From Plastics to Compute: Why Apple and Uber Are Positioned to Outlast OpenAI and the AI Infrastructure Players

The artificial intelligence boom is real. Capital spending is exploding, compute demand is rising at unprecedented rates, and entire industries are reorganizing around machine intelligence. But history suggests a crucial distinction that investors often miss: growth in demand does not guarantee durable profits and the biggest winners are often not the ones building the most infrastructure.

We have seen this movie before.

In the mid-20th century, plastics were hailed as a miracle technology. Early producers earned extraordinary margins as polymers displaced glass, metal, and paper. Demand surged for decades. Yet by the late 1960s and 1970s, many of the companies that built the most capacity struggled with collapsing margins and poor returns on capital. The problem was commoditization.

AI hardware and data centers are now following a remarkably similar path. And if history is any guide, the most compelling long-term opportunities may lie not in the factories and fabs, but in the companies that control distribution, orchestration, and the interface between AI systems and the real world—including one platform quietly positioning itself as the operating system for physical movement, and another that may become the default gateway for consumer AI agents.

The Plastics Playbook: How Great Technologies Become Bad Businesses

In the early days of plastics, companies like Union Carbide, Dow, and ICI enjoyed a classic skimming phase. New materials solved specific industrial problems and commanded premium pricing. Process innovation and scale advantages translated directly into high margins.

But success bred competition. As more producers adopted similar processes, capacity multiplied. The industry shifted into a penetration phase, where pricing fell to drive volume. Output soared, but profitability stagnated or declined. This phenomenon, often called “profitless prosperity,” is a hallmark of capital-intensive commodity industries.

The Saudi petrochemical build-out accelerated this dynamic. Using cheap energy and massive export-oriented facilities, Saudi Arabia became a global low-cost supplier. Western firms licensed technology and entered joint ventures, effectively enabling their own future competitors. When global demand growth slowed, margins across the industry were permanently impaired.

AI Hardware Is Entering the Same Phase

AI chips began their life in a skimming phase. Nvidia’s early accelerators were scarce, mission-critical, and priced accordingly. Gross margins expanded rapidly as buyers had little alternative.

That phase is already ending.

Broadcom’s custom AI accelerators offer similar performance at materially lower prices. Hyperscalers, including Google, Amazon, Apple, Tesla, and others, are increasingly designing their own silicon to escape Nvidia’s margin stack. Broadcom’s own management has warned that custom AI systems will pressure margins, even as revenues surge, a textbook sign of transition from skimming to penetration pricing.

At the same time, the product cycle is accelerating. New generations of chips arrive every 18–24 months, rapidly devaluing installed capacity. Unlike plastics plants, which operated for decades, AI hardware becomes obsolete in just a few years. This shortens the window to earn back capital and amplifies downside risk when supply catches up.

The Saudi Data Center Parallel

Saudi Arabia is now replaying its petrochemical strategy in digital form.

Backed by sovereign capital and cheap energy, the kingdom is positioning itself as a global AI and data center hub. Massive investments are being made to attract hyperscalers, chip vendors, and AI developers. The goal is explicitly export-oriented: to process and host global workloads at structurally lower cost.

Western companies are again supplying technology and capital to build this capacity, while simultaneously expanding high-cost data centers in the U.S. and Europe. This mirrors the 1970s petrochemical misallocation almost point for point. The risk is not a collapse in AI usage, but a future where compute becomes abundant, standardized, and cheap—with margins crushed accordingly.

Where the Profits Actually Go When Infrastructure Commoditizes

When a technology commoditizes, value does not disappear. It moves up the stack.

In plastics, durable profits migrated toward branded products, specialty applications, and downstream integration. The same pattern is likely in AI. The long-term winners will not be those selling raw compute, but those using AI to deliver differentiated services, control distribution, or orchestrate complex systems.

Two companies stand out as clear beneficiaries of this shift.

Uber: The Agentic Fleet Operator

Uber is positioned to become the operational layer for autonomous agents in the physical world.

While companies like Waymo or Tesla may build the vehicles, the commodity hardware, Uber owns the demand network, fleet orchestration, and regulatory relationships. Replicating a global marketplace with more than 170 million active users is far harder than building a self-driving car.

Crucially, Uber is pivoting toward a capital-light model, managing third-party autonomous fleets rather than owning the vehicles themselves. This transforms Uber from a low-margin transportation provider into a high-margin software and coordination layer.

In an agentic future, your personal AI will not drive you. It will negotiate with Uber’s fleet agent to secure transportation. Uber effectively becomes the API for physical movement.

The primary risk is commoditization of the ride itself. If robotaxi supply explodes, price per mile will fall. Uber’s challenge will be maintaining its take rate by remaining the dominant marketplace. But structurally, Uber fits the profile of a specialty application winner .

Apple: The Consumer Agent Gateway

Apple represents the opposite, but equally powerful model.

As AI agents become more autonomous, trust and privacy become scarce resources. Apple’s hardware-level privacy architecture creates a walled garden where users are comfortable letting agents act on their behalf. Owning the device in the user’s pocket, and increasingly on their face, gives Apple control over context, identity, and distribution.

Apple Intelligence is likely to evolve into a subscription-style agent layer, monetizing productivity gains across its installed base. This is classic integrated-platform economics.

The historical analogy is Exxon or Microsoft at their peaks: controlling the resource, the processing layer, and the distribution point. That integration insulates Apple from commodity price wars in chips or cloud infrastructure and allows it to capture value regardless of which models or hardware dominate underneath.

The Bottom Line

AI demand will continue to grow. Capital spending will remain enormous. But history is clear: infrastructure booms rarely deliver lasting shareholder returns.

The real winners are companies that sit above commoditized layers—those that control applications, ecosystems, and operational coordination. Uber and Apple exemplify two different, but equally durable, paths to winning in an agentic AI world.

Investors who mistake capacity growth for value creation risk repeating the same error made in plastics, petrochemicals, and countless other capital cycles.