Airbnb's CEO just named the next platform war — and Apple's the one with the most to lose

Apple's Services business is the closest thing the company has to a recurring software annuity, and the App Store is its engine. In Q3 FY2026 (quarter ended June 27, 2026), Apple reported Services revenue of $30.74 billion, up 12% year-over-year but short of the $31.22 billion the Street expected. Total quarterly revenue was $109.42 billion, up 16.4%. The miss matters more than the beat: it is the first quarter in which management has explicitly cited App Store regulatory changes as a headwind, which is the regulatory front door opening into the agent economy.

In May 2026, Apple confirmed it was working on a framework to allow AI agent apps on the App Store while policing 'rogue' agent behavior. That is the iOS lockdown Chesky is implicitly criticizing: Apple wants agents inside its App Store, governed by its rules, paying its 30% commission, on its terms. If the agent OS emerges from a non-Apple layer — ChatGPT, Foundry, Agentforce — Apple's Services line is exposed to the same disruption that hit the iPhone's share of mobile internet revenue a decade ago.

Microsoft did not get ahead of the agent OS narrative by accident. At Build 2026 in June, the company turned Microsoft Foundry from a model catalog into an agent operating system with a runtime, tool layer, memory, governance, and a distribution channel called Agent 365. The pieces named in Microsoft's own Build blog are explicit: ASSERT (open-source policy evaluation), the Agent Control Specification (a YAML-portable safety contract enforced at five checkpoints — input, LLM, state, tool execution, output), Guided Guardrail Setup, a rubric evaluator, and an Agent ROI module that quantifies business impact in completed tasks and cost efficiency.

Crucially, Agent 365 is the integration surface for OpenAI's Dots, announced two days before Chesky's interview at OpenAI DevDay on September 29, 2026. Dots are 'always-on' agents with their own cloud computer, connecting to more than 4,000 apps, accessed from ChatGPT, Slack, or Microsoft Teams. Specialist Dots inherit Microsoft enterprise governance through Agent 365. The result: Microsoft owns the runtime, OpenAI owns the agent brand, and customers buy the bundle. That is not a partnership; it is the closest the agent era has come to a de facto OS.

Chesky's argument has a quietly obvious implication: the company that already owns identity, permissions, workflow, and audit in a vertical gets the easiest on-ramp to the agent OS in that vertical. Salesforce owns that on-ramp for sales, service, marketing, and commerce. In Q1 FY2027 (quarter ended April 30, 2026), Salesforce reported revenue of $11.13 billion, up 13% year-over-year, and disclosed that Agentforce annual recurring revenue had crossed $1 billion — with combined Agentforce and Data 360 ARR at $3.4 billion, $1.2 billion of that Agentforce alone. That is not a pilot number; that is a product line.

The consumer side of the agent OS race compressed into a six-week window in late September and early October 2026. OpenAI announced Dots at DevDay on September 29, an always-on agent that runs in its own cloud, learns from feedback, and connects to more than 4,000 apps via ChatGPT, Slack, and Microsoft Teams. The Guardian reported the app crossed 3 million downloads within days of launch, and Pro and Business Premium subscribers get a first Dot at no extra cost. The strategic move, as Sarah Perez argued on TechCrunch the same day, is that ChatGPT is becoming the place software is discovered, launched, and used — the role Apple defined for the App Store in 2008.

Google's response is the Gemini Enterprise Agent Platform and Gemini Spark, a 24/7 personal agent inside the Gemini app, unveiled at Google I/O in May 2026. Google has the data layer (Gmail, Calendar, Maps, Search history) and the Android base, but the platform it controls is shrinking as Apple Intelligence on iOS 27 keeps default agents inside Apple's perimeter. Amazon is repositioning Alexa+ as the commerce surface for agentic shopping, an explicit acknowledgment that the agent OS will be the next checkout — but Amazon's credibility took a hit when it killed the Rufus chatbot to consolidate on Alexa, a signal that consumer agent UX is still unproven.

Underneath every agent OS is a compute layer dominated by one supplier. NVIDIA reported trailing-twelve-month revenue of $303.0 billion, up 105.9% year-over-year, with quarterly revenue growth of 105.9% in the most recent quarter. The gross margin is 74.7% TTM; operating margin 65.2%. That is the financial signature of a company selling shovels in a gold rush where every gold-rush participant is still buying shovels. Whether the winning agent OS is Apple's walled garden, Microsoft's enterprise stack, Salesforce's vertical layer, or OpenAI's consumer surface, every inference that matters runs on NVIDIA hardware.

Chesky is not a neutral observer. At Skift Global Forum on September 23, 2026, he committed to launching an Airbnb-branded AI agent in 2027 — 'much richer than a chatbot,' as he put it. The pitch is that Airbnb's 200 million verified identities, its $100 billion-plus payment history, and the 90% message rate between guest and host are the substrate an agent needs, and the substrate the existing OSes do not provide. He is also asking every other travel platform a question: who is your identity provider, who authorizes the payment, and who audits the booking if the agent gets it wrong?

Airbnb's Q2 FY2026 revenue was $3.608 billion, up roughly 17% year-over-year, on a TTM base of $13.18 billion. Operating income margin in the quarter was 21.0%. The company is profitable, cash-generative (capex is essentially zero against operating cash flow), and trading at a forward P/E in the mid-twenties. The bull case for the stock is that the agent OS shift lets Airbnb capture a higher share of travel dollars per user by owning the concierge layer, not just the listing. The bear case is that any agent OS that wins will route around Airbnb's brand the way Google Search routed around Yahoo's directory.

Short-term (next 12 months): four moves will signal who is winning. (1) Does Microsoft monetize Agent 365 as a per-seat license, or fold it into existing E5 deals? (2) Does Salesforce grow Agentforce ARR past $3 billion by Q4 FY27? (3) Does Apple ship a credible agent channel on the App Store, or block most agents to preserve its commission? (4) Does OpenAI convert Dots downloads into paid Pro and Business Premium seats at the rate ChatGPT Plus did in 2023? Each of those four print as a revenue line by Q4 FY27.

Long-term (one to three years): the structural question is whether the agent OS emerges as a thin cross-platform layer (most likely owned by OpenAI and Microsoft together), as a vertical product layer (Salesforce in CRM, Airbnb in travel, ServiceNow in IT), or as a phone-resident layer (Apple Intelligence on iOS, Gemini on Android). The vertical-product outcome is the most boring and the most likely. The cross-platform outcome is the most ambitious and the most valuable. The phone-resident outcome is the one Chesky is publicly betting against — because in that world, Apple and Google own the agent OS the way they own the mobile one, and Chesky's Airbnb argument becomes a footnote.