Amara's Law

We overestimate technology in the short run. We underestimate it in the long run. The pattern is so consistent across centuries, industries, and geographies that it qualifies as something close to a law.

Roy Amara, a researcher and president of the Institute for the Future in Palo Alto, articulated the principle sometime in the 1960s or 1970s — the exact date is lost, which is fitting for a law about misjudging timelines. His formulation: "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run."

The statement sounds like a platitude. It isn't. It describes a specific, repeating, exploitable asymmetry in how humans process technological change — an asymmetry that has destroyed hundreds of billions of dollars in capital, ended careers, and simultaneously created some of the largest fortunes in history for those who understood the shape of the curve.

The short-run overestimation is driven by narrative. A new technology appears — the internet in 1994, artificial intelligence in 2023, autonomous vehicles in 2015 — and the story of its transformative potential propagates faster than the technology itself can deliver. Journalists, investors, and founders extrapolate linearly from early demonstrations to full societal transformation. Venture capitalists compete to fund the narrative before it becomes reality. Analysts publish adoption curves that mistake enthusiasm for demand. Capital floods in. Valuations detach from revenue. Timelines compress to absurdity.

The dot-com bubble is the canonical illustration. By March 2000, the NASDAQ had risen 400% in five years. Pets.com, which sold pet supplies online, achieved a market capitalisation of $300 million despite never generating meaningful revenue. Webvan spent $1.2 billion building automated warehouses for online grocery delivery. The technology — e-commerce — was real. The timeline was wrong by roughly a decade. The NASDAQ crashed 78% between March 2000 and October 2002, erasing $5 trillion in market value. The short-run overestimation was spectacular.

The long-run underestimation was even more spectacular. By 2024, Amazon had a market capitalisation exceeding $1.8 trillion. Online grocery delivery — the exact business model that killed Webvan in 2001 — generated over $100 billion in annual U.S. revenue through services like Instacart, Amazon Fresh, and Walmart+. Every assumption the dot-com dreamers made about the internet's transformative power turned out to be correct. Their error was temporal, not directional. They were right about what. They were wrong about when.

Artificial intelligence followed the same arc, twice. The first AI summer (1956–1973) produced breathless predictions: Herbert Simon forecast in 1957 that a computer would beat the world chess champion within ten years and that machines would discover and prove significant mathematical theorems within the same timeframe. Marvin Minsky told Life magazine in 1970 that a machine with "the general intelligence of an average human being" was three to eight years away. When these timelines proved wildly optimistic, funding collapsed. The "AI winter" of the 1970s and 1980s was the trough — the period when the long-run potential was most severely underestimated because the short-run disappointment was still fresh.

The second cycle repeated in the 2010s. Deep learning breakthroughs at the University of Toronto in 2012 launched another wave of overestimation. Self-driving cars would be ubiquitous by 2020. Radiologists would be obsolete by 2025. General artificial intelligence was five years away — a prediction that has been five years away since 1970. The short-run overestimation was, once again, substantial. Waymo, Cruise, and Uber's self-driving division spent a combined $30+ billion without achieving the fully autonomous consumer experience that had been promised.

And the long-run underestimation is, once again, emerging. By 2025, large language models had restructured knowledge work in ways that no mainstream forecast anticipated even three years prior. GitHub Copilot was writing roughly 46% of code on the platform. AI-generated content was flooding search engines, creative markets, and customer service operations. The people who dismissed AI after self-driving cars missed their deadline were making the same mistake the people who dismissed e-commerce after Pets.com went bankrupt had made a generation earlier.

There's a structural irony embedded in these cycles. The overestimation phase, for all its waste, funds the infrastructure that makes the long-run transformation possible. The dot-com bubble funded the fibre-optic cables that made broadband internet possible. The AI hype of the 2010s funded the GPU clusters and training datasets that made large language models possible. The speculative capital that pours in during the overestimation phase builds the physical and digital infrastructure that the next generation of companies will exploit during the long-run acceleration. The tourists lose money. But their money builds the roads.

The non-obvious insight in Amara's Law is that both errors — the overestimation and the underestimation — stem from the same cognitive limitation. Humans think linearly. Technology develops along S-curves with long flat tails that suddenly inflect. The short-run overestimation happens because we project the excitement of the inflection onto a timeline that assumes linear delivery. The long-run underestimation happens because, after the disappointment of the trough, we project the flatness of the disillusionment phase into the future — failing to account for the compounding infrastructure, network effects, and institutional adaptation that will eventually drive adoption past the original hype.

The pattern holds across centuries. The railway mania of the 1840s saw British investors pour £250 million — roughly a third of GDP — into speculative railway schemes, many of which never laid a single mile of track. The bubble burst in 1847. Fortunes were destroyed. Parliament investigated. And then, quietly, the railways that survived rewired the British economy — collapsing the cost of transporting goods, enabling national newspapers, creating standardised time zones, and making modern industrial capitalism physically possible. The short-run overestimation was financial madness. The long-run underestimation was that railways would restructure civilisation.

The telegraph, the telephone, the automobile, radio, television, the personal computer — each technology followed an Amara arc. Each attracted premature capital, experienced a correction or period of disillusionment, and then delivered a transformation that exceeded the original hype by an order of magnitude that nobody in the hype phase had the imagination to forecast. The internet didn't just replace the Sears catalogue, as 1990s analysts predicted. It replaced the newspaper, the travel agent, the record store, the taxi dispatcher, the encyclopaedia, and the classified ad — and created trillion-dollar categories that didn't exist before.

Understanding this shape — hype, crash, slow build, eventual transformation beyond what anyone predicted — is one of the most valuable pattern-recognition skills available to founders and investors. It doesn't tell you which technologies will succeed. But it tells you that the moment of maximum pessimism about a real technology is almost always the moment of maximum opportunity.