The Shape of Bubbles

Every generation believes its bubble is different. The dot-com investors thought the internet changed everything. The housing speculators thought real estate only goes up. The crypto evangelists thought decentralisation would replace banks. The AI enthusiasts think intelligence itself has been automated.

They were all partly right about the technology. They were all wrong about the timeline. Can we measure the pattern?

We collected daily price data for four major speculative episodes — the NASDAQ dot-com bubble, the US housing bubble, the Bitcoin crypto mania, and the NVIDIA-led AI hype — and overlaid them on a single timeline aligned to each bubble’s peak. The shape is unmistakable.

This is the chart that tells the whole story. Each bubble is normalised to the same scale (start = 0, peak = 100) and aligned so that the peak occurs at month zero. The vertical axis shows how far prices rose; the horizontal axis shows months before and after the peak.

The pattern is strikingly similar. A gradual build-up, an exponential mania phase, a peak, and then a crash. What changes is the speed. Dot-com took 17 months of mania. Crypto took 13. AI compressed the same arc into what appears to be an even shorter window.

The mania multiplier measures how much the asset rose from mania onset to peak. Each successive bubble has been more extreme: dot-com tripled, crypto rose 6x, and NVIDIA rose over 9x in its mania phase.

Housing is the outlier — a slower, gentler bubble (only 1.3x) but with a much longer crash (67 months vs 12-31 for the others). When an entire economy is the bubble, the unwind is slower but more destructive.

Dot-com (1995–2003): The NASDAQ rose from 1,000 to 5,048 in five years, driven by the belief that the internet would immediately transform every industry. It took 15 years to recover its March 2000 peak. The technology was real — the timeline was a fantasy.

Housing (2000–2012): The slowest bubble and the most damaging. Case-Shiller rose just 1.3x from mania onset — modest by bubble standards. But housing is leveraged: a 27% drop in home values wiped out trillions in household wealth and triggered a global financial crisis. Slow bubbles break more things.

Crypto (2017–2024): Bitcoin went from $10,000 to $67,500 in 13 months, then lost 77% in 12 months. The fastest complete bubble cycle in our dataset. Crypto also demonstrated something new: the bubble can re-inflate. Bitcoin crashed 84% in 2018, recovered, and ran an even bigger bubble in 2021.

AI Hype (2022–present): NVIDIA rose 9.3x from its October 2022 low to its June 2024 peak — the most extreme mania multiplier in our dataset. Whether this is a bubble or a genuine paradigm shift remains to be seen. The technology is real. The question is whether the market has priced in a decade of growth in 18 months.

Google Trends captures the moment a bubble enters public consciousness. For crypto, search interest for “bitcoin” spiked to 100 at the exact peak of the 2017 and 2021 bubbles — maximum public attention coincides perfectly with the price top.

This is the textbook “buy the rumour, sell the news” pattern: by the time everyone is searching for it, the smart money has already left.

For AI, the search data tells a more complex story. “ChatGPT” peaked in early 2023 and has been declining since — but NVIDIA stock kept climbing for another 18 months. The hype and the investment are decoupling: public fascination faded, but institutional money kept flowing.

Three findings survived our analysis:

1. Bubbles have a signature shape. All four follow the same pattern: gradual build, exponential mania, sharp peak, steep crash. The shape is so consistent you can overlay them and they nearly match. This isn’t coincidence — it’s human psychology expressing itself through markets.

2. Each bubble is faster and more extreme. The mania multiplier has grown from 3.1x (dot-com) to 9.3x (AI). The crash duration has compressed from 31 months to 12. Information travels faster, capital moves faster, and collective delusion now builds and collapses in internet time.

3. The technology is always real. The prices are always wrong. The internet did transform everything — just not by 2001. Housing is a legitimate store of value — just not at 2006 prices. Crypto does enable permissionless finance — just not at $67,000 per Bitcoin (at that time). AI will change the world — the question is whether NVIDIA at 9.3x its pre-hype price already accounts for that future.

Data: Daily price data from Yahoo Finance (NASDAQ, BTC, ETH, NVDA). Monthly housing data from FRED (Case-Shiller National HPI). Search interest from Google Trends.

Normalisation: Each bubble normalised to 0–100 scale (pre-mania start = 0, peak = 100). Aligned on peak date (month 0) for overlay comparison. Weekly resampling for chart clarity.

Mania phase: Defined as the period of accelerating price growth from an identified inflection point to the peak. Inflection points selected based on price acceleration, not arbitrary dates.