Apple Just Positioned Itself For The Next Trillion Dollars

TITLE: Apple Just Positioned Itself for the Next Trillion Dollars CHANNEL: AI News & Strategy Daily | Nate B Jones DATE: 2026-04-26 ---TRANSCRIPT--- Tim Cook has stepped down as Apple CEO. The coverage so far is focused on the continuity, right? Apple lifer taking over, smooth handoff, stable transition. That’s all accurate, and there’s a version of this story where that’s the whole story. But, I want to add the piece that comes into focus if you look at the org chart underneath the announcement. The person taking over is a hardware engineer. His second-in-command designs chips, and that combination at the exact moment Apple is losing the AI race is Apple declaring they’re going to fight that race on completely different terms than everybody else in the industry. What that means for you depends on where you sit, and we’ll get to the specifics. But, I want to walk you today through five things about this Apple bet. One of them is actually about Apple. The rest are about what’s happening under the surface of the AI industry right now, and they matter whether you own an iPhone or not. Start with the news because the org chart tells you more than the succession does. The new CEO is John Ternus. For 25 years, he’s been an Apple hardware engineer. He ran the team that moved the Mac from Intel to Apple silicon, which is arguably the most successful silicon transition any consumer company has ever pulled off. Right below him, Johny Srouji, the guy who’s been running all of Apple’s chip design for the last decade. He’s getting elevated to a new role called the chief hardware officer. So, the top two execs at Apple are both hardware engineers at core, both silicon people. Neither of them comes from software, from services, or from AI. And that matters, and to see why you have to understand how Apple has been organized for the last 15 years. Tim Cook’s Apple is what’s called a functional organization. There’s no phone team, there’s no Mac team, there’s no watch team. Instead, there’s a hardware team, a software team, a services team, and a design team. No one team owns the product. The iPhone is where all those teams meet and argue and integrate. Steve Jobs built the company that way on purpose when he came back in the late ’90s because he thought product owned orgs tended to produce incoherent devices. If the phone team owns the phone, the phone gets optimized for what’s good for the phone team. If no team owns it, the phone gets optimized for the intersection of all the things, which was important to Steve cuz he wanted the complete experience. That’s how you get a device where the hardware, the software, and the services feel like the same thing as long as you have Apple’s culture. Because I will say here, I have seen this strategy tried at other companies and it fails without Apple’s culture. That model worked for Apple for 15 years. It built the iPhone, the watch, the AirPods, the whole empire. It also produced the Apple intelligence failure. And that’s the part I want you to sit with because it’s not a story about Apple being bad at AI. It’s a story about structures. Generative AI is not an integration product. It’s a capability race. The thing that matters isn’t how well the pieces fit together necessarily, it’s how fast you can ship the next model. If you’re a hyper scalar, how fast you can close the gap to whoever’s ahead, how fast your model development loop turns. The frontier labs ship a new model every quarter not because their engineers are smarter than Apple’s, sometimes every month now, right? But because their org charts let one person decide and push. Under Tim, Apple’s org chart doesn’t work that way. Consensus has to assemble horizontally, especially at the SVP level and above. Every major decision gets argued across functions before it ships and that’s how you get the iPhone. Again, as long as your culture is good. It’s also how you fall a year behind on AI features while the labs keep shipping or two years or three years cuz they’re behind. So, the Apple board had a tough choice. They could put a software leader in charge and try to force Apple’s software and services orgs to ship at a frontier lab cadence and try and reinvent themselves. Break the consensus model, put one person in charge of AI, or they could decide the race they were losing isn’t the race they wanted to play and they could change the game. They picked option two. They put hardware on the top. Instead of trying to match the frontier labs on their terms, which Apple wasn’t doing anyway, Apple just said, “You know what? We’re really not running the race. And I think that’s more interesting news than just like who was picked or why Tim stepped down. The Turnus pick is Apple admitting structurally that they cannot win a software velocity race in the age of AI and they’re betting on a different race entirely for the future of the company, which brings me to the second of the five big things I want to talk about. Because the race Apple wants to run, the one they think they can win, is being made possible by something that’s been reported on extensively in the financial press, but that hasn’t really been connected to the Apple store yet. AI business, the way it exists today, does not work at scale. I want to be specific because this matters. Every major frontier lab is losing money on the top tier of their consumer subscription. Sam Altman has said publicly that OpenAI is losing money on ChatGPT Pro even at 200 bucks a month. It’s not because those users are abusing the service, it’s because a capable model serving a serious user doing real work costs more to run than any consumer subscription price covers. The math is upside down and it’s being hidden right now by a few things, right? Investor capital is subsidizing the losses, GPU supply is expanding roughly in pace maybe a little behind demand, and the and the assumption that per token prices will keep falling faster than frontier capability continues to keep the market encouraged. But all of those assumptions are starting to get shakier. Investor appetite, I know it seems like it’s infinite because of all the capital these companies raised, but it’s not. At some point investors are going to start to expect a return, especially as Anthropic and OpenAI start to look to the public markets. GPU supply is constrained by power and fab capacity more than by Nvidia’s willingness to ship chips and that is a tough constraint to break. And harder than that is power. GPU supply is constrained by power and by fab capacity more than by Nvidia’s willingness to ship chips. Jensen wants to ship chips. And of the two, power is the harder of the two constraints. Per token prices are falling, yes, but frontier capability is scaling way faster than prices are dropping right now. So, the math is getting worse on a per serious user basis, not better. And where this ends, if nothing changes, is a two-class AI system. The top class, big enterprises signing seven-figure contracts or eight-figure contracts, they get the real AI. They get long context, agents running for days or weeks, dedicated capacity, and the rest of us get metered, throttled, consumer-tier access, because that’s what the labs can afford to serve us. The things you can do with AI start to depend on which class you can afford to be in. And that’s not really a prediction. You can see it happening today, right? Every consumer-tier rate limit that’s getting tightened in the last few months is the unit economics starting to speak, right? The labs are not necessarily being greedy here. They’re just choosing to bleed less. And that’s the direction all of these curves are pointing. And if you’re Apple, that’s kind of a scary curve, because your customers’ AI experience is about to be bounded by what the labs can afford to serve them for 20 bucks a month or less. You can’t build a 10-year product story on top of somebody else’s loss-making business with a customer-squeezed baked debt. Because so much of the value of the iPhone is the software, right? And you need AI to work on the iPhone to make that true. So, Apple needs an alternative, and there’s only one alternative available. And that brings me to the third thing that I wanted to talk about today. Apple is actually betting on a bet they’ve made a half a century ago, and it’s worth understanding why that precedent matters for this company in this moment. The alternative to cloud AI is to move compute off the cloud and onto the device. It’s really intuitive. This is what people mean when they say local AI or on-device or on-prem AI. And the framing you usually hear is about privacy. Your data stays on the phone, Apple doesn’t see it, regulators are happy. All of that is true, and it’s a real benefit. What I want to add is the first-order benefit underneath the privacy story, which is the cost structure. Here’s the real difference. On-device inference has a fixed cost. You paid for the chip when you bought the phone. Once the model’s running locally, asking it a thousand questions costs the same as asking it one. Essentially, nothing per query at consumer scale. It’s just electricity. Cloud inference has variable cost. Somebody pays every time you ask a question. Right now, that somebody is the lab subsidized by their investors. Eventually, that cost will get passed to the users. The meter starts running in ways you notice. And that’s been especially true since December as long running agentic workflows has kicked off and exploded demand for tokens. Apple silicon is the escape hatch from that meter and it contributed to the incredible popularity of open claw. People want models they can run locally. That’s why the Mac Minis are sold out. Apple is not going to beat the best cloud model on its own chips and they probably will not try. What they’re betting on is the long tail of things most people really use AI for. Summarizing documents, drafting emails, transcribing meetings, translating stuff, searching your own stuff, answering questions about your life, running routine agents against your data, getting AI involved in their health side of the business. If those tasks happen on the device, they happen outside the meter. The cloud becomes a specialist for hard problems, not the default for everything. And if you think that this is only for consumer applications and there’s no future for devs, there’s no future for prosumers, there’s no future for people who are serious about AI trying on prem, I got news for you. Apple has made this bet before. 50 years ago, in the 1970s, computing was a service. You rented time on a mainframe. The compute lived in someone else’s building. You paid by the hour and the users who got the most out of the system were the institutions like AT&T who could afford to pay the meter. The ordinary person did not interact with computing at all. The Apple II didn’t beat the mainframe on raw capability. Obviously, it couldn’t. What it did was move a useful amount of compute onto a device that you owned. And once you bought the machine, using [clears throat] it more cost you nothing. And the power user, the person who could afford to leave the computer running all night because nothing was metered, turned out to be the population that pulled the whole category forward. The prosumers succeeded in making Apple what it is. The spreadsheet happened on the Apple II. VisiCalc was not a mainframe product. It was a product that could only exist on a machine you owned. Same company, same structural move 50 years later. A metered service model where the heaviest users are a cost problem for whoever’s running the service, a device-based ownership model that drops that marginal cost to near zero, and a bet that the power user on the owned device is going to invent uses that the metered service cannot afford to permit at scale. Apple thinks they’re the Apple II in this situation. The rest of the industry is betting on the mainframe. And before I tell you why I think Apple might actually be right about this, I want to walk you through the most important piece of this whole story. It’s a specific population of buyers who are already trying to build a local solution themselves, and this is the piece I’m bringing from inside the buying conversations I’ve had over the last few weeks and months, and I think it’s the biggest part of the story. This is piece four. There is a specific, targeted category of buyer that keeps showing up with a problem the industry doesn’t have a solution for yet. I’m going to name them. Law firms, medical practices, accounting firms, tax firms, financial advisers, therapists, any professional whose job carries a very high bar on data confidentiality. Attorney-client privilege, right? HIPAA in the US, fiduciary duty, therapeutic confidentiality rules. These firms are watching their competitors pull ahead using cloud AI, and they’re facing an enormous amount of pressure to catch up, but they can’t, right? Running a public cloud AI service against client work product is often a practice problem, a regulatory problem, or at best a massive technical headache even if it’s compliant. Their clients would be within their rights to walk if they ever found out that confidential information had been processed by a cloud model owned by a company two layers removed and two countries removed from the relationship. So, these firms have a genuine problem. They need AI, they can’t use the cloud. What are they actually doing? A lot of them are converging on the same answer. They’re buying Mac minis. Literally, a handful of M-series Mac minis clustered together has enough capacity to run useful generative models locally. For a small firm, for a few thousand bucks in hardware sitting in a closet or the firm’s own network not talking to anything outside. The data does not leave the building, the privilege holds, the compliance story works. Now, before anybody in the comments says, “But Apple has private cloud compute.” They absolutely do. Apple has a cloud AI offering called private cloud compute that’s cryptographically attested. So, even Apple’s own administrators cannot read your data. Now, that’s a meaningful upgrade from normal cloud AI on privacy. It is not the answer for this segment, and I’ll tell you why. The problem for a law firm is not can a rogue cloud admin see my data. The problem is can I represent to my client, my regulator, and my malpractice carrier that this data never left my physical control? No cloud service lets you make that representation, no matter how good the cryptography is. Apple has explicitly declined to disclose where PCC nodes are physically located for their own quite valid security reasons, which is fine for consumer and an absolute non-starter for a firm that needs to know which jurisdiction its data touched. So, in many cases, instead of trying to deal with a massive technical headache associated with trying to set up a compliant local cloud setup, firms are improvising. They’re going after retail Mac minis, they have their own orchestration glue, they’re hiring a guy they know, they’re using open weights models they fine-tune for their domain, and they’re hoping the whole thing holds together. They’re doing that because Apple has not built the product they need, and no one else has either. There’s no rackable enterprise form factor for Apple silicon, no clustering software, no admin tools for IT teams running managed local inference, no identity layer that mirrors iCloud but stays on prem, no HIPAA business associate agreements, no curated model ecosystem positioned for regulated professional workflows. None of the infrastructure a law firm IT contractor would expect from an enterprise vendor. And that’s the thing I want you to hold on to. A US professional services economy, US alone, right? It’s measured in trillions of dollars and tens of millions of workers and a meaningful slice of that economy has a structural need for AI that never goes to the cloud. They know it. They’re often trying to buy a solution many times max and nobody’s selling it cleanly. It’s a complete mess. What that means for Apple’s bet is that the local AI story isn’t just about consumers on phones. It’s about every regulated professional in the economy who’s been locked out of cloud AI for compliance reasons and is trying right now to build a local alternative. Apple’s silicon is the natural substrate for these populations. The same chips that let your phone draft an email let a four lawyer firm run an AI in the closet. That that makes Apple’s on device bet much larger than the phone argument alone would suggest. It also means there’s a big product gap here. Apple might build the enterprise stack these buyers are asking for. Now, they have not signaled they will and their services strategy might push them in the opposite direction. If Apple wants to sell you iCloud, they may not want to ship a product on prem. But that gap is big enough that someone’s going to fill it. Either Apple will or a startup that wraps Apple hardware in an enterprise layer Apple won’t build will do the job for them. Just like third-party companies used to wrap IBM hardware in the services layer IBM would not build. That window is open today. It’s probably open for a couple more years before Apple either builds the stack themselves or the Qualcomm ecosystem closes in from below. And it’s one of the most interesting unserved opportunities I see in the AI market today. So, that’s piece four of five. The last piece here is what this means for you. Because if you’ve watched this far, you probably want to know, “Okay, fine. Interesting analysis, but what do I do with this?” This is going to depend on where you sit, right? There are three different shapes of response to this news for different folks. If you’re a leader, if you’re somebody running a company or you’re part of one or you’re making strategy calls, your big takeaway is this. When you’re losing a race, you’re structurally set up to lose, the move is not to try harder, the move is to change the game, which is a lesson a lot of folks who are competing on AI need to learn. And that’s what Apple just did. Most boards respond to losing by doubling down on the thing they’re losing at. Apple did not do that. They restructured around a race they could win. If you’re watching your org fail at something AI-related today, the question worth asking is, do we have a talent problem? Is there a premise problem? If it’s a premise problem, don’t optimize the premise, change it. The other takeaway for leaders, keep your eye out for business models that are structurally unprofitable because you need to be building as if that is the case. The labs right now are treating their consumer inference losses as if they’re a ramp into profitability, but that may not be the case. I walked through suggest there just may be a flaw here because of the pricing. If your strategy depends on cloud AI getting cheaper faster than it’s getting smarter, it’s not a plan you should plan on. Plan for the alternative. Let’s switch gears. Let’s say you’re a builder, a founder, an engineer, somebody making product decisions. The takeaway from this story is really what category do you build in, right? Don’t build AI-enabled products, build native AI products. The interesting opportunity is not wrapping GPT into your app. The interesting opportunity is the class of product that only makes economic sense when inference is free. Continuous background agents, assistants that read your users’ entire history without worrying about context windows, tools invoked thousands of times per hour without anyone caring about the cost. None of that is economically sane today on cloud APIs. All of it becomes sane on silicon your user paid for. And the SMB compliance segment I walked you through is a shippable startup thesis right now. Those buyers exist, they’re trying to find a solution. Nobody’s selling them a clean solution yet. If you have the back- ground to build that product, you should be building it. One more thing for builders, because the valley has shipped iOS first on every new consumer software category for the last decade. Instagram was iOS-only for 18 months. ChatGPT’s first mobile app shipped on iPhone before Android. Threads, Blue Sky, every premium consumer app of the last decade has launched on Apple silicon first. The reasons have nothing to do with AI. They have to do with who pays for premium apps. But that pattern compounds with Apple’s on-device bet in a way that even the bullish Apple analysis is kind of underweighted. If local AI becomes a category, the developer momentum is already pointed at Apple silicon. Apple doesn’t have to convince anybody to build for them. They just have to not screw up the platform terms. That’s a different bar than the one most people are grading Apple against right now. Now, if you’re a prosumer, if you’re somebody who uses AI intensely as part of your daily work, here’s the thing you should take away. Your ceiling is about to stop being your subscription tier and it’s going to start your literacy, right? Everything you do right now to conserve tokens, keep your context short, run one agent at a time, don’t ask the model to read big docs because it won’t hold them. That’s a habit shaped by having cloud AI. That habit might get in your way if you start to run local AI. You need to start to think about which AI am I wanting to use long term? And if I want to use local AI, am I ready to ask the model to do more for me, right? Get used to the idea that cost of usage could be something that either goes to the roof with cloud or goes to zero and choose accordingly. The second thing for prosumers data hygiene really matters. A local model is most useful when it can read all your stuff. This is actually true already with Claude and Claude code and Claude co-work. But your stuff is often scattered across a bunch of different places and most of them don’t like to export. So the work of consolidating your knowledge, your notes, your calendar, your messages, that work pays big returns over the next few years and frankly, another building opportunity, there’s not a good way to do that right now. The people who have been doing the most consolidation have already had an unreasonably good year this year because they’ve been able to use that file system to drive good agent work. And one last thing for prosumers, for the first decade of the smartphone, the difference between a two-year-old phone and the current one was small. That era is ending. If the on-device AI thesis holds, the neural engine generation you’re on starts to matter for what you do. I’ve seen this first hand moving from an M2 chip to an M5 chip. The case for buying the flagship and upgrading more often gets stronger than it’s been in a decade. This is also true if your employees are on Apple silicon. That’s a different consumer relationship with a device than most of us have had in a long time. It is more compelling to upgrade now, which is by the way something that Apple shareholders will really love. So, where does all this land? The turnus pick is a retreat that might succeed. Apple just broke a company that had worked for 15 years because the company they were building could not win the AI race on the terms the industry set. The new company that they’re putting together now has a shot on very different terms because the hardware economics of AI are fundamentally different from cloud economics of AI. In most of the industry has been quietly underpricing the difference. The rest of the industry is is running the the cloud play, right? Bigger data centers, more compute, more capex, and maybe they’re right. Certainly there’s room for lots of winners in this world. Somebody has to build the frontier and and the build out isn’t wasted capital even if the consumer economics don’t work, but Apple said kind of out loud with this pick what most of the rest of the players haven’t said. The cloud is expensive, the cost is real, the thing in your pocket might be what ends up mattering for AI. And the company that figured out how to put useful computing in your pocket in the first place 50 years ago might be the company that does it again. So, that’s where that leaves us. I am curious. Are you in the local compute category? Are you exploring on prem in your own home, in your own office? Or are you someone who says, “No, I want the cloud.” Let me know which in the comments. And of course, if you disagree with any of this, I love it. Put that in the comments, too. If this was useful, subscribe for more of this kind of analysis. I’m writing this up in way more depth on Substack. I think it deserves a lengthier treatment particularly around the SMB side and how that market is an opportunity for somebody whether it’s Apple or not. The best conversations in this channel happen under videos. So, I’ll see you at the next one and I’ll see you in the comments.