The Stocks of These 'toll Takers' Could be the AI Boom's Next Big Winners

Just like memory went from an afterthought to AI's biggest bottleneck, the next boom could be in stocks of companies collecting recurring revenue from AI workloads

The next boom could be in stocks of companies collecting recurring revenue from AI workloads.

What can investors do to find the next big winners in artificial intelligence? They need to spend less time focusing on what has already worked and more time asking what the market will be talking about 18 months from now.

As the latest boom was starting back in January 2025, almost nobody was talking about memory as the next great AI trade.

Investors were fixated on Nvidia (NVDA), GPUs and the companies building data centers. At the time, memory still looked like the same brutally cyclical semiconductor business that had burned investors repeatedly.

That turned out to be a costly assumption. Over the next 18 months, memory went from an afterthought to one of the AI boom's biggest bottlenecks. Apple CEO Tim Cook described the memory shortage as a "100-year flood." And memory stocks reflected this, with shares of Sandisk $(SNDK)$ and Micron Technology $(MU)$ surging more than 1,000% in the last 18 months.

The lesson goes beyond memory stocks, however. The best investments rarely feel obvious before they happen. Energy stocks were toxic to most investors before their 2022 boom. Nobody wanted bitcoin or precious metals before their runs in 2023 and 2024. And memory looked like yesterday's cyclical trade before AI transformed its supply-and-demand picture.

But by the time price confirms a structural trend, the easy money has usually been made. And I think Wall Street is missing another obvious opportunity.

This isn't the dot-com bubble

The market is obsessed with how much Big Tech is spending on AI.

Bank of America expects Microsoft $(MSFT)$, Amazon.com (AMZN), Alphabet $(GOOGL)$, Meta Platforms (META) and Oracle $(ORCL)$ to spend over $2 trillion on chips, servers, networking equipment, power, cooling and data centers over the next two years. Those numbers are enormous and invite comparisons to the dot-com bubble.

But the comparison misses a crucial difference.

During the late 1990s, companies built massive amounts of telecommunications infrastructure based largely on forecasts of future internet demand. WorldCom, Global Crossing and Qwest spent billions laying fiber-optic cable, building long-haul networks and expanding data-center capacity in anticipation of this demand.

The problem was that infrastructure supply was growing far faster than customer demand. Telecom companies assumed internet traffic would continue exploding, so they raced to build nationwide and transoceanic fiber networks before the customers and revenue were there. By the time the bubble burst, the industry was sitting on huge amounts of unused "dark fiber," which collapsed prices and crushed heavily indebted operators.

Today's AI buildout looks completely different.

The demand is already there

Microsoft, Amazon, Alphabet and Oracle are not building AI infrastructure because they anticipate demand will materialize. Instead, customers are reserving compute capacity years in advance, often before the facilities required to provide it have even been completed.

The proof is sitting in their order books. Microsoft recently reported $627 billion in remaining commercial performance obligations, nearly double the figure in the prior year, with an average contract duration of roughly 2.5 years. Alphabet's Google Cloud backlog has reached $240 billion, also more than doubling year over year, as existing customers are consuming cloud services more than 30% faster than their original commitments. Amazon saw AWS grow at its fastest pace since the pandemic boom, while its AI business and custom-chip business surpassed $25 billion annual revenue run rates.

The key distinction is that much of the late 1990s telecom buildout created capacity in anticipation of future customers. Today's AI infrastructure is being built because many of the customers are already waiting.

In the near term, the market has punished many of these companies for their spending plans. I expect that to reverse over the next 18 months.

And this violent "re-rating" will inevitably benefit what I call the "AI toll takers."

Where I'd put money to work

If this thesis is right, the obvious beneficiaries are the companies collecting recurring revenue from AI workloads rather than simply supplying the hardware.

Amazon is my favorite pure-play toll taker. Every time a company trains a model, deploys an AI agent or runs inference through its cloud, Amazon makes money. Alphabet is another: Google Cloud is growing rapidly, its backlog is expanding and AI is increasingly being monetized across search, cloud and enterprise software. And Microsoft remains one of the highest-quality ways to own the theme. Azure sits at the center of enterprise AI adoption, while Microsoft can monetize AI through cloud infrastructure, Office and its broader software ecosystem.

These companies already carry trillion-dollar valuations, but their AI businesses are also growing at rates that can materially change their earnings power over the next several years. If that happens, today's valuations may prove far less demanding than they appear.

There are also less obvious toll takers beyond Amazon, Alphabet and Microsoft.

One in particular sits in the middle of nearly every advanced AI-chip supply chain. It gets paid regardless of whether Nvidia, Advanced Micro Devices $(AMD)$, Apple $(AAPL)$ or another chip designer ultimately wins. I recently started buying it in my own portfolio. I'll break down that position, along with the other AI toll takers on my buy list, in my next "Let's Analyze" newsletter on Substack.

Your goal as an investor should be to identify narratives before the masses do, just like the investors who identified the memory-trade bottleneck before everyone else did. I think the next big opportunity will come from identifying who collects the toll after all this infrastructure gets built.

Robert Ross is the founder of TikStocks and author of "A Beginner's Guide to High-Risk, High-Reward Investing" (Adams Media, 2022). A former chief equity analyst at Mauldin Economics, Ross writes the investment newsletter Let's Analyze on Substack and hosts the weekly "Room to Run" podcast. Disclosures: Ross owns Amazon, Microsoft, Alphabet and Apple shares.

More from Robert Ross:

SpaceX's stock is down nearly 50% from its high. Why it's still not a bargain.

Forget oil. A surging El Niño could kill Fed rate cuts - and these stocks stand to win.

-Robert Ross