Why AI is just big data again

Then Marc Andreessen wrote that software was eating the world and the pitch generalized. Data was gold, data could be applied to any industry, users generated data, data powered algorithms, algorithms produced innovation. Chegg's leadership summarized it exactly: take a giant category, use technology to deliver at scale, own the customer and the channel, keep the data, and build a moat nobody can cross.

The moat argument rested on companies that were already winning. Facebook, Google, Apple and Microsoft posted operating margins of 34%, 35%, 27% and 35% in 2009, and 45%, 26%, 28% and 24% by 2016 — figures the rest of the private sector had never seen. If data was what made FAANG work, a nimbler startup applying it to a smaller industry should manage a tenth of that. That was the whole syllogism.

By 2016 the results were in and they were uniform. Of the fourteen largest data-driven consumer startups, twelve lost money: Snap −$520M, Twitter −$367M, Pandora −$319M, Wayfair −$196M, Dropbox −$194M, Pinterest −$188M, LendingClub −$150M, Wish −$147M, Zynga −$114M, Groupon −$100M, Roku −$43M, TrueCar −$39M. Only GrubHub at +$84M made money.

Silicon Valley's response was to move the goalposts rather than concede. Data was still valuable — you simply did not have enough of it, or the means to interpret it. Basic analytics and personalization were no longer sufficient. What you needed was terabytes, sophisticated tooling and data scientists. That relabelling is what got called Big Data.

The quotes from the peak have not aged well. Groupon: *we have more than 9 petabytes of data and we A/B test every single feature — how many companies do this?* It bottomed out within five years. Wayfair: *4 terabytes a day, 40 billion customer actions a year, 1,900 engineers and data scientists, multiple platforms at a strong ROI.* It still loses close to a billion dollars a year.

Groupon's own numbers are the clearest arc in the industry: revenue $3,192M in 2014 falling to $2,573M, $2,955M, $3,014M, $2,844M, $2,637M, $2,219M, $1,417M, $967M, $599M and $515M by 2023 — with an operating loss in seven of those eleven years.

Blue Apron claimed it knew every individual's taste profile and could algorithmically predict demand for any meal kit. Revenue peaked at $881M in 2017 and fell every year after to $458M, with an operating loss in all nine years — −$189M at the peak of the data story.

The Fortune 500 joined out of fear, not merit. Walmart, ExxonMobil, Home Depot, AT&T, Comcast, Disney, PepsiCo, Chase and Citigroup each committed hundreds of millions — some billions — because no chief executive wanted to be the one caught saying their teams were not data-driven. Under Armour is the small-cap version, spending over half a billion on fitness apps chasing FAANG margins.

A decade later there is nothing in their accounts to show for it. Operating margins in 2013 against 2023: Walmart 5% → 3%, Costco 3% → 3%, Kroger 3% → 2%, Target 6% → 3%, PepsiCo 13% → 12%, Kellogg's 18% → 8%, General Mills 16% → 17%, Unilever 14% → 17%, Chase 27% → 36%, Citi 26% → 25%, Wells Fargo 40% → 20%, Goldman Sachs 37% → 28%. If big data had produced efficiencies, this is where they would appear.

During a gold rush, sell shovels. Of the tech stocks with the greatest appreciation since IPO between 2012 and 2024, the winners are almost entirely enterprise: Shopify at 7,207%, ServiceNow at 6,581%, The Trade Desk at 3,809%, Arista at 3,483%, Palo Alto Networks at 3,318%, Workday at 1,388%, Block at 1,348% and MongoDB at 1,268%.

Splunk grew from $303M to $3,654M and Datadog from $101M to $2,128M selling table-stakes software for capturing and monitoring data. Snowflake went $97M, $265M, $592M, $1,219M, $2,066M, $2,806M. None of that revenue required anyone downstream to derive a single insight.

Being B2B is not the same as being profitable, though. In 2024 the leading enterprise vendors posted operating income of −$1,418M at Shopify, −$1,095M at Snowflake, −$877M at Twilio, −$345M at Atlassian, −$270M at Asana, −$234M at MongoDB and −$33M at Datadog, with only Palo Alto Networks positive at $387M. The advantage is resilience: enterprise deals are large, churn is slow, and a tool embedded at a big company is hard to rip out.

But the biggest winners were the clouds, and it was not close. AWS went from $1 billion in 2011 to $99 billion in 2023 and Google Cloud from $4 billion to $33 billion — because the consumer startups burned venture money on cloud, the Fortune 500 used big data as the forcing function to migrate, and the B2B vendors built their tools there too. It does not matter whether the trend is big data, IoT, AR or AI, as long as it drives usage.

Engineers had their own reasons to keep it alive. The technology was new, so there were no best practices; everyone worked it out in the open on GitHub, and naming the right tools on a résumé converted into a higher-paying job within months. Big data industrialized résumé-driven development, and developers stopped being implementers and became the people choosing the clouds, tools and vendors on their employer's behalf.

And the AI premise is stranger than the one it replaced. Big data said the insights were in there and you needed better tools. AI says the insights are in there and humans cannot get them out — so trust a system almost nobody can inspect, accept the answer instead of seeking it, and value the packaging and the speed above the accuracy. That is a retreat dressed as a breakthrough.

The accounting is the same as it ever was. All talk, no show, all sizzle, no steak. The winners from big data were the founders, executives and venture capitalists who liquidated at IPO and the people at the B2B vendors. The losers were public investors and the employees. Nothing about the AI cycle has changed who sits in which seat.