What's with all these sophisticated investors coming out and calling artificial intelligence (AI) a bubble?

I counted three alone in the past week:

It’s almost as if you risk losing your seat at the smart-money table unless you predict an imminent crash.

Their argument usually begins with the same observation: AI stocks have soared a lot.

They have. The mistake these “experts” are making is ignoring why AI stock prices have surged: The fundamentals underneath them.

Thinking AI is a bubble can cost you a lot of money. This is clearly the megatrend of the decade. Exit it too early, and you'll leave years of returns on the table.

Today, I want to debunk the six biggest fears about AI being in a bubble…

• Fear #1: Where’s the revenue?

Skeptics say this is a replay of the dot-com bubble, when companies with little or no revenue soared to unreasonable highs.

Just look at Anthropic, the maker of Claude.

Three years ago, it earned its first dollar. By the end of 2024, its annual recurring revenue reached $1 billion. By April of this year, it had soared to roughly $34 billion. Then it jumped to $45 billion by May.

Let me put that explosive growth into perspective.

Palantir Technologies (PLTR), Snowflake (SNOW), and Databricks each spent about a decade building their businesses.

Anthropic added revenue equal to their combined total in just the last month.

• Fear #2: Nobody uses this stuff.

It's dark fiber all over again, right?

Telecom companies spent enormous sums burying fiber during the internet boom. When the bubble burst, an estimated 95% of that fiber sat unused for years.

So perhaps today’s companies are building data centers full of chips that nobody needs. There’s one problem with that comparison: There are no dark GPUs.

Nvidia (NVDA) cannot make enough chips to satisfy demand.

Alphabet (GOOGL) says some of its seven- and eight-year-old TPUs still operate at 100% utilization. And speaking of how nobody uses this stuff... ChatGPT serves more than 900 million people every week.

AI is the fastest-adopted consumer technology in history.

• Fear #3: These GPUs will be obsolete in two years.

Companies are spending billions of dollars on Nvidia chips. But new models arrive every year.

Therefore, today’s hardware will soon become outdated, forcing companies into an endless replacement cycle. That would make the economics of AI far less attractive.

Well… the opposite is happening.

Older GPUs are becoming more valuable because the world remains desperately short of computing power.

The Nvidia A100 launched in 2020. Six years later, these still sell for nearly the same price as when they launched. Microsoft (MSFT) continued operating cloud machines based on Nvidia’s V100 until September 2025. That chip launched in 2017.

An old chip is still better than no chip at all.

• Fear #4: This level of spending can't continue.

Big tech will splurge $750 billion on AI infrastructure this year alone.

Skeptics argue this spending boom looks a lot like the excesses of the dot-com bubble.

But look closer at who's writing the checks.

In the dot-com boom, Global Crossing never recorded a profitable year. WorldCom collapsed under $30 billion worth of debt and an $11 billion accounting fraud. Many telecom companies borrowed heavily to lay fiber before they had durable customers or cash flow.

Today’s biggest AI spenders are Microsoft, Alphabet, Amazon (AMZN), and Meta Platforms (META).

They’re the richest and most profitable companies ever created. And they’re funding most of the AI buildout out of pocket.

• Fear #5: The valuations are crazy.

AI infrastructure stocks have been dominating the charts for the last three years. Many have already handed investors four-digit returns.

Surely the valuations must be insane? Nope.

Look at Nvidia, the poster child of the AI boom. If AI speculation had completely detached from fundamentals, Nvidia should trade at one of the highest valuations in its history.

During 2021 and 2022, Nvidia traded near 70X forward earnings. Today, its P/E ratio sits in the low 20s.

A stock can climb 1,000% and become cheaper when profits compound faster. Price tells you how far a stock has climbed. Valuation tells you what investors are paying for the profits underneath it. And valuations aren’t crazy.

• Fear #6: We’re building too much.

Every major technology boom eventually produces excess capacity.

Railroad companies laid too much track. Telecom companies buried too much fiber. Excess capacity drove prices down and helped trigger the bust.

AI will likely experience its own period of overbuilding and digestion. But the current constraints make it difficult to get there soon.

The world is short two essential inputs: watts and wafers.

In Northern Virginia, data-center developers can wait years for a grid connection. Large transformers carry lead times ranging from 80 to 210 weeks. Gas turbines are sold out well into 2028 and 2029.

Some companies already have GPUs that cannot be switched on because the electricity is unavailable.

Then come wafers. Taiwan Semiconductor (TSM) produces most advanced AI chips and has little interest in flooding the market with excess capacity.

Power shortages slow data-center construction. Chip shortages limit what goes inside them. These constraints stretch the cycle and force demand to wait for supply.

An overbuild requires companies to build faster than customers can absorb capacity. For at least the next two or three years, the physical bottlenecks remain too severe.

• Write down these six words…

“Pessimists sound smart. Optimists make money.”

It’s a line I live and invest by.

Every megatrend attracts skeptics. They call it overhyped. Overpriced. A bubble waiting to burst.

And every now and then, they’re right.

But wealth is built by people who can see what might go right. The optimists who stay the course are always the ones who make the most money.

My research suggests the AI boom has years left to run. We’re investing accordingly.

Stephen McBride Chief Analyst, RiskHedge