I built the first iPhone with Steve Jobs. The AI industry risks repeating a mistake

While recent investments from tech giants like Amazon and Nvidia suggest a capital expenditure (CapEx) bubble remains far off, the possibility floods headlines. True insulation requires meaningfully redistributing investment from CapEx to creating the apps and services required to achieve sustainable ROI on billion- and trillion-dollar infrastructure bets. It involves launching a new phase of the AI era that Goldman Sachs prices at $8 trillion. 

Infrastructure is part of the equation, not the whole game

Rewiring Americaprojects the national demand for electricity will grow by 128 gigawatts over the next five years. Meanwhile, data centers—under construction and in planning phases—will add 93 GW to the grid.

It’s clear from back-of-the-napkin math that we need to continue investing capital to construct physical infrastructure that supports the AI race. America’s private sector must play a leading role in helping community and national grids handle massive levels of electricity and compute. Large companies like Amazon, Nvidia, and Google and VC-backed startups, like Crusoe and OpenAI, are actively working on it. 

However, there’s a massive opportunity here to build out the application layer that scales AI into the physical world and ensures data center investment is worth every penny.

And there’s precedence from previous cycles. AWS as we know it today spawned out of the internal Amazon version powering company operations. Jeff Bezos and team not only built scalable infrastructure, they saw a trillion-dollar opportunity to commercialize and sell it to enterprise customers. 

AWS is now the backbone of the modern web. A key reason: it was built to support developers working to create, then scale applications for potentially millions of users.

The other example is the App Store. When Steve Jobs and his engineering team (that I was lucky to be a part of) stood it up, no one predicted Uber, Instagram, or DoorDash would hit the market. But it laid the foundation for a future we knew we couldn’t even imagine. Launching the App Store wasn’t just about selling access to smartphone users — it was about enabling millions of developers to build their own scalable applications that interact with everyday life in the real world. 

On the funding side, Kleiner Perkins worked with Apple to launch the iFund, a $100M investment initiative that provided capital necessary to make shifting from CapEx to app creation possible. Today, there’s an app available on the App Store for everything—from identity security to banking to dating to confirming a ride to the airport. 

AI is poised to power the world’s applications, but there needs to be a wave of net-new apps big enough to fill an app store — each using AI to solve problems better, faster, cheaper, and more efficiently where tech meets the physical world. 

Ensuring data centers power America into the future 

When I was at Apple, we didn’t just launch the smartphone era—we built an ecosystem. The iPhone succeeded because of the software and services that hardware designed for the real world unlocked. As we engineered and acquired our way to app store dominance in addition to shipping a world-class iPhone, we saw the business merit in the application layer and worked to bring a massive community of builders into the fold. 

I see an open door to do the same for AI. But an AI app store is no good unless we can fill it with apps that plug into the physical systems shaping our daily lives, including: energy, food, waste, water, transportation, the built environment, and healthcare.

These sectors that define quality of life for Americans nationwide. They’re also subject to productivity stagnation and extremely hard to scale.

Take the management of waste, for example. At Mill, we’re working to make food waste circular — converting what was once viewed as trash into a resource that can reenter the food system. AI can make these kinds of systems dramatically more efficient, from reducing the time people spend on every day chores, to automating recycling and optimizing collection routes with computer vision.

In healthcare, we’ve started to see that AI can accelerate drug discovery, automate administrative tasks, and personalize patient care.

None of these are moonshots, but they all require builders and investors.

The next phase of industrial AI

America’s next trillion-dollar opportunities lie in applying AI to industries and use cases that touch every household and business—not in training ever-larger models. The next industrial AI wave will benefit from nationwide data centers and infrastructure. Of course. But it won’t realize its full potential unless we start building applications for the real world. Today.

Working with America’s best innovators to ship a scalable, adaptable AI application layer is how we do it. It’s also how we insulate data center build-outs and AI progress from turbulent market cycles. That’s how we survive any AI bubble moment. And, most importantly, it’s how we ensure America never reaches a level of data center excess that renders billions of dollars and years of physical labor useless.

I’m confident data centers will serve as factories for the next phase of innovation where AI meets the physical world across sectors powering America’s society and economy. But they need to deliver a return on investment for builders, companies, and communities by enabling AI to produce something of value. That can’t happen unless investors shift focus and capital from infrastructure toward apps and builders ship real world AI applications designed to leverage increased capacity to solve real problems impacting daily life. 

Let’s get to work.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.