Have you seen this stat making the rounds...

“Thanks to AI, the top 10 stocks now account for nearly 40% of the S&P 500.”

It gets repeated on CNBC every morning.

The implication is always the same: artificial intelligence (AI) is one giant bubble. And when it pops, it’ll take the whole market down with it.

That’s the wrong way to look at it. AI isn’t one trade. It’s a rolling series of waves.

Consider Nvidia (NVDA). It surged 10X in the early days of AI. Yet it’s flat since last October.

Meanwhile, a lesser-known group of AI winners is ripping higher right underneath the surface.

• The three best-performing AI stocks this year are…

Ichor Holdings Ltd. (ICHR): +223%.

AXT Inc. (AXTI): +373%.

Aehr Test Systems (AEHR): +309%.

Most investors won’t recognize them. All three are buried in the less-obvious parts of the AI supply chain.

Ichor makes the tiny plumbing systems used inside chip factories. Its valves and tubes move gases and chemicals into the machines that carve circuits onto silicon wafers.

AXT makes special “glass-like” wafers. These tiny components sit inside every high-speed optical cable that connects AI servers together.

Aehr makes the machines that "torture test" AI chips before they ship. They run each one at extreme heat and voltage for hours to weed out the ones that would fail inside a data center.

These are boring companies, but they’re just as critical to the AI buildout as Nvidia. And they’re surging because of a simple pattern I’ve been pointing out for the last two years.

• AI investing is like a python swallowing a pig.

The bulge of investor money moves slowly through the AI value chain, from one bottleneck to the next. Each time it passes through a new segment, that segment's stocks go vertical.

When the bulge moves on, the stocks that were on fire cool off, and an entirely new group takes the lead.

Here's how it’s played out so far…

In 2023, the bottleneck was Nvidia’s GPUs. You couldn’t build AI without them. Nvidia became the company selling shovels during the gold rush. Its stock surged more than 1,000% in 18 months.

In 2024, the pressure shifted to power. Training AI models requires enormous amounts of electricity. So boring ol’ utilities suddenly became AI winners. Vistra Corp. (VST) quadrupled. Constellation Energy Corp. (CEG) tripled.

In 2025, the bulge moved to memory. As AI models grew bigger and more capable, they needed massive amounts of high-bandwidth memory to move data faster. Without it, GPUs would sit idle waiting for work. Micron Technology (MU) surged 247%. SK Hynix sold out its entire 2026 production capacity by October.

Now, in 2026, the bottleneck has moved even deeper into the supply chain. We’re in the picks-and-shovels-of-the-picks-and-shovels phase. The bulge has moved from companies making AI chips to companies keeping those chipmakers running.

• Follow these two rules when hunting for the next AI bottleneck.

Rule #1: Find the choke points hidden inside the choke points.

Everyone knows AI needs chips. Few people ask what those chips need.

Before an AI server can run, the chips inside it must survive extreme stress testing. That’s a niche Aehr own.

Before those servers can talk to each other at high speeds, they need optical cables powered by laser chips. Those laser chips are built on specialized wafers, which only AXT and another company make.

And before any of those chips exist at all, semiconductor fabs need ultra-pure gases delivered with microscopic precision. That tiny market is dominated by Ichor.

Rule #2: Hunt for tiny companies with big customers.

Aehr brought in about $10 million worth of revenue last quarter. One order from an AI giant like OpenAI or Google (GOOGL) could change the entire company overnight.

Ichor supplies critical systems to Lam Research Corp. (LRCX), one of the chipmaking equipment giants. When Lam wins a multibillion-dollar equipment order from Taiwan Semiconductor (TSM)—the world’s largest chipmaker—money funnels to Ichor, too.

• Where is the bulge moving next?

Our research shows it’s moving into semiconductor equipment companies. The companies making the machines that make the chips.

Better AI requires better chips. And better chips are much harder to make.

The tiny features inside the newest AI chips are around 40,000X to 50,000X thinner than a human hair. At that scale, every new generation becomes harder to manufacture and demands far greater precision.

That means more spending on the machines inside the chipmaking factories.

Companies that make the machines that make every AI chip in the world have been a sleeper trade all year. That’s about to change.

Taiwan Semiconductor recently announced it will spend a record $56 billion on new chip factories this year, with a big chunk going toward making cutting-edge chips.

Samsung, the world’s second-largest chipmaker, is committing more than $73 billion to stay competitive in the AI chip race.

Our research suggests the back half of 2026 and the first half of 2027 could be the strongest 12-month period semi-cap companies have ever seen.

Stephen McBride

Chief Analyst, RiskHedge

PS: Right now, the best AI investing opportunities aren’t coming from the household names you’ll read about in the news. They’re coming from the overlooked corners of the market you’ll only hear about after most of the profits have already been made.

That’s why I write The Jolt. It’s my twice-weekly disruption letter where I break down the biggest megatrends—and how to get ahead of them. If you’d like to sign up, you can join our free mailing list here.