In his first shareholder letter in 1997, Jeff Bezos clearly chose long-term free cash flow over short-term accounting profits. Amazon accumulated nearly $3 billion in losses from its founding until the end of 2001, achieving its first quarterly profit only in the fourth quarter of that year; after the dot-com bubble burst, its stock price fell from $116 to $6. This odds-based thinking is built on the company's ability to finance through consecutive losses.

Amazon continuously directed cash flow generated from its retail business and capital market credit into new infrastructure: launching AWS in 2006, acquiring warehouse robotics company Kiva Systems for about $775 million in 2012, and purchasing Whole Foods for about $13.7 billion in 2017. The funding path extends from cloud computing to automation and physical retail, with the goal of first establishing infrastructure and then spreading fixed costs over larger-scale transactions.

The SoftBank Vision Fund similarly relies on a few super winners to cover many failures, but it has committed over $14.25 billion to WeWork and has recognized about $4.582 billion in losses on related investments in the 2019 fiscal year. Its problem is not low success rates, but rather the continued investment in a single failed project, which undermines the premise of "capped single losses and sufficient betting frequency"; Amazon, on the other hand, would stop the Fire Phone and transfer personnel and technology to Echo.

Essentially, this is about capital concentration: once extreme returns appear, profits, talent, and data will continue to flow to the winners, allowing them to achieve lower costs and more chances for trial and error. The path is "tail returns generate cash flow—cash flow expands experimental scale—scale increases the number of successful projects—winners continue to attract capital," ultimately transforming innovation capability from a technological advantage to a capital structure advantage.

ABAB News · Cognitive Laws

Success rate determines survival, odds determine class
Small losses can be repeated, big gains must be scalable
A true moat is having a few more failures than competitors