The goldmine and shovels

Everyone loves a good gold rush analogy. In the AI version, NVIDIA and the data center operators are the ones selling shovels, and they have already made their fortunes. The implication is clear: if you are building an AI company, you are the miner, and you are probably too late. It is a neat story. It is also incomplete.

The shovel sellers are real

There is no denying that the infrastructure layer has been wildly profitable. NVIDIA posted $68.1 billion in quarterly revenue in Q4 of fiscal 2026, with data center revenue alone hitting $62.3 billion, a 75% jump year over year. The company's full-year 2025 revenue reached $130.5 billion, more than doubling the prior year. The capital flowing into AI infrastructure is staggering. OpenAI's Stargate project is targeting $500 billion by 2029. Amazon, Google, Microsoft, and Meta are each committing tens of billions annually to build out data centers and compute capacity. NVIDIA itself pledged $500 billion over four years for U.S.-based AI manufacturing and infrastructure. So yes, the shovel sellers are doing extremely well. But does that mean the miners are doomed?

The gold rush analogy breaks down

The original California Gold Rush had a fixed resource. There was only so much gold in the ground, and early arrivals had a structural advantage. Once the easy deposits were claimed, latecomers were left sifting through scraps. AI does not work like that. The "gold" in AI is not a finite mineral. It is value created by solving problems, automating workflows, generating insights, and building products that people actually want to use. The surface area of problems worth solving with AI is expanding, not shrinking. Every industry, from healthcare to logistics to education, has layers of inefficiency that AI can address. New use cases keep emerging as the technology matures. In a gold rush, the resource depletes. In AI, the resource compounds.

The miners are not broke

The narrative that AI companies are burning cash with nothing to show for it does not hold up against the numbers. OpenAI's annualized revenue surged to $13 billion by mid-2025, up from $200 million in early 2023. Anthropic hit $14 billion in annual recurring revenue, growing by 1,000% in each of the last three years. Enterprise AI revenue reached $37 billion in 2025, more than tripling year over year. Private investment in generative AI reached $33.9 billion in 2024, and AI startups captured 57.9% of all global venture capital dollars in Q1 2025. That is not money chasing a dead end. That is capital following real traction. Of course, not every AI startup will survive. The market is crowded, and many companies are thin wrappers around existing models. But the same was true of the early internet, early mobile, and early cloud computing. The fact that many will fail does not mean the opportunity is gone. It means the opportunity is being sorted.

What actually matters now

If the shovel sellers have an advantage, it is because they positioned themselves at a chokepoint. NVIDIA controls the GPU supply. Cloud providers control the compute. These are real moats. But chokepoints shift. GPU marketplaces like CoreWeave and Lambda are democratizing access to compute. Open-source models are reducing dependence on a handful of providers. The cost of inference is dropping rapidly. What was a bottleneck in 2023 is becoming a commodity in 2026. The companies that will win from here are not necessarily the ones with the most GPUs. They are the ones that understand their domain deeply, have access to proprietary data, and can build products that deliver measurable value. As one investor put it, the winners will not be the most AI-native operators, but the ones who understand capital structure, incentives, and distribution.

The real lesson from the gold rush

The gold rush analogy gets one thing right: selling infrastructure during a boom is a great business. NVIDIA, AWS, and Azure have proven that beyond any doubt. But the analogy gets the conclusion wrong. It implies that the opportunity is binary, that you either sell shovels or you lose. History tells a different story. The California Gold Rush built San Francisco. It created banks, railroads, agriculture empires, and an entire state economy. The miners who failed were the ones chasing easy riches without a plan. The ones who succeeded were the ones who built something lasting. The same is true in AI. If you are building a company that depends on hype and a ChatGPT wrapper, then yes, you are probably too late. But if you are solving a real problem, with a defensible approach and a clear path to value, the gold rush is far from over. The shovels have been sold. The real building is just getting started.

References