The ‘killer AI product’ myth, explained
Steven Levy’s Wired piece, published after Apple’s CEO transition announcement, carried the headline “Apple’s Next CEO Needs to Launch a Killer AI Product.” The framing is seductive and wrong.
Levy’s argument follows a familiar pattern: Apple succeeded because it invented category-defining hardware — the iPod, the iPhone, the iPad — and therefore AI demands the same formula. Find the breakthrough device, ship it, win the decade. It treats AI the way tech media treated wearables in 2014: as a product category waiting for Apple to enter it.
Apple’s own executives rejected this framing directly. Hardware chief John Ternus, when pressed on Apple’s AI strategy, described AI as “an immense kind of inflection point” but placed it alongside the other platform shifts Apple has navigated — not above them. His more revealing line: “We never think about shipping a technology. We want to ship amazing products.” Marketing head Greg Joswiak echoed the same posture. Neither man named a specific AI product on the horizon, and that absence is the point.
The history Ternus cited reinforces why the “killer product” lens fails here. The Apple II, the Mac, iTunes, the iPod, the iPhone — each one piggybacked on a previous product and a previous technological wave. None of them were the technology itself. The transistor wasn’t a product. The internet wasn’t a product. AI sits in the same category: an enabling layer that makes other things possible, not a thing you box and sell.
Demanding a single killer AI product from Apple misunderstands what AI actually is. It conflates the layer with the application, the infrastructure with the experience built on top of it. Levy’s framing makes for a sharp headline. It does not make for a useful diagnosis of what Apple needs to get right — or what it’s actually getting wrong.
AI as infrastructure, not invention
John Gruber’s argument cuts through the noise cleanly: AI is not a product. It’s a technology platform — closer in kind to electricity or the internet than to any device Apple has ever shipped.
That distinction matters enormously. Electricity didn’t produce one killer appliance. The internet didn’t produce one killer website. These technologies restructured everything they touched, gradually and then completely, embedding themselves into existing systems until the infrastructure became invisible. The value wasn’t in a single invention — it was in how deeply the technology wove itself into daily life.
The “killer AI product” framing assumes AI follows the pattern of the iPod or the iPhone: a discrete object that changes the market on launch day. Apple’s own hardware chief Jeff Ternus pushed back on exactly this reading in a conversation with Steven Levy at Wired. Ternus described AI as “an immense kind of inflection point” but framed it as one Apple navigates the same way it always has — by building products, not by shipping raw technology. Every major Apple success, from the Apple II through the Mac, iTunes, the iPod, the iPhone, and the iPad, piggybacked on a previous product. The product was always the vehicle.
That instinct is correct, but it also contains a trap. When technology functions as infrastructure, the companies that win aren’t necessarily the ones with the most elegant standalone device. They’re the ones that embed the capability so thoroughly across their existing ecosystem that pulling it out becomes unthinkable. Amazon didn’t win cloud computing by announcing a flashy server. It won by making compute capacity available everywhere, at any scale, quietly and reliably.
For Apple, this means the AI challenge isn’t about announcing something new. It’s about how deeply AI restructures what iPhones, Macs, AirPods, and Apple Watches already do — and whether that integration happens fast enough, and works well enough, that users never have to think about it at all.
What most tech coverage is getting wrong
When Wired published its piece under the headline “Apple’s Next CEO Needs to Launch a Killer AI Product,” it crystallized exactly how most tech journalism frames the AI race: as a product competition, scored by launches and feature announcements. That framing is wrong, and it distorts what readers and investors actually need to understand.
The product-centric lens creates a perverse incentive. Companies learn that performing AI progress — staging demos, shipping half-baked features, holding splashy keynotes — generates the same headlines as building real AI capability. Apple announced Apple Intelligence. Google announced Gemini integrations. Amazon refreshed Alexa. Each announcement triggers a fresh round of winner-and-loser coverage, regardless of what the underlying technology actually does.
What that coverage consistently ignores is the more consequential competition: who controls the foundational models, the custom silicon to run them efficiently, and the data pipelines to improve them. Those three layers determine long-term leverage. A company that licenses its core model from a partner — as Apple does with OpenAI for certain Siri functions — has handed a structural advantage to that partner, no matter how polished the product sitting on top looks.
Benchmarking by product launches also obscures the chip question almost entirely. Nvidia’s data center revenue crossed $47 billion in a single quarter in 2024. That number tells you more about where AI power is concentrating than any feature comparison between Siri and Google Assistant. The companies controlling inference hardware control the economics of every AI product built on top of it.
Readers following product-centric coverage end up measuring the wrong scoreboard. Investors pricing AI winners by which company shipped the most compelling consumer feature last quarter are missing the structural shifts happening one layer down. The killer-product framing isn’t just analytically shallow — it actively misdirects attention away from the decisions that will determine which companies have durable AI positions five years from now.
Apple’s specific AI dilemma in this context
Apple Intelligence launched to a chorus of disappointment. Features arrived late, Siri remained clunky, and the notification summaries that did ship produced embarrassing errors. Critics framed this as Apple falling behind OpenAI, Google, and Anthropic. That framing misses the actual problem.
The real question is whether Apple is building durable AI infrastructure or just bolting features onto existing products. Those are fundamentally different bets. Feature development produces press releases. Infrastructure development produces leverage — the kind that compounds across every product line for years. Apple’s history with silicon, with the App Store, with iCloud, suggests the company understands this distinction. Whether Apple Intelligence reflects that understanding is still genuinely unclear.
The harder structural problem is privacy. Apple’s entire architecture is built around keeping data on-device and out of Apple’s hands. That design choice earns customer trust and regulatory goodwill. It also cuts Apple off from the cloud-scale data pipelines that make frontier models powerful. Training better models requires more data. More data requires centralized collection. Centralized collection contradicts Apple’s core privacy positioning. Private Cloud Compute is Apple’s attempt to thread this needle, but it remains unproven at the scale that would make it competitively meaningful against Google’s data infrastructure or Microsoft’s OpenAI partnership.
The CEO transition sharpens every tension. Tim Cook built Apple into the most valuable company in the world by executing on hardware margins, supply chain discipline, and services growth. His successor — now confirmed as operations chief Jeff Williams — inherits a company that must make a binary strategic choice: compete on AI as a product, meaning ship a breakout consumer application, or compete on AI as a platform, meaning build the infrastructure layer that developers and enterprises depend on.
Those paths require different resource allocations, different acquisitions, different organizational priorities, and different definitions of success. Apple cannot pursue both with full conviction simultaneously. The choice Williams makes in the next two to three years will determine whether Apple leads the next computing era or optimizes a very profitable position inside someone else’s.
The right questions to ask about any company’s AI strategy
Stop asking “what is Apple’s AI product?” Start asking where AI is embedded, and whether it makes existing products meaningfully better. Those are different questions with different answers, and the second set is the one that actually predicts competitive staying power.
Apple’s own executives signal this distinction directly. Hardware chief Jeff Williams has described Apple’s approach as never shipping a technology — shipping products. The Apple II, the Mac, the iPod, the iPhone: each one absorbed the technology of its moment and made it disappear into something people wanted to own. That pattern doesn’t produce a standalone AI app. It produces a better iPhone, a more capable Siri, a faster chip that runs models on-device without burning battery or sending data to a server.
Which points to the real scorecard for evaluating any company’s AI position: custom silicon, model ownership, and developer ecosystem. Not headline features. Apple’s Neural Engine, built into every A-series and M-series chip, gives it on-device inference at a scale no competitor matches by volume. That’s infrastructure. ChatGPT is a product. These are not equivalent assets.
Google and Microsoft illustrate how two serious AI investments can take completely different shapes. Google is rebuilding its core search product around AI-generated answers — a direct reinvention of its primary revenue engine, with real risk attached. Microsoft embedded Copilot across Office, Azure, and Windows — an AI-as-productivity-layer strategy that doesn’t require any single killer product to work. Neither company launched an “AI product” that explained their strategy. Both made bets on where AI would compound value inside things people already paid for.
The killer-product framing fails because it imports a smartphone-era mental model into a different kind of technological shift. The iPhone was a product you bought. AI is infrastructure you evaluate through the products it improves. Judging Apple’s AI strategy by whether it has a ChatGPT competitor is like judging its chip strategy by whether it sells CPUs retail. The question was always wrong.
Why this distinction matters right now
The product-versus-technology confusion isn’t a philosophical quibble — it’s a resource allocation problem with real consequences. Companies chasing “killer AI products” are building strategies around the wrong unit of analysis, and the bill is coming due. Investors who poured money into standalone AI applications expecting iPhone-scale adoption are now asking harder questions. Consumers who downloaded AI assistants expecting transformative experiences and got glorified autocomplete are tuning out. The hype cycle is entering a more unforgiving phase, and the framing you start with determines how badly you get burned.
Regulators face the same trap. Treating AI as a discrete product pushes oversight toward surface-level feature audits rather than scrutiny of the underlying systems, incentives, and data pipelines that actually determine how AI behaves. That’s the wrong level of abstraction to be regulating at. The EU AI Act, for all its ambition, wrestles with exactly this problem — drawing risk boundaries around applications while the foundational technology cuts across all of them.
For investors, the distinction shapes which companies deserve conviction. A company that understands AI as a capability layer to embed across its existing product surface — hardware, software, services — is building compounding leverage. A company searching for the standalone killer AI app is chasing a category that may never materialize the way smartphones did.
Apple’s incoming CEO Jeff Williams, or whoever ultimately takes the role, will face this framing test immediately. The question Wired and others keep asking — where is Apple’s killer AI product? — is the wrong question, and answering it on those terms is a trap. Jeff Ternus said Apple never ships a technology, it ships products. That instinct is correct. The challenge is that executing on it requires organizational clarity about what AI actually is: infrastructure, not inventory.
Getting this right isn’t semantic tidiness. It determines which companies earn trust, which get funded through the next correction, and which ones regulators hold accountable for outcomes rather than features. The companies that internalize the distinction now will be harder to blame for the wrong things and easier to credit for the right ones.
