TL;DR
The Shiller CAPE ratio stands at 38-40, the second-highest in 155 years behind only the dot-com peak of 44.19, and S&P 500 top-10 concentration exceeds dot-com levels by nearly 50%. But AI companies are massively profitable unlike their dot-com predecessors, with Nvidia alone earning $120 billion in net income and the tech sector trading at 30x forward earnings versus 50x at the 2000 peak. The resolution depends on whether $660-690 billion in annual hyperscaler capex generates returns that justify the investment, a question that cannot be answered until the infrastructure cycle produces results.
The Shiller cyclically adjusted price-to-earnings ratio for the S&P 500 stands at approximately 38 to 40, depending on the day you check. In 155 years of recorded data, the CAPE has been higher exactly once: March 2000, when it reached 44.19, one month before the Nasdaq began a decline that would erase 78% of its value over the following 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 capitalisation, nearly 50% above the dot-com peak concentration of roughly 27%. Deutsche Bank’s latest fund manager survey found that 57% of institutional investors now identify an AI valuation crash as the single greatest risk to markets. Jeremy Grantham, the co-founder of GMO who correctly called the dot-com and housing bubbles, has said there is “slim to none” chance the current AI rally does not end in a bust. These are the numbers that make the comparison to 2000 feel inevitable. They are also, by themselves, incomplete.
The case for alarm
The structural parallels between the current AI equity rally and the dot-com bubble are not superficial. They are mechanical. Market concentration has exceeded dot-com levels by a wide margin. The Nasdaq-100’s performance is dominated by a handful of companies whose valuations are predicated on AI revenue growth that has not yet fully materialised at the scale the market is pricing. Hyperscaler capital expenditure, the combined infrastructure spending of Microsoft, Google, Amazon, and Meta, is approaching $660 billion to $690 billion in 2026, a figure that represents the largest corporate investment programme in history outside of wartime mobilisation. That spending is being funded, in part, by converting human labour into AI infrastructure: Meta and Microsoft collectively cut up to 23,000 jobs while simultaneously committing to record capital expenditure, a direct transfer from payroll to data centre construction.
Bank of America’s Savita Subramanian has set a year-end S&P 500 target of 7,100, with a bear case of 5,500, and expects multiple compression as earnings growth slows in the second half of 2026. The Motley Fool identified four factors it associates with bubble conditions: retail investor euphoria, speculative capital concentration, decoupling of valuations from fundamentals, and a narrative so compelling that scepticism feels intellectually disreputable. All four are present. OpenAI’s $852 billion valuation prices a company that has never earned a profit at roughly double the market capitalisation of Coca-Cola, a company that has earned profits continuously since the 1890s. Accel’s $5 billion AI-focused fund, the largest in venture capital history, exemplifies the capital flooding into AI at the private market level. The public and private markets are reinforcing each other: venture-backed AI companies raise at extraordinary valuations, public AI companies spend at extraordinary rates to stay ahead of them, and the cycle pushes both valuations and capital expenditure higher.
The case for calm
The most important difference between 2000 and 2026 is profitability. At the dot-com peak, the technology companies driving the market were, in aggregate, destroying capital. Cisco traded at 200 times earnings. Pets.com had no earnings. The entire thesis rested on future revenue from an internet economy that, while real, was years from generating the cash flows the market was discounting. In 2026, the companies driving the AI rally are among the most profitable in corporate history. Nvidia reported net income exceeding $120 billion for fiscal 2026. Its forward price-to-earnings ratio is approximately 41, elevated but not in the same postcode as Cisco at 200. The technology sector’s aggregate forward P/E is roughly 30, compared with 50 at the dot-com peak. Apple, Microsoft, Alphabet, Amazon, and Meta generated a combined $350 billion in free cash flow in their most recent fiscal years. These are not speculative enterprises burning venture capital. They are cash-generating machines that have chosen to reinvest at historically unusual rates.
Capital Economics analyst John Higgins has made the most nuanced version of this argument. He distinguishes between a “stock bubble” and a “fundamental bubble.” The stock bubble, in his analysis, may already be deflating: the Nasdaq-100 corrected more than 10% from its February 2026 highs before recovering on trade deal optimism and strong earnings. But the fundamental bubble, the one built on actual earnings growth, is still expanding. Nasdaq-100 earnings grew 19% year over year in the most recent quarter. As long as AI-related revenue continues growing at that pace, the earnings justify elevated multiples. The bubble pops not when P/E ratios are high, but when the “E” stops growing. JPMorgan has suggested the S&P 500 could reach 8,000 if earnings momentum continues. Goldman Sachs sees a multi-year AI “supercycle.” The bull case is not that valuations are reasonable. It is that earnings growth will make today’s prices look reasonable in retrospect, the same argument that was wrong about Cisco in 2000 and right about Amazon.
The capex question
The variable that will determine which analogy holds is capital expenditure returns. Hyperscalers are spending $660 billion to $690 billion this year building AI infrastructure. Meta’s $27 billion deal with Nebius for AI cloud capacity is one transaction among dozens, each individually larger than most companies’ entire capital budgets. The question is not whether this infrastructure will be used. It almost certainly will. The question is whether it will generate returns that justify the investment at the price paid. The fibre-optic cables laid in 1999 carry today’s internet. The companies that laid them went bankrupt. The technology was correct. The financial model was not.
There are structural reasons to believe the AI capex cycle is better supported than the fibre-optic buildout. Cloud computing operates on a consumption model where customers pay for usage, providing revenue visibility that speculative fibre networks lacked. The hyperscalers building the infrastructure are also the primary consumers of it, reducing the demand uncertainty that destroyed independent fibre companies. Oracle’s $553 billion in remaining performance obligations, Microsoft’s Azure backlog, and Amazon’s AWS contract pipeline all represent committed future revenue. But committed revenue is not collected revenue, and the concentration of AI demand in a small number of large model developers and enterprise customers creates fragility. If OpenAI, the anchor tenant of Oracle’s Stargate project, were to experience financial difficulty, the ripple effect through the infrastructure financing chain would be severe. If enterprise AI adoption plateaus at the “copilot” stage without progressing to the autonomous agent workflows that justify the next order of magnitude in compute spending, the return on $660 billion in annual capex would fall below the cost of capital.
The verdict the market cannot reach
Both sides of the debate are correct, which is what makes the current moment so difficult to navigate. The bears are right that market concentration, CAPE ratios, and speculative euphoria have reached or exceeded dot-com levels. The bulls are right that the underlying companies are profitable in ways their dot-com predecessors were not. The resolution depends on a variable that neither side can observe directly: the long-term return on the hundreds of billions being invested in AI infrastructure this year. If those returns materialise, the current valuations will be seen as fair prices paid early for a genuine technological transformation. If they do not, the CAPE chart will add a second peak to match the one from March 2000, and the comparisons that feel alarmist today will feel prescient.
The Federal Reserve’s benchmark rate sits at 3.50% to 3.75%, providing less of a cushion than the near-zero rates that inflated asset prices between 2020 and 2022 but not the restrictive 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 escalation will affect corporate earnings forecasts and consumer spending. The trajectory that brought technology markets to this point, a year of accelerating AI investment, record venture funding, and corporate restructuring around artificial intelligence, has created conditions that resemble a late-stage expansion more than an early-stage bubble. Late-stage expansions can last longer than sceptics 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 capex cycle produces results, and the capex cycle has barely begun. Grantham is betting it ends badly. Goldman is betting it does not. The market is pricing in both possibilities simultaneously, which is why it has been volatile in both directions, and will remain so until the revenue either arrives or does not.
You must be logged in to post a comment Login