The New Standard Oil: How Google's AI Vertical Integration is Rockefeller's Playbook for the 21st Century

In 1870, John D. Rockefeller owned 4% of America's petroleum market. By 1880, he controlled 90-95% of all oil refining in the United States. His secret? He didn't just refine oil. He owned the barrels. The pipelines. The railcars. The chemical suppliers. The distribution networks. Everything.

Fast forward 145 years, and I'm watching the same playbook unfold—except this time it's not petroleum. It's artificial intelligence. And the new Rockefeller? Google.

History doesn't repeat, but damn, it rhymes.

The Rockefeller Method: Own the Stack

Let me tell you how Standard Oil actually worked, because it's weirdly relevant to understanding the AI landscape in 2025.

Rockefeller didn't just compete with other refiners on price. That's amateur hour. Instead, he integrated vertically—buying up everything in the supply chain until competitors couldn't function without going through him.

Think about that for a second. You want to compete with Rockefeller in oil refining? Great. Good luck buying barrels—he owns the barrel makers. Need to ship your oil? Hope you like paying his railcar rates. Need chemicals for refining? He bought those suppliers last month.

When you control every layer of the stack, competition becomes a choice you graciously allow others to pretend they're making.

Google's AI Stack: The Modern Standard Oil

Now look at what Google has built over the last decade:

Google is the only truly vertically integrated AI company. They can take a breakthrough from research, optimize it on custom hardware, deploy it at global scale through their cloud, and distribute it to billions of people through products they already own.

No one else can do that. No one.

The "Nvidia Tax": OpenAI's Barrel Problem

Here's where it gets brutal.

OpenAI—the company everyone thinks of when they hear "AI"—doesn't make its own chips. Neither does Anthropic. They're completely dependent on NVIDIA for the GPUs that power their models.

And it's expensive.

Now compare that to Google.

Industry analysis suggests Google obtains its AI compute at roughly 20% of the cost of those purchasing high-end NVIDIA GPUs. That's a 4-6x cost efficiency advantage at the hardware level.

This is the barrel problem. OpenAI is trying to compete in oil refining, but Google owns the barrel factory. Google owns the railcars. Google owns the pipeline.

When Your Competitors Rent Your Infrastructure

Here's the part that made me spit out my coffee:

OpenAI has started using Google's TPUs to power ChatGPT.

Read that again.

The company building the most famous AI product in the world is now renting compute from Google—its direct competitor—because NVIDIA GPUs are too expensive and Microsoft Azure can't provide enough capacity.

And Anthropic? The company behind Claude (the AI helping me write this)? They just signed a deal for up to 1 million Google TPUs—worth tens of billions of dollars.

Rockefeller would be proud. When your competitors have to buy your barrels to ship their oil, you've already won.

The Numbers Don't Lie

Let's talk about Google's newest TPU—Ironwood (7th generation):

• 10x peak performance improvement over previous generation
• 4x better performance per chip for both training and inference
• Scales to 9,216 chips in a single superpod
• 9.6 Tb/s inter-chip interconnect speed
• Access to 1.77 Petabytes of shared High Bandwidth Memory
• All-reduce operations 10x faster than Ethernet-based GPU clusters

Stacy Rasgon, analyst at Bernstein, puts it plainly: "Of the ASIC players, Google's the only one that's really deployed this stuff in huge volumes... They're the furthest along among the hyperscalers."

The Competitive Landscape: Who Has What?

See the pattern? OpenAI and Anthropic have great models. But they have no infrastructure of their own. They're renting from the big three.

It's like being the best refiner in 1885... while Standard Oil owns all the pipelines.

The Strategic Moat

Here's what vertical integration actually gives Google:

But Wait—What About Antitrust?

Here's where the Standard Oil parallel gets interesting.

In 1911, after years of litigation, the Supreme Court broke up Standard Oil into 39 independent companies. Rockefeller's monopoly was declared "unreasonable" under the Sherman Antitrust Act.

The result? Historians argue the former monopoly simply "transformed into an oligopoly of large vertically integrated companies." Two of those fragments—Standard Oil of New Jersey and Standard Oil of New York—eventually became Exxon and Mobil. Which later merged back together.

Google is already facing antitrust scrutiny. But even if regulators act, history suggests the fundamental advantages of vertical integration don't disappear. They just get... reshuffled.

The Risks: Where This Could Go Wrong

Look, I'm not saying Google is invincible. The Rockefeller playbook has vulnerabilities:

• Massive Capital Requirements: Google is spending nearly $100 billion on infrastructure. If demand slows, those billions become stranded assets. Free cash flow already dropped 61% in one quarter.
• Software Ecosystem Gap: NVIDIA's CUDA platform has 15+ years of developer tools. Google's TPU ecosystem is catching up, but it's not there yet.
• Architecture Risk: What if transformer architectures get replaced by something else? Custom silicon optimized for today's models could become tomorrow's e-waste.
• Talent Wars: OpenAI and Anthropic attract top researchers. Control over hardware doesn't automatically mean control over innovation.

But here's the thing: these are all execution risks. The strategy is sound. Rockefeller faced risks too. He still dominated for 40 years.

What This Means for Everyone Else

If you're building an AI startup right now, here's the uncomfortable truth:

You're probably renting your competitive advantage from someone who could become your competitor tomorrow.

Using Google Cloud? Google sees your usage patterns. Using OpenAI's API? You're dependent on their pricing and availability. Using NVIDIA GPUs? You're at the mercy of allocation schedules and supply constraints.

The companies that thrive in this landscape will be the ones that either:

1. Build defensible moats in specific applications (where the infrastructure provider has no interest in competing)
2. Achieve distribution advantages that raw AI capability can't replicate
3. Create proprietary data flywheels that make their products better the more they're used

In the Standard Oil era, the winners weren't the companies that tried to out-refine Rockefeller. They were the ones who found niches he didn't care about—or who got so big in their own domains that vertical integration made sense for them too.

The Bottom Line

The future of AI "won't just be decided by who builds the smartest model—but by who can run it the fastest, at the largest scale, for the lowest cost."

Right now, that's Google.

OpenAI has the brand. Anthropic has the safety research. Meta has the open-source momentum. But Google has the stack.

And in a race that's increasingly about infrastructure economics, owning the stack matters more than owning the hype.

The question isn't whether Google will dominate the AI infrastructure layer. They already do. The question is whether that infrastructure dominance translates into application dominance—or whether nimble competitors can find cracks in the foundation.

Standard Oil lasted 40 years before the courts broke it up. Even then, the fragments became some of the largest companies in the world.

Google's AI empire is just getting started.

Welcome to the new Gilded Age. The barrels are just made of silicon now.