The AI industry has found its favorite metaphor: AI is the new smartphone. VCs use it to justify billion-dollar bets, executives lean on it to explain platform strategies, and analysts wave it around to forecast market dynamics. There's just one problem - it's wrong in all the ways that matter.

Good metaphors are both helpful and dangerous. They're helpful because they let us pattern-match against the familiar, but dangerous because they convince us to refight the last war instead of preparing for the next. Mistaking a narrative device for an analytical framework is how we end up building the wrong products, backing the wrong horses, and fundamentally misreading where power will accumulate.

The smartphone analogy is easy to fall back on. It's clean, familiar, and brimming with possibility. AI does exhibit the hallmarks of a platform shift and feels like a generational consumer technology wave. But it's still an entire universe away from the smartphone revolution, with structural, economic, and regulatory differences that make the analogy more misleading than illuminating.

I don't want to dismiss the metaphor entirely - it captures something about the excitement / anticipation of the moment. But if we're going to use it, we need to understand precisely where it illuminates and where it distorts.

When the iPhone launched in 2007, it wasn’t obvious that it would reset the computing stack. Early iPhones lacked 3G connectivity, copy-paste, and even an App Store. But Apple’s genius was in creating the conditions for a platform: a combination of hardware, software, and distribution that could attract developers, reshape user habits, pull in adjacent industries and upend tech hardware.

It's easy to forget how fast it all happened, but by 2010, the smartphone was less a phone than a hub for daily life. Maps killed Garmin. Instagram killed point-and-shoots. Payments, rides, food delivery, entertainment - all bent toward the device.

Crucially, the smartphone cycle was measurable. Hardware unit sales gave analysts metrics. Platform lock-in created defensible moats. Consumers opted into the new era with a single purchase. That clarity shaped the business models and competitive dynamics that followed.

Like the smartphone, AI is a horizontal technology: it's not one product - more a substrate for many. The iPhone enabled Uber; AI enables GitHub Copilot. The App Store created a distribution funnel for developers; AI models create a distribution funnel for applications. Consumers don’t care about the model weights; they care about the experiences those models make possible.

The analogy works when thinking about platform dynamics. A handful of firms dominate the frontier. OpenAI, Google DeepMind, Anthropic, Meta - these are the equivalents of Apple and Google circa 2010. Their models are the operating systems of the new era, setting standards, APIs, and rules of engagement. Applications, whether startups or incumbents, depend on them. The same tension between platform and developer is already replaying at the model layer.

And that’s largely where the similarities end. The smartphone was a hardware revolution. AI is not. There is no single product you buy and carry that announces: I’ve entered the AI era. Instead, AI slips in through features. Gmail autocompletes your sentences. Photoshop offers generative fill. Microsoft Office adds Copilot. Consumers don’t opt into AI with a purchase; they wake up one day and realize they're already surrounded.

That difference changes adoption dynamics. Smartphone growth was a classic S-curve: new device, early adopters, mainstream explosion, saturation. AI growth is messier. It spreads across software categories, not hardware shipments. Measuring adoption means looking at feature usage, not unit sales. That makes the business harder to analyze and easier to hype.

Smartphones produced enormous hardware profits. Apple, Samsung, and carriers captured value through devices and distribution. AI is dominated by infrastructure costs. Training models consumes billions in GPU clusters and energy. Inference at scale is expensive. Value capture tilts toward the firms with the capital and distribution channels to sustain that burn. Comparing this to a consumer hardware boom misleads both investors and founders about where profits will (eventually) land.

The smartphone era had clear distribution: app stores. That channel became the choke point for power, allowing Apple and Google to take 30% of revenue and control discovery. AI has no equivalent. There is no “AI store” - distribution is embedded. Microsoft pipes models into Office and Teams. Google pipes them into Search. Startups depend on viral growth, enterprise sales, or integration into incumbent workflows. Discovery looks less like browsing a storefront and more like stumbling upon a new autocomplete feature.

Monetization diverges. Apps could be free with ads or $0.99 downloads. AI applications can’t sustain that model; inference costs are too high. Subscriptions and enterprise licensing are the natural answer, but that difference tilts the ecosystem toward B2B adoption first. GitHub Copilot, customer service automation, and productivity integrations are leading the way, while consumer chatbots still search for a durable revenue model.

And no, $200 a month for access to a browser doesn't count.

If the smartphone metaphor obscures, what serves better?

The cloud? AI is invisible infrastructure that changes how software is delivered and consumed. You don’t “buy” cloud; you use apps that rely on it.

Energy? Once electrified, industries transformed quietly but irreversibly. AI may follow that path: ambient, infrastructural, and everywhere.

Neither (and no) metaphor is perfect. Cloud captures the economic structure, energy the pervasiveness. The smartphone captures the platform dynamics. The problem, exacerbated by decades of narrative shorthand, is resisting the urge to pick one and call it definitive. AI, like any general-purpose technology, refracts through multiple lenses at once.

If you think AI is the new smartphone, you might hunt for a killer app to rival Uber or Instagram. But the better strategy may be to build enduring workflows, deeply embedded in enterprises, where switching costs and data flywheels matter more than consumer virality. If you think AI = energy, you focus on infrastructure and distribution. If you think AI is cloud, you build middleware and integration.

For incumbents, the analogy can lull you into complacency. Apple and Google cemented dominance in the smartphone era because they owned the platforms. But in AI, incumbents already own distribution. Microsoft doesn’t need to create a new device; it only needs to make Copilot the default. Meta, already behind, may be learning this the hard way.

“AI as the new smartphone” will stick around - because it’s easy, because it's both descriptive and communicative, because we're lazy and obsessed with patterns, etc. But strategy can't be built on metaphor alone. The smartphone comparison is true in scale but false in structure. AI will be bigger, broader, and less visible than the iPhone ever was. Companies that understand that distinction - the folks who can resist chasing a consumer S-curve that isn’t coming - will be the only ones left with a chair when the hype cycle gives way to durable economics.