Every trillion-dollar company owns a museum of the moments it almost died, and a longer corridor of the fortunes it declined. Excite turned down all of Google for $750,000. Intel passed on making the iPhone chip and said no to buying Nvidia, twin decisions that cost it two computing eras. Yahoo declined to buy Google, then Facebook, then sold itself for less than it once refused to pay. In this industry the biggest losses are usually invisible: they are the deals not done and the memos not believed. Sometimes, though, the loss is a deal that was done: Microsoft paid $8.5 billion for Skype and then quietly switched it off.
The pattern behind those misses is always the same, and it is not stupidity. Smart executives valued the future by what it added to their existing model, instead of asking what it would do to that model. Search looked like a commodity feature to a portal company. Graphics chips looked like a toy market to a CPU monopoly. By the time the frame proved wrong, the price had grown a thousandfold. The reverse error exists too: paying $44 billion while waiving due diligence shows what happens when conviction outruns scrutiny.
The dark mirror of those missed fortunes is the data these companies gather in the meantime. When 23andMe went bankrupt in 2025, the genetic code of 15 million people was sold as an ordinary asset, a reminder that in Big Tech the users are frequently the inventory. And in the AI gold rush the oldest trick returned in the newest clothes: Builder.ai raised nearly $450 million on an "AI" that was largely 700 human engineers before it collapsed in 2025. These investigations treat the industry the way historians treat empires: not as inevitabilities but as a sequence of choices that could have gone the other way. Power, in this archive, is just the compound interest on a few decisions that happened to be right.