The silicon chokepoint: why the 2026 AI chip shortage is different — and how to invest in it

If you’ve been following markets lately, you’ll know that AI chips have become some of the most valuable assets in the global economy. But as we move through March 2026, the conversation is changing. It’s no longer just about who can build the smartest AI. It’s about who can actually get the hardware to run it. And that requires something like silicon!

For investors, this has created a strange situation.

Some experts think we’re in a giant bubble that is about to pop. Others see a “gold rush” that could last for a decade.

To understand what’s really happening, we need to look at the maths behind the chips and the companies that make them.

The Michael Burry warning: An “earnings bubble”?

Michael Burry is the investor famous for predicting the 2008 housing crash. Recently, he’s turned his attention to AI, and he’s not happy with what he sees. His main concern is depreciation.

In simple terms, depreciation is how a company accounts for the loss of value of its equipment over time. If a company buys a $40,000 AI chip, they don’t list that full cost as an expense all at once. Instead, they spread the cost out over several years. Most big tech companies—like Microsoft and Google—are spreading the cost of their AI chips over five or six years.

Burry argues this is a “trick.” He believes that in the fast-paced world of AI, a chip bought today will be “old news” and basically useless in just two or three years because newer, faster chips are coming out so quickly. By pretending the chips last six years, these companies are making their yearly profits look much higher than they actually are. Burry thinks that when these companies finally admit the chips are out of date, they will have to report billions of dollars in losses.

The counter-argument: Why old chips are still gold

While Burry’s maths makes sense on paper, the real world is telling a different story. If these older chips (like the Nvidia H100) were truly losing their value, it should be very cheap to rent them.

But it isn’t.

As of March 2026, the demand for AI is so high that people are still lining up to rent “old” H100 chips. On major cloud platforms, renting an H100 still costs between $3 and $4 per hour. While this is lower than the $10 per hour we saw two years ago, it is still very expensive for “outdated” tech.

Why is this happening?

It’s because there simply aren’t enough new chips to go around.

If you can’t get your hands on Nvidia’s newest “Vera Rubin” chip, you have no choice but to keep using the older ones. This means the older chips are still making a lot of money for the companies that own them, which suggests Burry might be too early with his “bubble” prediction.

Nvidia’s trillion-dollar promise

The reason the shortage is so bad was made clear at Nvidia’s GTC conference earlier this week.

CEO Jensen Huang announced that the company now has a $1 trillion backlog of orders. This means customers have already asked for $1 trillion worth of chips that Nvidia hasn’t even built yet.

Nvidia also showed off its newest system, named after the astronomer Vera Rubin. These chips are designed for a new phase of AI called “inference.” In the past, companies spent most of their time “training” AI (teaching it). Now, the world is moving to “inference”—actually using the AI to answer questions and run autonomous agents 24/7. This requires a massive, constant supply of power and chips.

Nvidia is effectively sold out of these new systems through the middle of 2027.

The $650 billion arms race

And while investors worry about the cost, the “Big Four” tech giants (Amazon, Microsoft, Google, and Meta) are not slowing down.

In fact, they are speeding up. Together, these four companies plan to spend about $650 billion on AI infrastructure this year alone:
• Amazon is spending $200 billion to build data centres and even its own satellites.
• Microsoft is spending $145 billion, largely because so much of its future business depends on its partnership with OpenAI.
• Google is doubling its spending to nearly $185 billion.
This is a classic “arms race.”

If one company stops buying chips, they risk falling behind in the most important technology of the century. They’re willing to overspend today to make sure they aren’t extinct tomorrow.

The ultimate play: Why ASML is the safest bet

If you’re a retail investor looking for a way to play this shortage without betting on which AI model wins, the smartest move might be looking at ASML.

ASML is a Dutch company that has a total monopoly on the world’s most important machine: the EUV Lithography machine. This machine uses a specialised light to “print” tiny patterns onto silicon. Without this machine, it is physically impossible to make a high-end AI chip.

Here is why the case for ASML is so strong in 2026:
• The only supplier: If Nvidia, Intel, or Samsung want to make a 2-nanometer chip (the tiniest and fastest kind), they must buy a machine from ASML. There is no second choice.
• Huge backlog: ASML currently has a backlog of nearly €39 billion in orders. They literally cannot build these machines fast enough to meet demand.
• High-NA technology: ASML just started shipping its newest “High-NA” machines. These cost about $380 million each. They are so complex that they require thousands of parts from all over the world. This makes it almost impossible for a competitor to ever catch up.
• Beyond the chips: It isn’t just about the “brains” of the AI. AI also needs massive amounts of memory (HBM4). These memory chips also require ASML’s machines to build.
In short, while Michael Burry might be right that some tech companies are playing games with their accounting, the physical reality is that the world is starving for chips.

As long as that hunger exists, every single dollar spent on AI eventually flows through ASML.

They aren’t just in the game; they own the stadium.

ASML (is currently listed on two exchanges: The NASDAQ (#ASML), where it trades in US dollars, and Euronext Amsterdam (#ASML), where it trades in euros. You can, however, purchase it through a local account on our online platform (recently named South Africa’s Best Trading Platform at the Financial Mail’s Top Securities Broker Awards).