Google X's Astro Teller: The $1B Bet No CEO Will Back, Moonshots 3x Cheaper in 16 Yrs, and Clean Water at 1¢/L| EP #300

Astro Teller defines a moonshot as an answer to a huge problem, built around a science-fiction-sounding solution and breakthrough technology that offers at least some chance of working. The idea is a testable hypothesis, not a promise: teams need enough audacity to pursue unlikely possibilities and enough humility to test them quickly, learn from evidence, and stop when the case weakens. X tries to establish whether a project’s core premise is feasible before investing in supporting infrastructure, so weak projects do not linger as “zombies.”

X gives code names to about 100–200 ideas a year, but only around two typically graduate as moonshots after five or six years. Across more than 16 years, Teller estimates that roughly 2,000 ideas received code names and 35–50 graduated, a success rate close to 2%. Projects can be rejected for being too small, too predictable, potentially harmful, or economically unworkable. Early analysis of materials, costs, and what customers might pay can show whether an idea could become an enduring business. Teller argues that purpose and profit can reinforce each other: a money-losing idea is unlikely to change the world at scale.

Teller illustrates organizational risk aversion with a choice between a guaranteed $1 million and a 1-in-100 chance of $1 billion. The riskier option has 10 times the expected value, and nearly every CXO he tested chose it. But when asked whether their manager, CEO, or board would support that choice, no one kept a hand raised. Teller argues that high-risk work should sit at the edge of an organization, with teams able to report directly to the CEO instead of being governed by operations optimized for routine returns. He says Google’s “mothership” left X alone for long periods, while other companies’ organizational immune systems can pull explorers back into the parent company and disrupt their work.

The cost of getting an individual moonshot through X’s process has fallen by about a factor of three over 16 years. Teller estimates a decline of roughly 10–20% a year, while stressing that the rate is uncertain and reflects more efficient execution, earlier project graduation, and AI—not broad economic growth. X keeps teams small; Google Brain had about 18 people when it graduated. Teller says AI could help X test ideas faster, increase teams’ audacity, and accomplish more in less time, while the basic moonshot process remains intact. He does not expect X to be 100% AI in the next decade: people remain necessary to select problems and judge whether solutions are socially acceptable, as well as to support sales and community adoption.

For clean water, Teller says the all-in cost must reach about 1 cent per liter to meaningfully change the world. He estimates nearly 3 billion people are water-stressed and lack clean drinking water, and expects climate-worsened shortages to contribute to hundreds of millions of climate refugees. X has returned to the problem repeatedly, but its analysis put water at about 10 cents per liter, so the team stopped and tried again. Atmospheric water and much cheaper desalination could help bring costs toward the target, but the episode presents that price as a goal, not an achieved result.

Teller says Google Brain began at X about 15 and a half years before the conversation, when Andrew Ng and Jeff Dean explored whether scaling neural networks could revive the field. He credits the project with helping seed deep learning and says TPUs and the Transformer came from it, illustrating that innovations may require 15–20 years of work before they appear to succeed overnight. X has also tested several ways to shift energy across time and location without finding results it is excited about, and Teller says it will keep revisiting education, which he considers a major potential moonshot because it is not working well even in developed countries. He identifies sustained leadership commitment to an engineering culture, including protected space for explorers, as the hardest part for outsiders to replicate.