Blog

There is a war being fought for the future of artificial intelligence, and the loudest combatants — OpenAI, Google, Anthropic, Meta — are spending billions training ever-larger models, racing to claim the title of the world’s most capable AI. Apple is not in that race. It doesn’t need to be. While its rivals burn cash on compute clusters and research talent, Cupertino is doing something far more lucrative: building the tollbooth every one of them may eventually have to pay to pass through.

The strategic logic is elegant, if quietly audacious. Apple controls the most valuable consumer computing real estate on the planet — over 2.2 billion active devices, as disclosed by CFO Luca Maestri on the company’s Q1 2024 earnings call. That installed base, bound together by iCloud, iMessage, the App Store, and an ecosystem designed to make leaving feel expensive, gives Apple something no AI lab has: a direct, trusted, and deeply habitual relationship with mainstream consumers. The question Apple appears to be answering is not „how do we build the best AI?“ but rather „how do we make sure every AI has to come through us?“

The Orchestrator Play

The clearest public signal of Apple’s strategy arrived at WWDC 2024, when the company unveiled Apple Intelligence — its suite of on-device AI features for iPhone, iPad, and Mac. The product itself was notable less for its raw capabilities than for its architecture. Apple Intelligence is designed to handle simpler tasks locally and, crucially, to decide when to route more complex queries to third-party AI models. In Apple’s own language from its press release: „Siri can tap into ChatGPT when it might be helpful for the user.“

Read that sentence again slowly. Apple — not the user, not OpenAI — is the traffic cop. It decides which AI gets the request, which model gets access to the context, and which provider gets the chance to turn that interaction into a paying customer. That is not a feature. That is a structural market position, and it is the foundation of what may become the most consequential business arrangement in the AI industry.

The first, and so far most visible, implementation of this model is the integration of OpenAI’s ChatGPT into iOS 18, iPadOS 18, and macOS Sequoia. Users can invoke ChatGPT through Siri for complex queries or use it within the system-wide writing tools. Free users can access it without logging in; ChatGPT Plus subscribers can link their accounts for expanded capabilities.

A Deal That Tells You Everything

The financial terms of the Apple-OpenAI arrangement are deliberately opaque, but what has been reported is instructive. According to Bloomberg’s Mark Gurman, Apple is not paying OpenAI for the integration — at least not initially. Instead, OpenAI is receiving something arguably more valuable in the short term: distribution across hundreds of millions of premium devices, delivered to users at the exact moment they need a capable AI. Apple, in turn, is positioned to collect a share of any monetization that flows through its platform — including, potentially, subscriptions to ChatGPT Plus — consistent with its established App Store economics.

This is not a partnership of equals. It is a landlord-tenant arrangement dressed in the language of collaboration. OpenAI gets a shop on the most expensive high street in technology. Apple gets a percentage of everything sold inside it, without having spent a dollar on the inventory.

The precedent is not hard to find. For years, Google has paid Apple an estimated $15 to $20 billion annually to remain the default search engine in Safari, a figure that emerged from testimony in the U.S. Department of Justice’s antitrust case against Google. That arrangement — a dominant platform extracting billions from a dominant service provider simply for default placement — is precisely the template now being laid over the AI industry. The „search tax“ may be about to become an „intelligence tax.“

The App Store as Blueprint

To understand where this is heading, it helps to understand where it has already been. Apple’s App Store generates its leverage through a simple, well-tested mechanism: if you want to sell digital goods or services to iPhone users, Apple takes its cut. Historically, that cut has been 30 percent on digital purchases and subscriptions in the first year, dropping to 15 percent thereafter — and 15 percent for developers in its Small Business Program. The result: Apple’s services segment, which includes App Store commissions, reached $85.2 billion in revenue for fiscal year 2023, according to its annual 10-K filing, up from $78.1 billion the prior year.

Regulators have taken notice. The European Commission’s Digital Markets Act formally designated Apple as a „gatekeeper“ in September 2023, forcing changes to App Store rules and default settings within the EU. Epic Games has fought Apple in court over the same dynamics. And yet the model has proven remarkably durable — because the underlying reality has not changed. Developers may resent the commission, but the alternative — forgoing access to Apple’s user base — is commercially worse. That calculus, applied to AI, is why the AI tax may be even harder to resist than the App Store tax ever was.

Why AI Companies May Have No Choice

The AI industry has a customer acquisition problem. Training a frontier model costs hundreds of millions of dollars. But getting that model in front of mainstream consumers — not tech enthusiasts, not developers, but ordinary people who will pay $20 a month for a subscription — is a different and equally difficult challenge. Apple has already solved that problem. It has over one billion paid subscriptions flowing through its services ecosystem, as Tim Cook disclosed on the company’s Q3 2023 earnings call, and a user base that trusts its hardware, its privacy narrative, and its defaults.

For an AI company trying to convert free users into paying subscribers, Apple’s distribution is not merely attractive — it may be close to essential. The consumer who encounters ChatGPT through a seamless Siri handoff, on a device they already trust, in a workflow they already use, is a far warmer prospect than one reached through a paid ad or a cold app-store search. Apple knows this. And if the OpenAI deal is the template, the terms of access will only become more explicit — and more expensive — as the AI market matures and the value of that distribution becomes impossible to ignore.

The Privacy Angle as Competitive Moat

Apple’s leverage is not purely structural — it is also reputational. The company has spent years positioning itself as the privacy-respecting alternative in a data-hungry industry, and that positioning now serves as a key selling point for its AI orchestration model. Apple Intelligence’s „Private Cloud Compute“ architecture, which processes sensitive requests on Apple-controlled servers without storing user data, gives consumers a reason to trust AI features they might otherwise be wary of. For AI providers, it offers a form of legitimacy-by-association: your model, delivered through Apple’s trusted infrastructure.

But that trust also deepens the lock-in. Users who engage with AI through Apple’s system — through Siri, through system writing tools, through seamlessly integrated third-party models — are engaging with Apple’s interface, Apple’s permissions framework, and Apple’s data relationships. The underlying model may be OpenAI’s or Anthropic’s or Google’s. The relationship, as far as the user is concerned, is with Apple.

The Silent Tax, Compounding

None of this requires Apple to build a better chatbot than OpenAI, a more capable model than Gemini, or a more creative system than Anthropic’s Claude. It requires only that Apple remain what it already is: the preferred device of the world’s highest-spending consumers, the gatekeeper of the interfaces they use most, and the platform through which digital goods and services are increasingly bought and sold.

The „AI tax“ is silent precisely because it does not look like a tax. It looks like a partnership, an integration, a feature. But the economic logic is the same as it has always been in platform capitalism: he who controls the pipe does not need to own the water. As AI becomes the defining software layer of the next decade, Apple’s strategic bet is that every model — however powerful, however well-funded — will eventually need to flow through its pipes. And in Cupertino, pipes come with a toll.

There is a war being fought for the future of artificial intelligence, and the most strategically positioned combatant isn’t the one firing the most shots. While Microsoft, Google, Meta, and a constellation of well-funded startups race to build ever-larger language models — spending tens of billions on data centers, GPUs, and compute infrastructure — Apple has been doing something quieter, and potentially more lucrative: paving the road everyone else will have to drive on, and reserving the right to charge a toll.

The thesis is simple, even if its execution is anything but. Apple doesn’t need to win the AI model race. It needs to own the distribution layer — the screen, the voice interface, the default search bar, the app marketplace — through which AI reaches the most valuable consumer segment on earth. If that sounds familiar, it should. Apple has already done it once, with search. The question now is whether it can do it again, at scale, with AI.

The Capital-Light Strategy in a Capital-Hungry Race

To understand Apple’s position, you first have to understand what it is choosing not to do. On Apple’s Q1 2024 earnings call, Tim Cook described AI as „a major focus“ for the company, but pointedly framed it as something embedded in products rather than a standalone moonshot. That measured language was not accidental. While Cook was speaking, his peers were making very different kinds of statements.

Microsoft’s CFO guided investors toward sharply rising AI infrastructure capital expenditure to support Azure and its partnership with OpenAI. Alphabet flagged „elevated levels of investment in AI compute“ across its data centers and custom silicon. Mark Zuckerberg publicly committed to spending tens of billions on AI infrastructure at Meta. Meanwhile, OpenAI’s relationship with Microsoft involves multi-billion-dollar compute commitments, and Anthropic’s strategic collaboration with Amazon is explicitly tied to cloud usage at scale.

Apple, by contrast, has leaned aggressively into on-device AI processing — a strategy documented across its official machine learning research publications and developer documentation — emphasizing its Neural Engine silicon and the Core ML framework. The narrative Apple is selling is privacy and latency. The business logic underneath it is margin preservation and control. On-device processing means Apple doesn’t need to pay for cloud inference. It means user data stays on the device, away from regulators and rivals. And critically, it means Apple retains the orchestration layer: the piece of software that decides which AI model gets called, when, and under what commercial arrangement.

The Toll Booth That Already Exists

Before examining what Apple’s AI toll might look like, it is worth lingering on the one that already exists — because it is hiding in plain sight, and it is already, in a meaningful sense, an AI tax.

The U.S. Department of Justice’s antitrust case against Google revealed that Google pays Apple an estimated $18 to $20 billion annually to remain the default search engine on Safari and across Apple devices, according to reporting by The New York Times based on trial testimony. Apple contributes almost nothing to the search product itself. It simply controls the on-ramp — the default search bar through which hundreds of millions of high-income users enter their queries — and charges Google handsomely for the privilege.

Now consider what Google’s search product is becoming. With the rollout of its Search Generative Experience and Gemini integration, Google Search is increasingly an AI product. The queries flowing through that Apple-controlled default bar are increasingly being answered by a large language model. Which means that, functionally, Apple is already collecting a toll on AI-generated answers — it simply hasn’t renegotiated the contract to reflect that reality. Yet.

When that renegotiation comes — and it is a matter of when, not if — the payment flowing from an AI-native search product to Apple’s balance sheet will look very different from a deal struck in the era of blue links. Apple knows this. So does Google.

Siri as the Silent Gatekeeper

The Google search deal is the clearest existing precedent, but Apple’s deeper leverage lies in Siri and the system-level frameworks that mediate how users interact with their devices. Siri is not merely a voice assistant; it is a privileged operating-system process with access to contacts, messages, calendar data, device settings, and — increasingly — third-party app actions via Apple’s SiriKit and Intents frameworks. No third-party AI assistant, however capable, is granted that level of system access on an iPhone without Apple’s explicit permission.

This creates an architectural advantage that no amount of model capability can easily overcome. A user might prefer the quality of a third-party AI’s answers, but if that AI cannot set a timer, send a message, control smart home devices, or access on-device context without routing through Apple’s own layer, its utility is structurally limited. Apple, in other words, controls the pipes. And companies that want to be plumbed into those pipes will increasingly find that access has a price.

Apple’s own developer documentation makes clear that the Intents and App Extensions system is how third-party apps participate in system-level AI features. The company that writes the rules for that participation also sets the commercial terms — or reserves the right to do so.

The App Store Blueprint

If you want to understand how Apple would structure an AI toll, you need look no further than the App Store, which has served as a remarkably durable proof of concept for the entire model. Apple charges a commission of up to 30% on digital goods and services sold through iOS apps, with a reduced 15% rate for small developers and for subscriptions after their first year, as detailed in Apple’s own developer program documentation. That toll has survived years of regulatory scrutiny, developer rebellion, and high-profile litigation — emerging battered but fundamentally intact.

The scale of the economy Apple taxes is enormous. In May 2023, Apple announced that the App Store facilitated over $1.1 trillion in developer billings and sales in 2022. The majority of that commerce is not subject to Apple’s commission — physical goods, for instance, are excluded — but the figure illustrates the gravitational pull of Apple’s platform. AI-powered subscription services, chatbot apps, AI productivity tools, and AI-assisted in-app purchases sold through the App Store would all, under current policy, be subject to Apple’s standard commission structure. That is not a hypothetical future tax. It is a tax that is already being collected today, on every ChatGPT Plus subscription renewed through an iOS app, on every Perplexity Pro plan purchased on an iPhone.

The question for the next phase of the AI economy is not whether Apple will levy a toll — it already does — but how comprehensively and creatively it can expand that toll as AI becomes the dominant mode of digital interaction.

Financial Stakes: Services, Margins, and the Upgrade Flywheel

Apple’s services segment — which encompasses the App Store, Apple Music, iCloud, Apple TV+, and a growing portfolio of subscription and licensing revenue — has become the company’s highest-margin business line and an increasingly important driver of its overall valuation. The strategic logic of the AI toll is not merely additive to that business; it is potentially transformative.

Consider the compounding effect of three revenue streams converging simultaneously. First, licensing and default-placement deals with AI model providers — the evolved form of the Google search arrangement — represent recurring, high-margin income that requires no incremental capital investment from Apple. Second, App Store commissions on AI-powered subscriptions and in-app purchases grow automatically as AI apps proliferate and monetize. Third, the premium AI features embedded in new iPhone and Mac hardware — made possible by the Neural Engine — provide a compelling consumer upgrade narrative, sustaining the hardware refresh cycle that underpins Apple’s unit economics.

Apple’s installed base of over 2.2 billion active devices, disclosed by Tim Cook on the Q1 2024 earnings call, is the foundation on which all three revenue streams rest. No AI model provider — not OpenAI, not Google, not Anthropic — commands anything close to that direct consumer relationship. They need Apple’s distribution. Apple, as a result, has pricing power it has barely begun to exercise.

The Regulatory Shadow

No analysis of Apple’s toll booth strategy is complete without a serious reckoning with the regulatory environment, which represents the most significant structural risk to the entire thesis. The App Store commission model has already attracted antitrust scrutiny across multiple jurisdictions. The EU’s Digital Markets Act has compelled Apple to allow alternative app marketplaces and payment systems in Europe — a meaningful concession that, if replicated globally, would erode the commission model’s reach.

The Google search default deal faces its own legal jeopardy. The U.S. v. Google antitrust case — centered on exactly the kind of default-placement payments that underpin Apple’s search toll — could result in remedies that restructure or prohibit those arrangements entirely. A ruling against Google’s default search payments would not eliminate Apple’s ability to negotiate AI distribution deals, but it would constrain the form those deals can take and invite immediate regulatory scrutiny of any successor arrangement.

Apple is not naive about this risk. Its public emphasis on privacy, on-device processing, and user choice is partly genuine — the company has made real architectural commitments to on-device AI — and partly a regulatory positioning strategy. A company that can credibly argue it is distributing AI to protect user privacy rather than to extract monopoly rents is in a meaningfully stronger position before competition regulators than one that cannot.

The Competitors‘ Dilemma

For the AI model providers, Apple’s toll booth presents a dilemma with no clean resolution. Refusing to participate in Apple’s ecosystem means forgoing access to over two billion devices and the most affluent consumer demographic in the world. Accepting Apple’s terms means ceding margin, data, and strategic leverage to a platform owner whose long-term interests may not align with their own.

Google is perhaps best positioned to navigate this tension, because it already has a negotiated relationship with Apple and because its own device ecosystem — Android, Pixel — provides an alternative distribution channel. But Google’s dependence on Apple’s Safari default for a significant portion of its search revenue illustrates just how powerful Apple’s position is, even for the world’s dominant search company. For smaller AI players without Google’s scale or negotiating leverage, Apple’s terms will be substantially less favorable.

The most intriguing wildcard is what Apple does with its own first-party AI ambitions. The company has been building foundational model capabilities internally, and its on-device AI investments — in the Neural Engine, in Apple Silicon, in on-device language models — suggest it is not content to be merely a passive toll collector forever. At some point, Apple’s own AI products may compete directly with the third-party models it currently routes user queries toward. When that happens, the neutrality of the toll booth becomes very difficult to sustain — and regulators will take notice.

The Quiet Accumulation of Power

What makes Apple’s AI strategy so strategically elegant — and so difficult to counter — is that it does not require Apple to win any single battle in the AI arms race. It does not need the best model, the most parameters, or the flashiest demo. It needs only to remain the trusted, default interface through which hundreds of millions of consumers encounter AI for the first time, every day.

That position is built on a decade of hardware investment, ecosystem construction, and brand equity that cannot be replicated quickly. Apple’s competitors can spend more on compute. They cannot easily replicate 2.2 billion loyal devices, a developer ecosystem trained to pay Apple’s toll, and a consumer brand associated with privacy and trust at a moment when AI’s relationship with user data is deeply contested.

The frontier model race is loud, expensive, and genuinely uncertain in its outcome. Apple’s strategy is quiet, asset-light, and increasingly legible. While the rest of the industry debates who will build the most powerful AI, Apple is focused on a different question entirely: who controls the door through which AI reaches the people who matter most. In Cupertino, they believe they already know the answer — and they are quietly setting the price of admission.

In the summer of 2024, Apple stood on a stage at its Worldwide Developers Conference and announced something that, on the surface, looked like a concession. The company that had long insisted on controlling every layer of its ecosystem — hardware, software, services — was inviting OpenAI’s ChatGPT directly into its operating system. To some, it looked like Apple admitting it had fallen behind in the AI race. To others, it looked like something else entirely: a toll booth opening for business.

Apple doesn’t need to win the AI model war. It needs to own the road.

The Most Valuable Audience in Technology

Any serious analysis of Apple’s AI strategy has to begin not with algorithms or data centers, but with people. According to Apple CFO Luca Maestri on the company’s Q1 2024 earnings call, Apple’s active installed base had surpassed 2.2 billion devices globally — a figure that includes more than 1.2 billion active iPhones, a number management has described as sitting at an „all-time high.“ These aren’t passive users. Research consistently shows that iPhone users skew toward higher income brackets and higher average revenue per user compared to their Android counterparts, a demographic reality that Apple itself underscores by pointing to record services revenue per user quarter after quarter.

This is the substrate of what might be called Apple’s „silent AI tax.“ The company doesn’t need to charge users a visible fee for artificial intelligence. It simply needs to sit between AI providers and the most commercially attractive digital audience on the planet — and let the economics do the rest.

A Proven Playbook: The Services Precedent

Apple has done this before. Its Services business — encompassing the App Store, Apple Pay, iCloud, Apple TV+, and a constellation of subscriptions — generated a record $23.1 billion in revenue in Q1 2024 alone, up from $20.8 billion in the same quarter a year earlier, according to Apple’s earnings release and Form 10-Q for the quarter ended December 30, 2023. Services now represent roughly one-fifth to one-quarter of Apple’s total revenue depending on the quarter, and carry significantly higher margins than hardware, a mix shift Apple’s management has openly credited for boosting overall gross margins.

The business logic is elegant and well-rehearsed: Apple creates the platform, sets the rules, and takes a cut. For most of its App Store history, that cut has been 30 percent on paid apps and in-app purchases, reduced to 15 percent for small developers and for subscriptions after the first year. Apple doesn’t need to make the apps. It simply needs to control who gets to distribute them — and on what terms.

AI is shaping up to follow the same structural logic, with the potential for even greater leverage. Unlike individual apps, AI is increasingly woven into the operating system itself, which means the „distribution“ Apple controls is not just an app store shelf — it’s the interface layer through which users experience intelligence itself.

The Google Deal: A Window Into What „Default“ Is Worth

Perhaps the clearest window into what Apple’s AI gateway position could eventually be worth is the company’s arrangement with Google Search — a deal that became a focal point of the U.S. Department of Justice’s antitrust case against Google. Court proceedings revealed that Google pays Apple an estimated $18 to $20 billion or more annually simply to remain the default search engine on Safari. That figure — staggering in its own right — is payment not for a product, not for advertising, not for content, but purely for placement. For the privilege of being the default answer when an Apple user wants to find something.

If a search engine is worth that kind of annual tribute for default status on Apple devices, the question worth asking is: what is a default AI model worth? The answer, as AI becomes the primary interface through which people access information, complete tasks, and make decisions, could dwarf the Google search arrangement entirely.

Apple Intelligence: Orchestrator, Not Competitor

Apple’s 2024 AI announcement made the strategic architecture explicit. Unveiled at WWDC and detailed in an Apple Newsroom release on June 10, 2024, „Apple Intelligence“ is not a single AI model competing with GPT-4 or Gemini. It is a layered system: on-device models running on Apple Silicon handle lighter tasks like summarization and rewriting; a „Private Cloud Compute“ infrastructure routes more intensive queries to Apple-controlled servers; and for tasks that exceed even that, Siri can now hand off to external models — starting with ChatGPT, with Apple explicitly signaling that additional AI providers could follow.

This architecture is not an admission of weakness. It is a deliberate choice to occupy the orchestration layer rather than compete in the expensive, capital-intensive frontier model race being run by OpenAI, Google, and Anthropic. Apple is not trying to out-spend those companies on GPU clusters and training runs. It is positioning itself as the system that decides which model a user reaches, when, and under what conditions — with user consent flowing through Apple’s own privacy framework and data remaining, as far as possible, within Apple’s controlled environment.

As Seeking Alpha’s analysis of Apple’s AI positioning has noted, this approach allows Apple to monetize AI through services and ecosystem engagement rather than hardware alone — deepening consumer lock-in while avoiding the balance sheet risks of frontier model development.

The Regulatory Shadow Over the Toll Booth

Apple’s ability to extract economic value from platform control has not gone uncontested. The company’s App Store model has faced sustained legal and regulatory challenges that offer a cautionary counterpoint to any assumption that the AI toll booth will operate without friction.

In the U.S., Epic Games‘ multi-year antitrust battle against Apple largely upheld the company’s right to control iOS distribution, though the district court required Apple to allow developers to direct users toward alternative payment methods — the so-called anti-steering provision. The Supreme Court declined to hear appeals in early 2024, leaving most of Apple’s core model intact. In Europe, however, the picture is more complicated. The EU’s Digital Markets Act, which formally designated Apple as a „gatekeeper“ in September 2023, requires the company to allow alternative app stores and sideloading in EU markets. By March 2024, the European Commission had opened non-compliance investigations into Apple’s response to those requirements.

The DMA’s implications for AI are still taking shape, but the direction of regulatory travel is clear: European authorities are explicitly suspicious of gatekeeper dynamics, and an AI distribution model that replicates the App Store’s economics will face scrutiny. Apple’s response — as it has been with every previous regulatory challenge — will likely involve the minimum structural concession necessary to satisfy legal requirements while preserving as much economic leverage as possible.

The Hardware Upgrade Cycle, Quietly Reloaded

There is a dimension to Apple’s AI strategy that receives less attention than the services angle but may prove equally significant: hardware. Many of Apple’s most capable Apple Intelligence features require Apple Silicon of a certain generation — specifically, devices with an A17 Pro chip or later, which means, in practical terms, an iPhone 15 Pro or newer. On-device AI processing at the level Apple has architected demands memory bandwidth and neural engine performance that older hardware simply cannot provide.

This creates a second, less visible revenue mechanism. Users who want the full Apple Intelligence experience are nudged — not forced, but nudged — toward upgrading their devices. In a smartphone market where upgrade cycles have been lengthening for years as hardware improvements became more incremental, AI represents a genuinely compelling reason to replace a device that still works perfectly well. Apple doesn’t need to advertise this. The feature gates speak for themselves.

A Tax by Any Other Name

No Apple user will ever see a line item on their bill reading „AI gateway fee.“ That is precisely what makes the model so durable. The toll Apple collects on the AI era will be embedded in App Store commissions on AI-powered applications, in the implicit value exchanged when AI providers pay — in cash, in favorable terms, or in strategic concessions — for privileged access to Apple’s ecosystem, in hardware upgrade revenue driven by AI feature requirements, and in the deepened services engagement that AI-powered Siri and system features are designed to generate.

What Apple has understood, perhaps better than any other company in the industry, is that the most profitable position in a platform economy is rarely the one building the most impressive technology. It is the one controlling access to the most valuable users. With 2.2 billion active devices, margins that are the envy of the industry, and a track record of converting platform control into recurring, high-margin revenue, Apple has quietly positioned itself at the most important chokepoint in artificial intelligence — not by racing to the frontier, but by owning the gate in front of it.

The AI companies spending billions training ever-larger models may be building the engines. Apple is building the road — and collecting the tolls.