TL;DR
The Shiller CAPE ratio is at 38-40, the second highest in 155 years behind only the dot-com peak of 44.19, and the S&P 500 top-10 concentration exceeds dot-com levels by about 50%. But AI companies, unlike their dot-com predecessors, are massively profitable, with Nvidia alone generating $120 billion in net income and the tech sector trading at 30x forward earnings compared to 50x at its peak in 2000. The resolution hinges on whether annual hyperscaler capital generates returns that justify the $660-690 billion investment, a question that cannot be answered until the infrastructure cycle delivers results.
The Shiller’s periodically adjusted price-to-earnings ratio for the S&P 500 is around 38-40, depending on the day you check. CAPE has been higher once in 155 years of recorded data: in March 2000, when it hit 44.19, a month before the Nasdaq began a slide that would wipe 78% of its value over the next two and a half years. The ten largest companies in the S&P 500 now account for 36% to 40% of the index’s total market capitalization, up from about 50% at the dot-com peak concentration of about 27%. Deutsche Bank’s latest fund manager survey showed this 57% of institutional investors now identify a valuation crash in AI as the biggest risk to the markets. GMO co-founder Jeremy Grantham, who aptly called the dot-com and housing bubbles, said, “ever so thin“coincidentally, the current AI rally does not end with a bust. These are the numbers that make comparisons with 2000 inevitable. They are also incomplete in themselves.
Box for excitement
The structural parallels between the current AI stock rally and the dot-com bubble are not superficial. They are mechanical. Market concentration has surpassed dot-com levels by a wide margin. The performance of the Nasdaq-100 is dominated by several companies whose valuations are based on AI revenue growth that has yet to fully materialize at the scale the market is pricing in. Hyperscaler capital spending, the combined infrastructure spending of Microsoft, Google, Amazon and Meta, is approaching $660-690 billion in 2026, a figure that represents the largest corporate investment program in history outside of wartime mobilization. These costs are partially financed by turning human labor into AI infrastructure: Together, Meta and Microsoft cut up to 23,000 jobs a direct transfer from payroll to data center construction while simultaneously committing to record capital expenditures.
Bank of America’s Savita Subramanian has set a year-end S&P 500 target of 7,100, down from 5,500, and expects plenty of compression in the second half of 2026 as earnings growth slows. The Motley Fool identified four factors it attributed to bubble conditions: the concentration of retail investors, the draining of core capital. and a narrative so compelling that skepticism feels intellectually discredited. There are four. OpenAI is valued at $852 billion It nearly doubled the market capitalization of Coca-Cola, a company that had never made a profit since the 1890s. Accel’s $5 billion AI-focused fundexemplifies the flow of capital into AI at the private market level, the largest in venture capital history. Public and private markets reinforce each other: venture-backed AI companies raise extraordinary prices, public AI companies spend extraordinary prices to stay ahead of them, and the cycle drives up both valuations and capital costs.
A case for calm
The most important difference between 2000 and 2026 is profitability. During the dot-com peak, the technology companies that dominated the market were generally destroying capital. Cisco traded at 200 times earnings. Pets.com had no profits. The entire thesis was based on future income from the Internet economy, which, while real, is years of generating cash flows that the market discounts. The companies driving the AI rally in 2026 are among the most profitable in corporate history. Nvidia reported that its net income for the 2026 fiscal year exceeded $120 billion. Its forward price-to-earnings ratio is around 41, high, but not in the same zip code as Cisco, which is 200. The tech sector’s common forward P/E is about 30, compared to 50 at the dot-com peak. Apple, Microsoft, Alphabet, Amazon and Meta had a combined free cash flow of $350 billion in their most recent fiscal year. These are not speculative ventures burning venture capital. They are money making machines that have historically chosen to reinvest at extraordinary rates.
Capital Economics analyst John Higgins made the most nuanced version of this argument. He distinguishes between a “stock market bubble” and a “fundamental bubble”. According to his analysis, the stock market bubble may already be deflating: the Nasdaq-100 has corrected more than 10% since February 2026, before trade deals recovered on optimism and strong earnings. But the underlying bubble built on real earnings growth is still expanding. Nasdaq-100 revenues rose 19% year-over-year in the most recent quarter. As AI-related revenue continues to grow at this pace, earnings justify elevated multiples. Bubble pops when “E” stops growing, not when P/E ratios are high. JPMorgan suggested the S&P 500 could hit 8,000 if the earnings momentum continues. Goldman Sachs Sees Multi-Year Artificial Intelligence”supercycle.” It’s not that valuations are reasonable. It is earnings growth that will make today’s prices look reasonable in retrospect, the same argument that was wrong about Cisco in 2000 and right about Amazon.
Capex question
The variable that will determine which counterparty is appropriate is the return on capital expenditure. Hyperscalers are spending $660-690 billion on building AI infrastructure this year. Meta’s $27 billion deal with Nebius Cloud capacity for AI is an operation between dozens, each individually larger than the entire capital budget of most companies. It is not about whether this infrastructure is used or not. It almost certainly will. The question is, will it deliver a return that justifies the investment for the price paid? Fiber optic cables laid in 1999 carry today’s internet. The companies that installed them went bankrupt. The technology was right. This was not the financial model.
There are structural reasons to believe that the AI capex cycle is better supported than the fiber-optic structure. Cloud computing works on a consumption model where customers pay for usage, providing revenue visibility that speculative fiber networks lack. Infrastructure-building hyperscalers are also its primary consumers, reducing the demand uncertainty that destroys independent fiber companies. Oracle’s $553 billion in remaining performance commitments, Microsoft’s Azure backlog and Amazon’s AWS contract pipeline reflect future revenue. But invested revenue is not accumulated revenue, and the concentration of AI demand in a small number of large model developers and enterprise customers creates fragility. If OpenAI, an anchor tenant of Oracle’s Stargate project, were to experience financial distress, the ripple effect through the infrastructure funding chain would be severe. Enterprise AI Adoption Plateaus “co-pilotWithout moving to autonomous agent workflows that justify the next scale of computing costs, the $660 billion annual return on capital will fall below the cost of capital.
A judgment beyond the reach of the market
Both sides of the debate are correct, which makes the current moment difficult to manage. The bears are right that market concentration, CAPE ratios and speculative euphoria have reached or surpassed dot-com levels. The bulls are right that mainstream companies are profitable in a way that their dot-com predecessors were not. The resolution hinges on a variable neither side can directly observe: the long-term return on the hundreds of billions invested in AI infrastructure this year. If these returns materialize, current valuations will be considered fair prices paid early for true technological transformation. If they don’t, the CAPE graph will add a second peak corresponding to March 2000, and today’s alarming comparisons will seem advanced.
The Federal Reserve’s benchmark rate is 3.50% to 3.75%, providing less cushion from the near-zero rates that inflate asset prices between 2020 and 2022, but not the cap rates that typically trigger corrections. Section 122 tariffs of 10% to 15% on a range of imports expire on July 24, 2026, and their renewal or increase will affect corporate earnings forecasts and consumer spending. The trajectory that brought the technology markets to this pointA year of accelerating AI investment, record venture funding and corporate restructuring around AI has created conditions that look more like a late-stage expansion than an early-stage bubble. Late-stage expansions may take longer than skeptics expect. They also end more abruptly than optimists imagine. The honest answer to whether AI stocks are in a bubble is that the question cannot be answered until the capital cycle delivers results and the capital cycle has barely begun. Grantham bets it will end badly. Goldman is betting it doesn’t. The market is simultaneously pricing in both opportunities, so it has been volatile in both directions and will remain so until the return either comes or doesn’t come.
