Investors Were Wrong About the Cloud Twice

In 2006, Amazon began renting out two basic building blocks over the internet, a place to store files and a way to borrow computers by the hour. The next year, Jeff Bezos, writing in Amazon's annual letter, explained at length why this business, called Amazon Web Services, mattered.

The investor reaction was just not good. Eric Vishria, a partner at the venture firm Benchmark, says that if you had "put 30 of the smartest investors" of that era in a room in 2007 and asked whether rented computers could be a durable, high margin business, "you would have gone zero for 30."

That dismissal is the first half of a mistake that investors then made in reverse, and it is why Vishria thinks Wall Street is misreading artificial intelligence now.

By 2014, when Vishria joined the venture business, the consensus had flipped completely. The common line was that Amazon Web Services would eat everything, leaving no room for other business software companies. The fear made sense on paper. The beloved software companies of the time collected yearly subscriptions at about 85% gross margins, the share of each sale left after direct costs, and the worry was that Amazon, happy to operate at something like 8%, would simply offer every database, tool, and app cheaper and better.

It did not happen that way. From 2014 to today, Snowflake built a direct rival to Amazon's own database and ran it on Amazon, in Vishria's phrase, out-Amazoning Amazon on Amazon. Confluent, Elastic, Databricks, and Datadog all thrived at the infrastructure layer, and Datadog alone grew into a company worth about $100 billion. There was roadkill, and Amazon crushed plenty of smaller products, but the eat-everything thesis was, in his words, massively wrong.

It was wrong in a second way too. In 2014, Microsoft's Azure and Google's cloud business were nearly irrelevant. By 2026, they had become enormous businesses alongside Amazon in what Vishria describes as roughly a 40, 30, 20 split of the cloud market among Amazon, Microsoft, and Google.

Even outside that trio, Cloudflare, a cloud provider of a different sort, grew into another company worth about $100 billion.

The lesson, for Vishria, is what happens when investors think in zero-sum terms, meaning one winner takes everything so others lose, and forget to ask what if it all works. The market turned out to be so large that one vendor, even one as capable as Amazon, in Vishria's words, "could not scale and consume it all."

That is the rhyme he hears today when people say one artificial intelligence lab, such as Anthropic, will do everything. Scaling AI requires a huge physical buildup of power plants, chips, memory, energy, and new techniques, and it is growing far faster than cloud ever did. Vishria expects the same ending, an oligopoly, a handful of big companies, at the top plus smaller winners growing into companies worth about $100 billion.