Quora Answer: What are the Top Indicators That We are in a Technology Bubble?

The following is my answer to a Quora question: “What are the top indicators that we are in a technology bubble?  And what are the signs that the bubble is going to burst?”

The last time the world convinced itself that a transformative technology would rewrite the laws of economics, the Nasdaq fell 77 per cent.  The fibre-optic cables laid at colossal expense are still in the ground.  Mostly dark.  Mostly unused.  The companies that laid them are mostly gone.  We are doing it again.  This time with more money, more debt, and the same absence of a revenue model.  Why?  Because people are stupid, and they keep repeating the same damned mistakes because of hubris and greed.

A technology bubble has a recognisable anatomy.  It is not overvaluation alone.  It is overvaluation combined with narrative displacement.  Every bubble in history has had a story so compelling that the numbers stopped mattering.  The South Sea Company was going to monetise the trade routes of an entire continent.  The railways were going to transform British commerce.  The internet was going to make geography irrelevant.  The story was not wrong.  The valuations were.

The indicators are consistent across every bubble since the South Sea Company in 1720.  Valuations detach from earnings.  Capital concentrates in a small number of names.  Retail investors enter late and in volume.  Insiders sell.  Debt replaces equity as the preferred funding mechanism.  Revenue projections assume adoption curves that have never materialised in any prior technology cycle.  And when someone raises the question of when the investment will pay off, they are told they are missing the point.  We are at that stage.

The Valuation Picture

The Nasdaq-100 traded at 60 times forward earnings in March 2000.  That is the number most people cite as evidence of dot-com excess.  The S&P 500 forward price-to-earnings ratio in early 2026 was 23 times — stretched to the most extended level since the dot-com era, though well below the 2000 peak.  On headline earnings multiples, the AI trade is not 2000.

OpenAI, the company most associated with the AI revolution, was valued at US$730 billion in its latest funding round.  That is 56 times its 2025 revenue.  Anthropic, valued at US$380 billion, trades at 27 times revenue on US$14 billion annualised income.  Databricks is US$134 billion — 25 times its US$5.4 billion run-rate.  Some AI startups carry per-employee valuations of US$400 million to US$1.2 billion.  There is no dot-com precedent for that figure.  The metric did not exist in 2000 because nobody thought to calculate it.

Market concentration has reached levels not seen since the mid-1960s.  The ten largest companies in the S&P 500 account for 41 per cent of market capitalisation.  Those ten names drove 53 per cent of the index’s 2025 return.  Artificial intelligence, as a category, accounts for 40 per cent of the S&P 500’s total market capitalisation.  When four companies represent that much of a country’s equity market, diversification has lost its meaning.  Venture capital has abandoned any pretence of balance.  In 2025, artificial intelligence captured 61 per cent of all global venture capital — US$258.7 billion.  In 2022, the figure was 30 per cent.  The top five deals alone accounted for US$63 billion, or 25 per cent of the total.  Seventy-three per cent of AI venture capital value was absorbed by mega-deals.  A market in which the top five transactions represent a quarter of all activity is not a market.  It is a lottery with very large tickets.

The Capital Expenditure Problem

Alphabet, Microsoft, Amazon, and Meta are on track to spend US$724 billion on capital projects in 2026.  JPMorgan projects global AI capital expenditure will reach US$900 billion in 2027.  Goldman Sachs puts hyperscaler capital expenditure for 2025 to 2027 at a cumulative US$1.15 trillion — against US$477 billion for the three years prior.  Sustaining acceptable returns on capital for the hyperscalers demands an annual profit run-rate above US$1 trillion.  The 2026 consensus is US$450 billion — less than half what is needed.  Nine major technology companies spent US$600 billion on capital projects over the past year.  Goldman Sachs projects debt will finance more than a third of their AI spending by 2027.  A Wall Street Journal analysis found US$3 trillion in future AI-related commitments not yet on corporate balance sheets.

Amazon posted negative free cash flow of US$42.4 billion over the past twelve months on US$220 billion of projected 2026 capital expenditure.  OpenAI spent US$1.69 for every US$1.00 of 2025 revenue it generated.  Its projected cumulative cash burn through 2029 is US$115 billion.  The company expects to reach profitability around 2030.  Its investors, who valued it at US$730 billion, are therefore betting on a business that will consume US$115 billion in cash before it earns its first net dollar — and doing so at a valuation that assumes the profitability, when it arrives, will be extraordinary.  That is not investment analysis.  That is faith.  They are treating this like a religion.

The Misplaced Revenue Projection

The standard response to concerns about AI returns is that the technology is early.  The adoption curve will steepen.  The revenue will come.  Give it time.  The data does not support that optimism.  MIT’s 2025 study on enterprise AI adoption — “The GenAI Divide” — surveyed 52 executive interviews, 153 leader surveys, and 300 public deployments.  It found that 95 per cent of AI pilots showed no measurable impact on the profit and loss account.  Only 5 per cent of integrated AI systems created material value.  At the time of the study, US businesses had invested US$35 to US$40 billion in enterprise AI.

The National Bureau of Economic Research surveyed 6,000 executives across the United States, the United Kingdom, Germany, and Australia in February 2026.  Sixty-nine per cent of firms reported actively using artificial intelligence.  Ninety per cent reported no employment or productivity impact over three years.  The average executive used AI tools for 1.5 hours per week.  Executives forecast a 1.4 per cent productivity gain and 0.8 per cent output growth over the next three years.  RAND’s 2025 analysis found an 80.3 per cent AI project failure rate: 33.8 per cent of projects abandoned, 28.4 per cent delivering no value, 18.1 per cent failing to justify their costs.  S&P Global reported that 42 per cent of companies had scrapped most of their AI initiatives in 2025, up from 17 per cent the prior year.

Meanwhile, Microsoft, Meta, Tesla, Amazon, and Alphabet invested US$560 billion over two years and generated US$35 billion in AI-related revenue.  That is a sixteen-to-one ratio of investment to return.  In effect, AI’s contribution to US GDP growth in 2025 is basically nothing.  This is not the dot-com bubble.  It is worse.  In 2000, the underlying businesses were largely not yet real — the revenue model was aspirational by design.  Today, the underlying businesses are real and substantial.  Alphabet, Microsoft, Amazon, and Meta generate hundreds of billions in annual revenue from established franchises.  The danger is not that these companies will collapse.  The danger is that they will borrow their balance sheets into permanent constraint in pursuit of a return on AI investment that the evidence says is not arriving on the timeline the valuations require.  When a company destroys its balance sheet, it damages the real business.  That is what makes this worse.

The Debt Cliff

The transition from equity-funded optimism to debt-funded expansion is the structural signal that a bubble is entering its terminal phase.  Big Tech has begun funding much of its spending growth with debt.  Its own projection is that debt will finance more than a third of AI spending by 2027.  US$3 trillion in off-balance-sheet AI commitments represents obligations that have not yet appeared in any quarterly filing.  They will.  When they do, the refinancing question arrives.  At what rate?  Against what cash flows?  With what collateral?  For companies posting negative free cash flow on USD200 billion-plus capital expenditure programmes, those are not rhetorical questions.

The dot-com era saw 80 million miles of fibre-optic cable laid.  Between 85 and 95 per cent remains dark today — more than two decades later.  The capital expenditure to lay that cable was written off.  The companies that wrote it off either failed or survived as shadows of what they had been.  Total AI capital expenditure in this cycle is running at seventeen times dot-com era levels in inflation-adjusted terms.  The parallel is not perfect.  The hyperscalers have real businesses.  But the parallel does not need to be perfect to be instructive.  What it tells you is that capital deployed at this scale, against this return profile, in this debt structure, has a history of not ending well for the entities that deployed it.

The Signs the Bubble is Nearing Bursting

Bubbles do not burst because someone announces them.  They burst because a single event removes the assumption on which the entire structure was built.  In January 2025, a Chinese AI laboratory called DeepSeek released a large language model that matched the performance of leading American models at a fraction of the training cost.  NVIDIA lost US$588.8 billion in market capitalisation in a single trading session — the largest single-day loss for any stock in recorded financial history.  The market recovered.  But the assumption that AI infrastructure spending is permanently justified by competitive necessity had been punctured.  The Federal Reserve raised rates in September 2026 for the first time since July 2023.  Rate increases ended the dot-com bubble — not because they made individual companies unviable, but because they raised the discount rate applied to future earnings, which deflated every valuation that depended on profitability arriving eventually rather than now.

Insider behaviour is consistent with the late-stage pattern.  Fifty-four per cent of fund managers surveyed in October 2025 described AI as being in bubble territory — the same fund managers whose portfolios are still positioned long.  That is the structural contradiction of a bubble near its peak: everyone knows, and nobody moves first.  NVIDIA, the single stock most identified with the AI trade, trades at 47 times forward earnings as of early 2026.  That multiple requires years of sustained earnings growth.  Any miss in the capital expenditure cycle — any single hyperscaler that blinks — reduces the revenue that justifies that multiple.

A Timeline

Bubbles are easier to identify than to time.  But the data points cluster around a window.  The debt cliff announces itself when refinancing pressure meets declining free cash flow.  The US$1 trillion profit requirement against a US$450 billion consensus is a gap that must close by some point in the 2027 to 2028 period — or the capital spending must slow, which removes the revenue justification for companies like NVIDIA.  Bain projects that by 2030, AI firms will require US$2 trillion in annual revenue from compute alone and may fall US$800 billion short of that requirement.  OpenAI does not expect profitability until 2030, by which point it will have absorbed US$115 billion in cumulative losses.

The timeline is a range.  The conditions for a material correction — not a systemic collapse, but a reckoning — are in place between 2026 and 2028.  The most likely trigger is not a single catastrophic failure but a cascade: one major hyperscaler revises its capital expenditure guidance downward; NVIDIA misses a quarter; the refinancing of a large AI-committed balance sheet occurs at a rate that signals the market’s reassessment of the risk; enterprise AI adoption data for 2026 confirms rather than reverses the MIT and NBER findings.

Each of those events is individually survivable.  In sequence, they remove the assumption one layer at a time.  The assumption that the revenue will come, eventually, at scale, at the rate required to justify the investment.  The South Sea Company’s prospectus described its purpose as “carrying on an undertaking of great advantage, but no one to know what it is.”  The investors were told the returns would be extraordinary.  They were not told what the business did.  We know what AI does.  We know what it costs.  We are still being told the returns will be extraordinary.  The debt cliff does not care about the story.  It arrives when the numbers do not add up.  On current trajectories, the numbers stop adding up somewhere between 2026 and 2028.

---

Terence Nunis | Executive Chairman, Equinox Zenith | Author, The 1% Playbook: The Billionaire Cheat Code