Astro Teller says X tracks one number very closely: what it costs to get a project all the way through its pipeline. By that measure, Teller said, moonshots have become about three times cheaper over the last 16 years, and "a lot of it is because of artificial intelligence."
Teller is co-founder and "Captain of Moonshots" at X, the innovation lab owned by Alphabet, Google's parent company. He spoke on stage at Moonshots Live 2026 with the hosts of the Moonshots with Peter Diamandis podcast. The conversation was recorded on September 25, 2026, and published as episode 300 on October 5. Not everyone on stage accepted his number.
What a "graduate" costs
X works on long-shot technology projects. A project that survives is said to "graduate": it moves out of X and into an environment that suits its next stage of growth. That can be a business of its own or a home inside Google. The best-known example is the self-driving car project, which X announced in December 2016 was becoming Waymo, an independent company within Alphabet.
Alexander Wissner-Gross, a computer scientist and founder of Reified, asked Teller which way the real cost of moonshots is heading. He said he could argue either side. Rising expectations could push costs up. Or costs could be "hyper-deflating" because "we have superintelligence now" and hard problems are being solved left and right.
Teller said X is "obsessed with the efficiency of getting at moonshots" and that the cost of reaching each graduate is "down by about a factor three over the last 16 years." He added that the figure is "a bit of a lagging indicator" and that there is "a lot of waste and complexity" in it.
"That's inflation"
Wissner-Gross pushed back right away: "Wait, factor of three over the, that's inflation." Teller checked that he had been understood, saying the factor of three meant cheaper. Wissner-Gross answered that a threefold change over 16 years "rhymes to me like 10% per year," or about what markets return.
He then laid out the argument in full. Some people think the biggest problems in science and engineering will be solved by the whole economy growing more capable, not by a single breakthrough. If the stock market doubles roughly every 10 years, he said, a threefold drop in the cost of moonshots over 16 years "sounds to me like basically just our economy is capable of more things." In real terms, he concluded, "it's the same cost."
The numbers are close. A threefold drop over 16 years works out to about 7% a year, not 10%. Doubling every 10 years also comes to about 7% a year, and over 16 years that compounds to roughly a factor of three.
Teller rejected the comparison: "I think you're confusing the cost to paint a house with the value of the house. They're like apples and oranges." His figure was about how cheaply X can "paint our house," meaning what it costs X to get a project through its process. What the result is worth to the world is a separate question. He used Waymo as his example. Its value, he said, is huge relative to the cost of building that team and all the other projects that failed along the way. "It is a very good return on investment." For a sense of scale, Waymo raised a $5.6 billion investment round led by Alphabet in October 2024.
Teller also gave a second, faster rate. The cost of each project that makes it all the way through, he said, is falling "at, I don't know, 10, 20% year over year, something like that." He gave three reasons: X is getting better at its job, it is graduating projects a little earlier, and AI. He did not say how the two figures fit together. A steady 10 to 20% yearly decline over 16 years would compound to a drop of roughly five to 35 times, far more than the factor of three he cited.
Cheap tools, small teams
The cost question built on a premise from Salim Ismail, founder of Open ExO. Ismail pointed to what Peter Diamandis has called the demonetization of the world: technology turning expensive things cheap or nearly free. For most of history advanced technology was costly, Ismail said, but now solar energy and sensors are cheap and blockchain software is open source and free. A big company that once had to make one large bet, which a nervous chief financial officer would defund as fast as possible, can now run "a portfolio of very disruptive experiments at the edge." What remains, he said, is mindset.
Teller agreed. He said he often hears that X's projects happen because "Alphabet has lots of money," and "I guarantee you that is not why." X keeps its teams tiny. When Google Brain graduated from X, Teller said, it had "like 18 people," and that is pretty normal. X's own description of its factory talks about small teams drawing on shared labs and prototyping workshops.
The small size is deliberate. Teller said he tells teams: "I want you to find a cheat in the video game of life." He wants a shortcut that cuts through the problem, like cutting the Gordian knot. If a team can't find one, he would rather it stop than solve the problem "the normal way," and X will go find a different moonshot.
He also plugged X's method. X made a certificate course with MasterClass that he said takes about 20 hours with all the work, or three and a half hours to watch the videos. "It's our secret sauce is not secret," he said. MasterClass announced the ten-module course in October 2025.
AI as electricity
Dave Blundin, founder and general partner of Link Ventures, offered his own "dirty little secret": doing a moonshot is really easy. You take smart, fanatically focused people who care about a mission and pour money on them. Someday, he said, AGIs (artificial general intelligence, meaning AI as broadly capable as people) will replace the people and compute will replace the money. Until then there must be a transition, with some mix of AIs and people. Had X started down that road?
"Totally," Teller said, and then went on what he called a side rant. If a founder pitched a company by saying it uses electricity, he said, you'd think they were nuts. You'd be worried if the business didn't use electricity, but it would still be "an implementation detail." He feels the same about AI and tries to say the term as little as possible. When Diamandis joked that they should say "SI now, superintelligence," Teller said he got his PhD in artificial intelligence 30 years ago, "so a little stuck on that one."
Every one of X's moonshots, Teller said, is "a bunch of agents and a bunch of people working together to try to solve a problem." Agents are AI systems that can carry out tasks on their own. What X cares about is the problem: clean water, grid-scale energy storage, education, fixing the electric grid, or a more circular economy that doesn't send "trillions of dollars a year to landfill."
Salaries still beat the AI bill
Blundin turned this into a budget question. He used X's clean-water moonshot, which aims for about a penny a liter, the price Teller said would "really change the world." Teller confirmed that X has a team on it now. Which is bigger, Blundin asked, the team's AI bill or its salary bill?
"It's still their salary in that particular case," Teller said, though he didn't know by how much. He called the AI systems "absolutely teammates in the process of innovation," and said X keeps finding ways to automate parts of the moonshot factory so it can move faster.
Blundin asked when spending would reach 50-50, then 75% AI, then 100%. Teller said he is not sure it will be 100% at any point in the next decade. Choosing which problems to solve, making sure they are solved in ways society accepts, working with distributors, and landing a product in communities in ways they find acceptable all require people "for at least the next decade or two," he said.
The shift will go "as fast as is efficient," Teller said. "We're not going to use compute just to use compute." He said some people spin up agents "because it feels powerful to spin up agents," but "there is a difference between spinning up an agent and actually getting benefit."