Then he revealed the punchline. The three products were all the same device. Apple, Jobs said that day, was reinventing the phone.

Nearly two decades later nobody uses the iPod, although there are probably still some stuffed in the corner of a bedroom drawer. And that “internet communicator,” the thing the audience didn’t quite understand when Jobs first described it, was the “one more thing” that made history and built a $4 trillion Apple franchise.

The iPhone became the operating system that supports the App Store, the Apple Pay mobile wallet, streaming video and music, and a trillion-dollar services economy that connects people with the internet. The Apple apps tail wagged the iPhone dog so hard that the dog forgot it was ever a dog. Well, except four times a year when earnings day rolls around.

That same phenomenon is happening again. Right this very minute. Except this time, the tail is AI and agents. And the dog is still the smartphone. But which one?

The smartphone, first the iPhone and later Android handsets, were envisioned as the device that runs AI apps. Instead, AI is becoming the reason for the device. The companies that understand this are building phones, forging $50 billion alliances to explore that, and assembling the commerce ecosystems that will determine who captures the economic value of the most important technology shift since the smartphone itself.

AI and the agentic economy.

Read more: Why 30 Million US Consumers No Longer Search

That makes the question less about which model is smartest and more about who controls the device and the operating system in the consumer’s hand. Because that’s the device that determines the AI defaults, the distribution, and the revenue.

It’s the tail wagging the dog story once again.

The Data Story: What People Actually Do

Every month, PYMNTS Intelligence publishes our Consumer AI Adoption Benchmark. The latest was fielded in late March 2026 across a national sample of 2,111 U.S. adults and tracks 54 distinct personal AI tasks across nine categories. So far, we’ve captured the trendlines for nearly 20,000 consumers that statistically reflect the composition of the U.S. adult population. It is, as far as I know, the most granular longitudinal dataset on how real people — not the developers, not the academics, not marketing agencies and tech journalists — are using AI in their daily lives.

And because we track it monthly, we see trendlines in real time. Not in quarterly snapshots that smooth out the story, or random single data points, but in the month-over-month shifts that reveal what’s actually driving behavior.

And the picture it paints is unambiguous.

The top tasks the consumers ask frontier models to perform are not exotic or even one-offs. Editing personal writing. Drafting texts and emails. Finding product links. Looking up symptoms. Planning meals. Learning new skills. These are activities they increasingly do on their mobile phones. They happen on the couch, in line at the grocery store, on the commute home. They happen throughout the day, in between everything else, and almost always on their smartphone.

The long-term trendlines are even more fascinating.

The fastest-growing AI tasks over the past six months aren’t the flashy research and content creation functions that powered the initial adoption wave. They’re household logistics. Finding discount codes grew three percentage points. So did managing household logistics and meal and grocery planning. Reminders for bills and appointments grew a little more than two percentage points.

AI is becoming a consumer utility. And consumer utilities live on smartphones.

Read More: Gen AI: The Technology That Broke the Adoption Curve

What also makes our monthly tracking so valuable is that we can see how the world outside of AI usage shows up inside people’s AI usage.

In March, we saw travel planning tasks had pulled back. So did hotel and restaurant searches. And that tracks precisely with what we’re seeing in the broader economy. Rising travel costs, shrinking discretionary budgets and consumers pulling back on the trips they planned six months ago. Resume and cover letter writing is climbing, which maps directly onto a labor market shaped by layoffs and graduation season. Our models are becoming more predictive than simply informative.

That means AI usage doesn’t exist in a vacuum. It’s a mirror of the economy. The smartphone is the looking glass through which those insights can be found.

The Distribution Illusion

Here’s where the story gets a little more interesting. And where the smartphone-as-gateway to AI ecosystem finds a few important competitive implications.

The PYMNTS Intelligence data shows that Android users are 24 percentage points more likely to use Google Gemini than iOS users — 73% versus 49%. That’s not a preference signal. It’s an embedded distribution advantage. Gemini ships as the default AI on every Android phone, pre-signed into the same Google account that holds the user’s  Gmail, Maps, Calendar, Wallet. Like Apple Pay on iPhones, it doesn’t need to be chosen. It just shows up there.

ChatGPT, meanwhile, is the de facto AI preference on iOS, where it faces no native OS-level competitor. iOS users prefer ChatGPT by 10 percentage points over Android users: 88% versus 78%.

What this suggests is that the phone you carry strongly influences the AI you mostly use.

Read More: The First Chatbot Consumers Try May Be the One They Stick With

And then there’s Copilot.

Microsoft’s workplace AI story is perhaps the most obvious example of what distribution without demand looks like. As of early 2026, Copilot has somewhere between 15 and 20 million paid enterprise seats. But the workplace conversion rate, the share of provisioned users who actively choose to use it, according to our data, is only a little more than a third (36%). ChatGPT’s is nearly three times that at 83%.

And according to the Recon Analytics dataset, when employees have access to both Copilot and ChatGPT, only 18% choose Copilot. When all three major platforms are available, Copilot’s share drops to 8%. But when Copilot is the only tool the employer provides? Adoption reaches 68%.

Translation: Two-thirds of Copilot’s workplace usage exists because workers have no other option. Is that really adoption? Or lock-in captivity?

What these captive workers do when they want the AI they actually prefer is to pick up their phone and open ChatGPT.

This is consistent with what I’ve written about before. Consumers use more than one AI model, but they have a favorite. The average active AI consumer engages with more than two platforms; power users engage with nearly four.

Read More: The Battle for AI Isn’t About Models. It’s About Habits

But engaging with multiple platforms is not the same as dividing time equally among them. The pattern mirrors how most people shop. One go-to that handles the everyday, and a short list of specialists that earn their slots by doing something the go-to doesn’t do as well. ChatGPT is that anchor for most consumers today. The phone they carry determines how easily they reach it. And how hard the alternatives have to work to earn a slot in the rotation.

The Engagement Paradox

There’s another layer of the PYMNTS Intelligence consumer AI adoption data that makes the smartphone-as-AI platform argument even more complicated. And that’s how AI-fluency varies across operating systems.

iPhone users are 12 percentage points more likely to be active AI users — to use our vocabulary, power users or mainstream users — than Android users. Fifty-two percent versus 40%. On the desktop side, 60% of Mac users are active AI users versus 47% of Windows users. Macs score 21% power users versus 12% on Windows.

What separates these segments isn’t just how often consumers open an AI app. It’s the depth and breadth of that usage. How many of the 54 tasks they’ve moved to AI and across how many categories of their daily lives.

Read More: The Data Behind AI’s Shift to Everyday Consumer Use

Light users average roughly two activities per month. Mainstream users average eight. Power users perform more than 25 distinct AI-driven tasks a month spanning everything from shopping to health to financial management. That intensity is how we measure how embedded AI has become. And it’s where the smartphone-as-AI-platform thesis gets its economic teeth.

iPhone users are the most AI-engaged consumer base, the users most ready for always-on AI that users can query, or for agents that deliver suggestions and outcomes on their behalf. (We call this anticipatory AI.)

But Apple’s own AI (Siri and Apple Intelligence) is the weakest of the bunch. Google has the strongest AI integration, Gemini baked into every layer of Android, but the largest pocket of non-users to convert. Forty-nine percent of Android users are AI non-users, compared to 37% of iOS users.

Google has the integration. Apple has the audience.

And the smartphone is where those two things intersect.

What the Consumer Actually Wants

We asked AI users in March for the first time to tell us whether they want AI that comes to them (embedded and proactive) or AI that responds to their initiated prompts (agents that are reactive at the prompt).

The answer is a near-even split.

Forty-one percent of users prefer to activate AI themselves, the “AI they go to” model. A little more than a third, 36%, want AI that anticipates their needs and surfaces information and actions automatically, the “AI that comes to them.” Twenty-three percent are on the fence.

But peel back the averages and we find a richer story.

Among power users, 55% want AI that comes to them. Among light users, just 18% do. Among Gen Z, 48% want AI that comes to them. Among boomers, just 21% do. Among consumers who have already fully replaced their old search and discovery methods with AI, 49% want the proactive model. Among those who only lightly complement their old methods, 25% do.

Here’s where the plot thickens. iPhone users are 7 percentage points more likely than Android users to want AI that comes to them. That’s despite Apple currently lacking a native proactive AI surface that matches Gemini’s reach.

The demand for embedded, always-on AI is sitting inside the Apple ecosystem, waiting for someone to serve it.

That someone doesn’t have to be Apple.

The Smartphone Wars Are Coming, Part Two

On April 27, analyst Ming-Chi Kuo reported that OpenAI is developing a smartphone in collaboration with Qualcomm, MediaTek, and Luxshare. Mass production is targeted for 2028. The device is said to reportedly replace apps on the home screen with an AI-agent interface. The kind where ChatGPT agents handle tasks directly rather than routing users through individual apps.

The same day, Sam Altman posted on X that it “feels like a good time to seriously rethink how operating systems and user interfaces are designed.”

None of this should come as a surprise to regular readers of my columns.

Two years ago, I wrote that a new dream team was reimagining the future of smartphones with GenAI, and it wasn’t at Apple. Sam Altman and Jony Ive, Apple’s original design visionary, were raising a billion dollars to create an AI-powered device. I asked then whether this new device could become as transformative as the iPhone was in 2007. I also noted, with some irony, that one of the rumored investors was Laurene Powell Jobs, Steve Jobs’ widow. The dream team has since become a dream company. OpenAI acquired Ive’s firm for $6.5 billion. Since then, the ambition has shifted from a device to a phone.

A month before the Kuo report, Reuters reported that Amazon is developing a smartphone within its Devices and Services division. That effort is led by a group called ZeroOne under the leadership of a former Xbox executive. It’s said that the device would integrate with Amazon’s shopping, Prime Video, Prime Music, and Alexa+ ecosystems.

The conventional analysis of these moves focuses on whether OpenAI or Amazon can compete with the iPhone. History would say, whoa, not so fast. The Fire Phone. Windows Phone. Facebook phone. The graveyard of failed smartphone challengers is deep and well-populated with some of the biggest names in tech.

Then again, maybe that was then and this could be what’s next.

The question isn’t whether OpenAI can outsell Apple. It’s what happens if the company with 900 million weekly active users and 50 million paying subscribers decides that it no longer wants to access those users through someone else’s hardware.

ChatGPT was the most-downloaded app of 2025, with 770 million installs. Its mobile app alone generated $1.35 billion in revenue for Apple that year. That would make OpenAI the largest tenant in Apple’s App Store by any AI measure. And Apple takes a cut of every subscription sold through it. Every ChatGPT query flowing through an iPhone is an interaction that OpenAI controls and Apple monetizes.

That’s a business relationship that works until it doesn’t. And OpenAI is clearly signaling that it wants to own the full stack.

The Google Problem

If OpenAI builds a phone, it almost certainly runs Android. Building a proprietary OS from scratch is the path that killed every previous challenger. You need the app ecosystem, the carrier relationships, the developer tools. Android provides all of that.

Amazon learned this lesson the hard way with the Fire Phone. It ran Fire OS, a fork of Android that lacked access to the Google Play Store and popular apps. It was a technical achievement and a commercial disaster.

The OpenAI phone would need to somehow maintain Android app compatibility while replacing the AI layer with ChatGPT. That’s possible in theory but challenging in practice. Google controls the default search engine, the default assistant, the default maps, and the default mail client on every Android phone. Displacing all of that requires a value proposition that goes far beyond “our chatbot is smarter and our phone is cooler looking.”

Amazon’s Different Bet. And the Agentic Commerce Land Grab

Amazon’s phone thesis is different from OpenAI’s and grounded in the behaviors that now drive nearly 60% of all online sales in the U.S.

Amazon is betting that the commerce ecosystem is the product and the phone is the lock-in mechanism to access it. And deliver the personalized and proactive AI experiences that consumers say they want more of.

Amazon doesn’t need consumers to think its AI is the smartest. It just needs consumers to buy more things from Amazon, watch more Prime Video, use Alexa+ to manage their households and Rufus to chase down what they want to buy. Their phone becomes a dedicated remote control for the Amazon economy. An economy that has expanded beyond retail and groceries to healthcare, cars, food delivery, pharmacy and insurance, both in and outside of its own ecosystem.

Read More: Amazon Says Rufus Gives It an Edge in Agentic Commerce Race

That’s a narrower pitch, and maybe even a more practical one. It’s also the one that faces the highest bar to clear. People don’t need a dedicated phone to be part of the Amazon economy today. For Amazon’s phone to succeed, Amazon has to do in 2026 what Steve Jobs did with the iPhone in 2007.

Convince consumers to try something new that they never asked for and didn’t think they needed. And to be blown away when they try it.

But there is one thing that makes Amazon’s position a little stronger than “there they go again.” Amazon wouldn’t just be building a phone. It would be building the connective tissue to the agentic commerce ecosystem that everyone else is trying to replicate, the ecosystem Amazon already has.

In March, Amazon expanded its Shop Direct program to 100 million products from more than 400,000 merchants. Most of those merchants don’t sell on Amazon’s traditional marketplace. Using AI agents, Amazon can now complete purchases on third-party websites on behalf of consumers through its Buy for Me feature, using the shopper’s saved Amazon address and payment credentials. Amazon is going outside its own walls, reaching into the open web, and pulling commerce back through its ecosystem. It isn’t waiting for merchants to come to the marketplace. It’s sending agents to the merchants and delivering sales. And the logistics that close the loop from sale to delivered product.

And then, in February, Amazon announced a $50 billion investment in OpenAI. The company with the most popular AI answer engine, ChatGPT and its 900 million weekly users, is now joined at the hip with the company that owns the most popular product search engine and the most sophisticated fulfillment infrastructure on the planet.

OpenAI’s models will power Amazon’s customer-facing applications. Amazon’s commerce fundamentals, which include real-time inventory, delivery logistics, payment credentials, and purchase history, will power OpenAI’s agents.

This is the dimension of the AI competition that most of the “who has the best model” coverage whiffs. The real land grab isn’t the AI model. It’s the ecosystem. It’s who builds the platform where AI agents don’t just answer questions but actually do things. That shop, book healthcare appointments, manage groceries, schedule entertainment, handle insurance claims, and reorder prescriptions. The consumer’s entire agentic experience, mediated by AI and agents, flowing through a single set of commerce rails.

Read More: Why AI Shopping Is Still Just a Smarter Search Bar

Google has search and ads but not fulfillment or payments at scale. Apple has the device but not the commerce infrastructure. OpenAI has the users and the model but not the transaction layer. Amazon has the whole stack. And now it has OpenAI too.

Apple’s Accidental Genius

And then there’s Apple.

Apple’s fiscal Q2 2026 earnings, reported on April 30, tell an AI story that is hiding in plain sight. Services revenue reached a record $31 billion in the quarter, up 16% year over year. iPhone revenue hit $57 billion, a March quarter record. Total revenue was $111.2 billion. Gross margins were 49.3%.

Services revenue now exceeds Mac, iPad, and Wearables revenue combined. It’s the monetization layer that sits on top of the hardware installed base. And AI is becoming a meaningful, and growing, component of that layer.

By being smart in a very clever-like-a-fox kind of way.

Apple doesn’t have to be the smartest AI to capture the economics of AI. By plan or pure luck, you decide, they just need to be the device through which someone else’s smart AI is accessed.

Every ChatGPT subscription purchased through the App Store. Every Gemini query routed through iOS. Every AI-powered app that lives on the iPhone. Apple takes a percentage of each transaction without having to build a model, train a model, host a model, or explain why a model hallucinated.

Apple’s the AI toll bridge, not the agentic highway.

The PYMNTS Intelligence data shows why this position is sticky.

At least right now, iOS users are the most AI-engaged demographic and the most likely to want anticipatory AI. That’s the kind of experience where an agent knows the consumer well enough to surface three dresses and a matching hat for the wedding in London in July, consistent with their preferences, favorite color, budget and the event, before they’ve started looking.

But the agentic trust picture is more complicated than Apple might like to admit. AI-native brands like OpenAI and Anthropic are the most trusted providers of AI assistance over handset manufacturers. The phone manufacturer comes in slightly behind that for both iOS and Android.

That suggests that Apple’s hardware trust gives it permission to be the intermediary, but consumers trust the AI companies themselves more than they trust the device maker. At least, that’s what consumers say  now.

It’s true that Apple’s installed base is more active with AI and more open to the idea of a trusted device intermediary than Android’s. And that’s a real advantage. But it’s a tenuous one. One built on the absence of a better alternative, not on deep conviction that Apple is the right company to deliver an intuitive, smart AI experience itself.

Tenuous positions have a way of unraveling from multiple directions at once. Apple’s is unraveling from three: the courts, the regulators, and the AI companies whose subscription revenue it clips.

The Tollbooth Problem

Apple’s App Store commission on AI subscriptions is, functionally, interchange. A percentage of every transaction that flows through the apps ecosystem Apple controls, charged to merchants who have no alternative if they want to reach the consumers on the other side.

Thirty percent the first year, 15% thereafter. For a company that didn’t build the AI, didn’t train the AI, and struggles to ship a version of Siri that can hold a conversation, it’s a pretty good deal.

The numbers are getting large enough to matter. Apple’s cut of just ChatGPT’s subscription revenue represents a revenue stream that most AI startups would envy. And Apple earns it by doing nothing more than owning the handset and operating the App Store.

I flagged this dynamic back in October 2024, when Apple Pay turned ten. Then I wrote about the fault lines forming beneath Apple’s iPhone-centric business model. The combination of regulatory pressure and GenAI, I argued, had the potential to become the biggest disruption to the smartphone ecosystem since Apple launched the iPhone in 2007. Those fault lines appear wider now.

Read More: Apple Pay’s 10-Year Journey and Its Next Decade of Decisions

The problem is that interchange has a history. And that history is one of regulatory pushback.

Anyone who has watched the payments industry over the past fifteen years knows the gory details. Interchange starts as a fair fee for services rendered. Network access, fraud protection, transaction processing. It becomes a significant revenue engine for the network operator. Merchants complain. Regulators investigate. Caps get imposed. The Durbin Amendment. The EU interchange regulation. Australia. India. The pattern repeats across geographies and decades. And it always ends the same way. The party taking the clip gets clipped.

That pattern is now playing out with Apple.

The Epic Games case has been working through the courts for years, and the trajectory is clear. In April 2025, the District Court found that Apple had willfully violated an earlier injunction by systematically erecting barriers to prevent developers and consumers from actually using external payment options. The Ninth Circuit backed that finding. Apple’s attempt to charge a 27% commission on purchases made outside the App Store, a hardly magnanimous three-point discount for the privilege of not using Apple’s payment system, was rejected. The court ruled that any commission must be limited to costs that are genuinely and reasonably necessary, not a de-facto tax on all commerce that touches the iPhone.

Just days ago, Apple lost its bid to pause these changes. Developers can now link to external payment options without Apple charging a commission on those purchases.

And Epic is not alone. In the U.K., a Competition Appeal Tribunal ruled against Apple in a £1.5 billion class-action representing 36 million consumers, finding Apple abused its dominant position. In the U.S., the Department of Justice and fifteen state attorneys general have filed suit. The legal pressure is global, it’s accelerating, and it’s aimed squarely at the revenue model that makes Apple’s “toll bridge” strategy a cash register.

If you’re OpenAI, this is the reason the AI phone project might be a hedge rather than a gamble.

If the courts force open payments, and the trend line strongly suggests they will, OpenAI can route subscribers directly and keep more of its revenue without building a single handset. The phone becomes less urgent. But if Apple finds ways to maintain its commission, or if system-level AI integration becomes the competitive moat, or, or, or, then owning the hardware is the only way out.

Just Don’t Ask Siri

Which brings us to the question that’s becoming harder for Apple to answer. What happens when the toll bridge operator is also supposed to be the AI provider?

Last July, I wrote that Apple Intelligence didn’t appear in any serious public GenAI ranking. That it lagged behind even Grok. I called it a strategic misstep. Maybe even a crater-sized hole that could prove Apple’s fatal flaw. I asked whether Apple was at risk of repeating IBM’s trajectory. That of an early innovator that squandered its AI lead and became irrelevant in the revolution it helped start. Siri was chat before there were chatbots. And then Apple watched while everyone else — Google, Amazon, OpenAI — built the future around conversational AI.

Read More: Apple’s $10B AI Crisis. 3 Bold Moves To Reinvent Its Future

Ten months later, not much has changed. Apple Intelligence remains a branding exercise more than a product breakthrough.

That strategy works when the phone is the universal remote control for AI. When consumers are happy to use the iPhone as the device that launches ChatGPT, summons Gemini, opens Claude. Apple doesn’t need its own AI to be good. It just needs to be the best hardware through which other people’s AI gets accessed.

Read More: Apple Says It Will Let Siri Access Multiple AI Assistants

But the shift toward agentic AI could change that thesis. Agents don’t just answer questions. They book flights, manage calendars, execute transactions, coordinate across apps. They need system-level permissions that Apple has historically guarded with religious zeal. An AI agent that can’t access contacts, or read notifications, can’t initiate a payment without three permission dialogs isn’t an agent. It’s a friendly chatbot with agentic ambition.

The agentic future isn’t theoretical anymore. Sure, we’ve seen our share of initial agentic commerce missteps. Does anyone remember how miserable the first iPhone experience was? But technology improves and the flywheel starts to rev and use and adoption take off.

Read More: Consumers Stop Sampling AI and Start Relying on It

The companies building these agents — OpenAI, Google, Amazon, Anthropic — all understand that the agentic future requires either deep OS integration or ownership of the OS itself. And Apple is the company least likely to grant that integration to a third party.

So, the question becomes something like this: Is the iPhone the universal remote control for the AI era? Or is it the cable box, the device that once controlled the living room but became irrelevant when the apps it was supposed to deliver started going directly to the TV?

Cable boxes didn’t die because they stopped working. They died because they stopped being necessary. The content providers, first Netflix and then everyone else, realized they could reach consumers without the intermediary, and the intermediary’s toll became unjustifiable. They had a financial incentive to explore those alternatives. The cable company’s response was to make the box “smarter,” to add apps, to integrate streaming services into the guide.

Apple’s bet is that the iPhone is stickier than the cable box ever was. That’s probably right today. People don’t switch phones the way they swapped out set-top boxes. The switching costs are higher, the ecosystem lock-in is deeper, the muscle memory is more entrenched.

But the history of platforms teaches us that the intermediary’s power is a function of how much value it adds versus how much value it extracts.

Right now, Apple adds a lot. The hardware, the security, the ecosystem, the trust. The courts are asking whether the toll, the 30% of every AI subscription, every in-app purchase, every transaction, is proportionate to that value.

And the AI companies building the next generation of agents are asking whether the intermediary is necessary at all.

The Real Question

The AI industry is obsessed with the model race. Who’s smarter? Who’s faster? Who’s cheaper per token? Those questions matter, but they’re not the ones that will determine who captures the economic value of AI over the next decade.

Read More: The Battle for AI Isn’t About Models. It’s About Habits

There are two questions that might.

The first: Who controls the device in the consumer’s hand? Because the device determines the defaults, the defaults determine the distribution, the distribution determines the usage, and the usage determines the revenue.

The second: Who builds the ecosystem where AI agents actually transact? It’s not enough to have the best model or even the best phone. You need the commerce ecosystem. The merchant relationships, the payment credentials, the fulfillment infrastructure. You need the whole agentic economy. Not just the chatbot concierge at the front door.

The phone, in this context, becomes something more than a screen, a camera and a music player. It becomes the command center for the agentic economy. The device where the consumer’s AI agent lives, where it has their permissions, their payment credentials, their preferences, their context.

Whoever controls that device controls the default agent. And whoever controls the default agent controls the commerce that flows through it.

Google understands the device question. That’s why Gemini is embedded at the OS level on Android. Microsoft understood a version of it. That’s why Copilot is bundled into every Microsoft 365 license, although the data shows that bundling without quality produces adoption without loyalty. OpenAI is starting to understand both questions. That’s why it’s building a phone and partnering with Amazon and others building AI-native marketplaces like Marc Lore’s Wizard.

Amazon understands the ecosystem question better than anyone. That’s why it’s investing $50 billion in OpenAI while sending AI agents to buy from merchants outside its own marketplace.

Apple understood the device question first. It called it the iPhone. But toll collectors don’t build economies. They collect rents until someone builds a better highway that avoids them. And the companies racing to build an agentic economy aren’t asking Apple’s permission.

At the same time the courts, the competitors, and the consumers are all starting to ask the same question Steve Jobs answered on that stage in 2007 with his “internet communicator.”

Just from the other direction.

What if the AI tail wags the smartphone dog so hard that the whole thing has to be reinvented?

Including by Apple itself.

 

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.