on the cost side. Copilot sits on the distribution side. That gives Microsoft influence over both inputs and access. No other major AI company is exposed in quite that way.
| Player | Owns the OS / device | Owns the model | Owns last-mile distribution |
|---|---|---|---|
| Apple | Yes (2.2B devices) | Building | Yes |
| Microsoft | Yes (Windows) | Building (MAI) | Yes (Copilot) |
| Yes (Android, Chrome) | Yes | Yes | |
| OpenAI | No | Yes | Rented |
Apple, Google, and Microsoft pair distribution with model development. OpenAI and Anthropic still depend on surfaces controlled by others. In smartphones, Apple kept the richest margins because it owned the device, the OS, and the customer relationship. That is the side of the market OpenAI does not control. Sequoia put the broader problem at roughly $600 billion: the gap between AI infrastructure spending and actual end-user revenue.14 That hole does not close unless someone owns the surface that turns AI capability into daily usage. The incumbents already do.
This is not OpenAI's problem alone, and the rest of the field makes the pattern clear. Anthropic rents the same way, with no device, OS, or social app of its own, which makes it the other exposed pure-play. xAI is the counterexample: Grok rides X, the platform Musk folded into xAI, and ships inside Tesla cars, so it reaches users on surfaces it controls. Meta goes further, pairing its open Llama models with Facebook, Instagram, and WhatsApp and their billions of users. The split is clean. The labs with a surface, Google, Microsoft, Meta, and xAI, own the road to the user; the labs without one, OpenAI and Anthropic, rent it. OpenAI is the most exposed of the two, because it is the largest consumer business built entirely on someone else's land.
Reading the Product Blitz as Freehold Panic
The Jony Ive device talks only make sense if OpenAI thinks rented distribution is closing on it. The hardware project matters more than the browser rumor or the commerce tie-up because it is the only move that could give OpenAI a surface it actually owns. The commerce side shows the same trap: OpenAI quietly shut its own Instant Checkout in March 2026, then leased Visa's payment rails that June, renting the transaction layer it could not make work on its own. Freehold panic is what happens when a tenant realizes too late that rented access to users is a dead end. The reported $1 billion raise for a dedicated AI device that "bypasses the traditional smartphone interface" fits that reading exactly.15
Altman has nearly said it out loud. "The smartphone is incredible," he told Bill Gates, "but to fully realize the potential of these agents, we will likely need new hardware paradigms."16 OpenAI needs its own device because iOS will never give it full control over routing or personal context. Without the OS, it stays boxed into general queries while Apple keeps the user data that makes an agent sticky. The hardware bet is not ambition spilling over. It is a land grab by a company that finally understands it does not own the road to its users.
If a rival controls the default entry point and can copy your core feature, treat that channel as temporary, not strategic. Amazon did this to third-party sellers for years, using marketplace demand to spot categories worth entering itself. Platform owners move in once a category proves demand. AI will be no different.
If Apple or Microsoft can answer most user requests in-house by 2027, OpenAI’s app stops looking like a destination and starts looking like a backup. And if OpenAI fails to secure its own surface before Apple, Google, and Microsoft close the loop, 900 million weekly users will look less like a moat and more like traffic flowing through someone else’s tollbooth.
