AI Powers Historic Stock Rally Even As AI Math Becomes "Hallucinatory"

These are the good old days. The only question is how long they will last. AI is powering an historic rally in stocks. And AI skeptics appear to have largely surrendered to the conventional wisdom: don't fight the tape, the trend is your friend. 

But, to the extent there are lingering concerns, they are not if money will be made, about how. Because some of the math underpinning the rally doesn't make sense from an historical perspective: competition lowers prices, such as those for AI usage tokens, whose cost has become so exorbitant that even very deep-pocketed companies like Microsoft are rationing them. The prudent course seems to be: enjoy the ride, but keep an eye for inevitable changes to the underlying economics. JL

Jack Pitcher and Andy Kessler report in the Wall Street Journal:

AI chip stocks powered the S&P 500 up 16% across April and May, a surge matched only four times since 1950. This is driving an historic stock rally, sending shares of tech companies soaring that recalls the dot-com boom. Wall Street—and history—suggest the gains could keep coming: fund managers have reduced their cash levels by the most in a month since 2024, holding significantly more stocks than their benchmarks. Skeptics note there may be 47 million programmers in the world but very few have a budget of $100,000 to spend on AI tokens. Token prices have risen 65% since February. But shortages don’t last, meaning over the next five years we’ll see token costs decline another 90%, maybe 99%. The future AI math is more like 470 million coders with $1,000 of tokens. Even $100. The rosiest AI demand forecasts will be exceeded, but at much lower prices.

The growing frenzy over memory chips is driving a historic stock rally, sending shares of technology companies soaring in ways that recall the heights of the dot-com boom. Wall Street—and history—suggest the gains could keep coming. 

Chip stocks powered the S&P 500 up 16% across April and May, a two-month surge matched only four other times since 1950, according to Dow Jones Market Data. The index was higher six months later each time, by a median 17%.

Skeptics remain, with some troubled by the sheer scale of the moves, which span from Seoul to Boise, Idaho. Micron Technology, an Idaho-based memory-chip maker, has increased roughly 10-fold in value over the past 12 months to a market capitalization above $1 trillion. Samsung is up around 465% in South Korea over that period. The PHLX Semiconductor Index just posted its strongest performance through the first 100 trading days of any year on record.

Even some of the companies involved recall the tech-bubble era: Intel has tripled in 2026 to its first record since 2000. Cisco Systems and Qualcomm have both surged around 50%.

Yet many on Wall Street are riding the wave, lifting their year-end targets for the S&P 500 in defiance of the Middle East conflict and rising inflation. Analysts at Goldman Sachs just raised their year-end target for the index to 8000 from 7600, implying a 5.5% rise over the rest of the year on top of the 11% gain since the end of December.

Some analysts said the moves are justified by the blockbuster profits that chip makers have posted in recent weeks and expect the spectacular spending on chips to grow as the artificial-intelligence build-out progresses.

“The AI train is moving forward,” said Joe Tanious, chief investment strategist for North America at Northern Trust Asset Management. “You don’t necessarily want to try to stop it or sit on the sidelines.” 

Even doubters have grown reluctant to bet against an index that has just posted nine consecutive weekly gains. In late 1996, Federal Reserve Chairman Alan Greenspan famously warned of “irrational exuberance” in the stock market. Over the following three years, the Nasdaq Composite Index tripled.

George Vanderheiden, once a Fidelity portfolio manager with one of Wall Street’s best performance records, shunned tech stocks in the late 1990s in a bet that the bubble would soon burst, loading up on Treasurys instead, hoping to protect his investors. Instead, performance slumped and he retired in February 2000, scribbling a message on his office whiteboard: “Tulip bulbs for sale”—a reminder of 17th-century speculation in the Dutch flower market.

“You try to manage the portfolio for risk,” he told The Wall Street Journal at the time. “But the market hasn’t rewarded you for it. The market has no fear.”

AI fervor has been powering broad market gains since the release of ChatGPT in late 2022, but this year, the trade has gained even more steam. Advances in AI models’ ability to write code have wowed software engineers and convinced companies and investors that the technology will be useful and transformative. 

Concern around whether AI companies with enormous expenses could eventually turn big profits has eased accordingly. Anthropic, which is expected to go public this year, defied skeptics by posting explosive sales growth and its first operating profit in the second quarter—powered by coding tools that have become popular with corporate clients. Anthropic just closed a funding round at a $965 billion valuation. “We are squarely in the acceleration phase of the AI era,” said Kevin Shea, senior equity strategist at BNY Wealth. “You take a look at the revenue growth from some of these large language models, it’s faster than anything we’ve seen before.” 

Indeed, strong corporate earnings and upbeat forecasts for the year ahead are underpinning the advance. Companies in the S&P 500 traded Friday at around 22.5 times their projected earnings over the next 12 months, a slightly lower valuation than at the start of the year, because earnings projections have grown significantly. 

Micron Technology is up roughly 10-fold over 12 months to a market capitalization above $1 trillion. Todd Meier for WSJ

Jason Pride, chief of investment strategy and research at the wealth-management firm Glenmede, estimated that 75% to 80% of the rally over the past two months “has been fundamental in nature,” fueled by legitimate hopes for a resolution to the Iran conflict and excellent earnings, particularly among the tech companies that have led the market gains. 

It is only over the past week or so, he said, that the rally has become a little bit stretched, with investors trying to chase returns. Going forward, he added, there is every reason to think that stocks can keep on rising, if not at their recent breakneck pace.

The biggest threat to the market rally might be inflation, said Angelo Kourkafas, senior global strategist at Edward Jones.

Consumer prices have climbed much faster than expected this year, largely because of energy-supply disruption caused by the war with Iran but also the overwhelming demand for AI-related hardware including memory chips.

If the Fed moves toward raising interest rates, the yield on the 10-year U.S. Treasury note could potentially climb back up toward 5%. That, in turn, could “create some more indigestion or volatility,” Kourkafas said.

For now, investors appear more concerned with how they can get as much stock exposure as possible than they are with bracing for a bursting bubble. Recent Bank of America surveys show that fund managers just reduced their cash levels by the most in a month since 2024 and are holding significantly more stocks than their benchmarks.

“Who knows what’s going to happen in three, four, five years from now with these data centers, but at least in the immediate term…it doesn’t seem like it’s going to slow down anytime soon,” Northern Trust’s Tanious said. 

IPO mania has begun, and nothing kickstarts initial public offerings like spreadsheets flashing green to incite the crowd. SpaceX’s recent S-1 filing promoted an “actionable total addressable market” of $28.5 trillion, which includes $23 trillion from enterprise artificial-intelligence applications. Woo-hoo! You can hear the gears in investors’ brains grinding: With just 5% of the market, they’ll be a trillion-dollar company! The entire U.S. economy produces $31 trillion a year.

This is on the back of AI chip maker Cerebras’s May IPO, with the stock almost doubling on opening day. Andrew Feldman, Cerebras’s CEO, did more flimsy AI math. “There are 47 million software engineers in the world. If each one uses $100,000 of tokens a year, that’s nearly $5 trillion from just a single use case. Do that bottoms-up and you go, ‘Holy crap.’ ”

Anthropic projects $11 billion in revenue in the June quarter, more than double from March. That’s a $44 billion run rate. If (more brain-gear grinding) sales keep doubling quarter to quarter, even my old HP 12c calculator tells me that Anthropic will have a trillion-dollar run rate by September 2027. Of course, that’s another hallucination.

Still, this kind of thinking is driving investment in AI data centers, some $700 billion this year and $7 trillion (so says McKinsey) by 2030. Many are circular deals, with debt and off-balance-sheet funny accounting. Even that “Shark Tank” guy from the ping-pong movie who got $15 million from Sam Bankman-Fried and FTX is building data centers.

Don’t get me wrong, AI is really the greatest thing since sliced bread, but I’ll repeat, we don’t need 100 gigawatts of toasters.

Does history rhyme? Here’s a story filled with ghosts of Wall Street’s past. In 1997, Gary Winnick’s Global Crossing invested a few hundred million dollars to lay fiber optic cable under the Atlantic. At the time, undersea T-1 lines (which transmit data at 1.5 million bits per second) cost between $10,000 and $20,000 a month. Their AC-1 undersea fiber could handle 40 gigabits per second, or more than 10,000 T-1 lines, which could easily generate more than $100 million a month. Spreadsheets flashed green!

Salomon Brothers and Merrill Lynch took Global Crossing public in 1998, raising $400 million. The stock peaked at $55 billion in value. Global Crossing borrowed heavily to build out 100,000 miles of fiber. Mr. Winnick paid more than $60 million for a home in Bel Air. Heady times.

But it worked only with $10,000-a-month pricing. The world isn’t static. AT&T and MCI created a joint venture to lay their own undersea cables. Prices plummeted. By 2002 the cost of undersea T-1 lines dropped to $1,000 a month, down more than 90%. Spreadsheets flashed red. After reporting fake revenue that Arthur Andersen blessed, Global Crossing filed for bankruptcy.

As for AI, there may be 47 million programmers in the world, but very few have a budget of $100,000 to spend on tokens. Microsoft recently canceled Anthropic Claude licenses, and Uber has already blown through its 2026 AI budget.

Meanwhile, Nvidia chips are hard to find. Memory prices, which Moore’s Law says should drop 30% a year, have instead exploded upward. Last week Micron and SK Hynix joined Samsung in the $1 trillion stock gang. Token prices—meaning what it costs to use AI models—have risen 65% since February, after dropping 90% since ChatGPT was released.

But shortages don’t last. Competition plus new technology means that token price declines will return as the norm. My gut says over the next five years (the length of Global Crossing’s existence), and likely sooner, we’ll see token costs decline another 90%, maybe 99%. The future AI math is more like 470 million coders with $1,000 of tokens. Even $100.

Watch prices. Everything Global Crossing forecast in undersea demand—per bit—happened and more, but at way lower prices. Same for the overall internet, which grew and grew in demand, but not at 1999 prices that caused flashing green spreadsheets.

It’s starting. New training algorithms show up almost daily that promise to use less memory and less compute and have AI models learn 10 times as fast. This is the natural evolution of technology, not price increases. DeepSeek out of China recently cut prices of the company’s open-source AI model by 75%. Running certain benchmarks, OpenAI costs 12 times as much and Anthropic 19 times.

There were no recessions in bits, only in price per bit. With tokens, the rosiest AI demand forecasts will be exceeded, but at much lower prices. Today, investors are enjoying the ride. This AI rocket ship is flying high. Why get off? But be warned that many spreadsheets go from green to red in a hurry when prices drop.