During a gold rush, sell shovels — and AI is the current gold rush

In 1849, tens of thousands of prospectors flooded California with one dream: find gold, get rich. Most didn't. But the merchants who sold them pickaxes, shovels, and denim jeans? They became some of the wealthiest people in America. Levi Strauss built an empire not by mining, but by outfitting the miners.

Fast forward 175 years. We are living through a new kind of gold rush — Artificial Intelligence. Billions of dollars are flooding into AI models, AI startups, and AI applications. Every company wants a piece of it. And just like 1849, most will not strike gold. But the companies selling the "shovels" — the semiconductors, the lithography machines, the high-bandwidth memory, the storage — those are the ones building durable, essential businesses.

"During a gold rush, don't mine for gold — sell shovels."

This article maps the modern AI gold rush, identifies who the real "shovel sellers" are, and walks through three case studies — Micron (MU), ASML, and Western Digital / SanDisk — to understand which infrastructure plays have the strongest investment thesis.

The scale of the AI gold rush

This is not a typical tech cycle. The numbers are staggering. Hyperscalers — Amazon, Microsoft, Google, and Meta — collectively committed to spending over $300 billion on AI infrastructure in 2025 alone. That capital has to go somewhere. It flows into data centres, chips, memory, power systems, and networking. Unlike AI software companies (where profits are speculative and competition is brutal), the infrastructure companies are getting paid regardless of which AI "wins."

Who are the shovel sellers?

The AI supply chain has several layers, each with different risk and reward profiles. Here's a simplified breakdown of where the "shovel" opportunities sit:

1) EUV Lithography

The machines that print chips. Without them, there are no chips. ASML is the only company on earth that makes EUV machines.

2) Memory (DRAM / HBM)

AI models need massive amounts of fast memory to run. High-bandwidth memory (HBM) is the critical bottleneck today.

3) Storage (NAND Flash)

Training data, model checkpoints, and inference outputs all need storing. NAND flash is the default medium.

4) GPU / Logic Chips

NVIDIA dominates, but TSMC makes every chip. The logic layer is hyper-competitive — harder to play safely.

Case study 1: ASML — the only shovel manufacturer in town

If you want to build a modern AI chip, you need ASML's Extreme Ultraviolet (EUV) lithography machines. There is no alternative. These machines — which cost upwards of $200 million each — use light with a wavelength of just 13.5 nanometres to etch circuits onto silicon with atomic-level precision. It took ASML over two decades and €6 billion in R&D to build one. No competitor has come close.

This is the ultimate "pick-and-shovel" business: everyone racing to make cutting-edge chips — TSMC for Apple, TSMC for NVIDIA, Samsung for everyone else — must buy from ASML. The company has essentially created a legal, structural monopoly in the most critical step of the chip-making process.

The key risk with ASML is geopolitical. The Dutch government, under US pressure, has restricted ASML from exporting its most advanced EUV machines to China. This removed a meaningful revenue stream. Yet even accounting for this, ASML's backlog as of early 2025 stood at over €36 billion — essentially guaranteeing revenue for years ahead. For long-term investors, ASML is as close to an infrastructure monopoly as the stock market offers.

Case study 2: Micron Technology (MU) — the HBM sleeper

Micron is one of only three companies in the world that makes DRAM memory (along with Samsung and SK Hynix). But the more interesting story is what is happening with High-Bandwidth Memory, or HBM. HBM is a special type of DRAM stacked in 3D layers and placed directly beside the GPU die. It allows AI chips to access data dramatically faster than standard memory — which is why every H100 and Blackwell GPU is packed with it.

Demand for HBM has exploded. NVIDIA's GB200 systems require enormous quantities of HBM3e. SK Hynix was first to market and captured most of NVIDIA's initial orders. But Micron has been aggressively ramping HBM3e production, and its memory is now qualified for NVIDIA's systems. This positions Micron as a significant beneficiary — a company selling an essential commodity that happens to be in extreme, structural undersupply.

The key nuance with Micron is the cyclicality of the memory industry. Memory is a commodity — prices fluctuate sharply with supply and demand. In 2022–2023, memory prices crashed as the post-COVID inventory glut unwound. Micron's revenue fell by nearly 50%. Then the AI cycle hit, HBM demand exploded, and Micron's fortunes reversed dramatically. This is the risk: buying a cyclical company at the wrong point in the cycle can be painful. But for investors who understand the cycle — and buy when the industry is in the trough — the returns can be exceptional.

Case study 3: SanDisk / Western Digital — NAND's quiet AI tailwind

SanDisk, now a brand of Western Digital (WDC), is one of the world's largest manufacturers of NAND flash storage — the technology inside SSDs and USB drives. At first glance, NAND feels less "AI" than HBM or EUV. But consider what training a large language model actually requires: petabytes of training data, stored on fast SSDs. Constant model checkpointing. Continuous inference-result logging. Data centres are effectively massive storage farms, and AI has dramatically increased how much storage is needed per server rack.

Western Digital split its HDD and flash businesses in 2024, with the NAND/flash business emerging as the more AI-relevant entity. Like Micron, WD is a cyclical play — NAND prices crashed in 2022–2023 and have been recovering since. The recovery thesis: as AI-driven storage demand grows structurally, NAND suppliers benefit from a persistently tighter supply-demand balance than in previous cycles.

Comparing the three "shovels"

ASML

Structural monopoly on EUV lithography. Every leading-edge chip requires their machines. Massive backlog, pricing power, and no credible competitor. Premium valuation — you pay for certainty.

MU

HBM3e ramp positions Micron for the AI memory supercycle. Cyclical industry means volatile earnings, but the AI-driven demand shift is structural. Higher risk, higher potential reward.

WDC / SNDK

AI's quiet beneficiary via exploding data storage demand. NAND price recovery adds a near-term catalyst on top of the structural AI tailwind. Highest cyclicality of the three.

The historical lesson: infrastructure always wins in a mania

History repeatedly validates the shovel thesis. During the dot-com boom of the late 1990s, thousands of internet companies went public. Most went to zero. But Cisco — which sold the routers and networking equipment the internet ran on — saw its revenues grow from $1.2 billion in 1995 to over $18 billion in 2000. Infrastructure. During the shale oil boom of the 2010s, many drillers went bankrupt when oil prices crashed. But pipeline companies kept collecting their fees regardless of oil price. Infrastructure.

The pattern is consistent: in a speculative boom, the entities that provide the essential, non-optional inputs to the boom tend to generate more durable profits than the boom participants themselves. The gold miners gamble. The shovel sellers get paid.

The key risks to the shovel thesis

No investment thesis is complete without understanding the risks. The shovel strategy has three main vulnerabilities. First, valuation: ASML, for instance, trades at a premium multiple. If AI investment slows or a recession hits, even infrastructure companies can see their stock de-rate significantly — their business is still fundamentally tied to demand. Second, cyclicality: memory and NAND companies like Micron and Western Digital are exposed to commodity cycles. A supply glut — even in an AI boom — can crush margins. Third, geopolitics: export restrictions on semiconductor equipment and chips to China introduce real revenue uncertainty for ASML and others.

The mitigation? Diversify across the supply chain. Hold ASML for the monopoly moat. Hold Micron for the HBM cycle. Hold storage plays for the data accumulation thesis. No single position should bet the portfolio on one outcome.

Final thought: the shovels are already selling

The AI gold rush is real, it is enormous, and it will produce massive winners and losers. The companies building AI models are competing furiously — and most will be disrupted within a decade. But the companies making the chips, printing the silicon, storing the data, and powering the data centres? They get paid by everyone, regardless of who wins.

The 1849 miners needed Levi Strauss's denim. Today's AI companies need ASML's EUV machines, Micron's HBM, and Western Digital's flash storage. The insight hasn't changed in 175 years — and neither has the opportunity.

As always — do your own research, understand the cycles, and never invest more than you can afford to lose. But if you're looking for places where AI spending is near-guaranteed to flow? Start with the shovels.