Dwarkesh Podcast: Elon Musk – "In 36 Months, the Cheapest Place to Put AI Will Be Space"
Date: February 5, 2026
Host: Dwarkesh Patel
Guests: Elon Musk, Milan Cvitkovic
Episode Overview
This deeply-researched conversation centers on Elon Musk's thesis that space will soon become the most economical and scalable location for AI computation. The discussion delves into the constraints of terrestrial data centers, manufacturing bottlenecks, regulatory and geopolitical considerations, the future of AI hardware and chips, the role of humanoid robots, and larger societal impacts. The tone mixes technical depth, irreverent humor, and philosophical speculation.
Key Themes & Discussion Points
1. Why Put AI Data Centers in Space?
-
Energy is the Constraint
- Elon highlights that global electricity output (outside China) is flat, while AI GPU demand is soaring. Terrestrial data centers will soon hit a wall due to inability to scale up power.
"Where are you going to get your electricity? Especially as you scale, the output of chips is growing pretty much exponentially, but the output of electricity is flat." (00:45, Elon Musk)
- Elon highlights that global electricity output (outside China) is flat, while AI GPU demand is soaring. Terrestrial data centers will soon hit a wall due to inability to scale up power.
-
Solar in Space: Efficiency and Scalability
- 5x solar panel efficiency in space due to no atmosphere, clouds, or night. No need for batteries. Permitting issues, land use, and regulatory friction on Earth make space more appealing.
"It's always sunny in space... my prediction is that by far the cheapest place to put AI will be space in 36 months or less." (02:08, 03:07, Elon Musk)
- 5x solar panel efficiency in space due to no atmosphere, clouds, or night. No need for batteries. Permitting issues, land use, and regulatory friction on Earth make space more appealing.
-
Regulatory Arbitrage
- Building and scaling infrastructure in space will ultimately prove easier, as terrestrial expansion faces red tape, land constraints, and slow-moving utility companies.
"It's harder to scale on ground than in space... and you get about five times the effectiveness of solar panels in space." (02:08, Elon Musk)
- Building and scaling infrastructure in space will ultimately prove easier, as terrestrial expansion faces red tape, land constraints, and slow-moving utility companies.
2. Engineering and Supply Chain Bottlenecks
-
Limited Terrestrial Power Infrastructure
- Building big data centers needs not just power plants but also new industrial infrastructure (transformers, cooling, backup). The slow-moving pace of utilities impedes growth.
"Those who have lived in software land don't realize that they're about to have a hard lesson in hardware." (05:07, Elon Musk)
- Building big data centers needs not just power plants but also new industrial infrastructure (transformers, cooling, backup). The slow-moving pace of utilities impedes growth.
-
Solar Panels: Supply, Tariffs, Manufacturing
- Tariffs sharply limit US solar imports, and domestic capacity lags. SpaceX and Tesla are ramping to 100 GW/year of solar cell production, planning to cover the stack from raw polysilicon.
"We're going to make solar... it's easier for space because no glass or frames are needed." (07:30, Elon Musk)
- Tariffs sharply limit US solar imports, and domestic capacity lags. SpaceX and Tesla are ramping to 100 GW/year of solar cell production, planning to cover the stack from raw polysilicon.
-
Gas Turbine Backlogs
- The bottleneck for power plant construction is specialized turbine components (vanes and blades), with only a few casting companies worldwide.
"Only three casting companies in the world... they're massively backlogged." (13:06, Elon Musk)
- The bottleneck for power plant construction is specialized turbine components (vanes and blades), with only a few casting companies worldwide.
3. Scaling AI Compute and Manufacturing Chips
-
Matching Sun’s Output Requires Off-Planet Expansion
- Earth only receives a minuscule fraction of the sun’s energy; real Kardashev-scale growth demands harnessing solar power in space and launching hardware off-Earth.
"Earth only receives about half a billionth of the sun's energy... the only way to scale is to go to space." (21:23, Elon Musk)
- Earth only receives a minuscule fraction of the sun’s energy; real Kardashev-scale growth demands harnessing solar power in space and launching hardware off-Earth.
-
Chips: The (Next) Critical Limiting Factor
-
Current global compute is ~25 GW. Getting to a terawatt requires "very big chip fabs" and new manufacturing paradigms—existing companies cannot scale fast enough.
"You can't partner with existing fabs because they can't output enough, the chip volume is too low." (24:05, Elon Musk)
-
Supply of high-end chips (logic and memory) is severely constrained; fabs take 5 years from start to volume, and Musk is prepaying for all the capacity he can get.
"TSMC and Samsung... pedal to the metal. Still not fast enough." (30:22, Elon Musk)
-
-
Memory as Bottleneck
- It's harder to scale memory than logic, and Musk jokes about current memes where DDR chips are more valuable than rescue.
"That's why you see DDR prices going ballistic... you write DDR, memory ships come swarming in." (27:49, Elon Musk)
- It's harder to scale memory than logic, and Musk jokes about current memes where DDR chips are more valuable than rescue.
-
Terafab Vision
- Musk wants to build their own advanced fabs for logic, memory, packaging—starting with small-scale experiments, iterating, then scaling to millions of wafers/month.
"We make a little fab and see what happens. Make our mistakes at a small scale and then make a big one." (162:13, Elon Musk)
- Musk wants to build their own advanced fabs for logic, memory, packaging—starting with small-scale experiments, iterating, then scaling to millions of wafers/month.
4. AI in Space: Launch Cadence, Hardware, and Business Models
-
Starship Launch Economics
- Scaling up to 10,000+ launches/year is "just like airlines," eventually aiming to launch a Starship every hour, with ~20-30 reusable vehicles.
"You could probably do it with as few as 20 or 30 [Starships]... SpaceX is gearing up to do 10,000 launches a year." (17:35, 18:14, Elon Musk)
- Scaling up to 10,000+ launches/year is "just like airlines," eventually aiming to launch a Starship every hour, with ~20-30 reusable vehicles.
-
Data Flow & Resilience
- Neural nets are naturally resilient to radiation/bits flips, making them viable in orbit; chips need to run hotter (higher radiator efficiency).
"Neural nets are going to be very resilient to bit flips... just design it to run hot." (158:53, Elon Musk)
- Neural nets are naturally resilient to radiation/bits flips, making them viable in orbit; chips need to run hotter (higher radiator efficiency).
-
Vision for AI Supremacy
- SpaceX could outpace all terrestrial compute, and Musk positions X.AI and his companies to lead by solving hardware/power ramp up.
"SpaceX will launch more AI than the cumulative amount on Earth of everything else combined." (18:27, Elon Musk)
- SpaceX could outpace all terrestrial compute, and Musk positions X.AI and his companies to lead by solving hardware/power ramp up.
-
Business Model Revolution
- Fully digital corporations (AIs + robots) will quickly outcompete any human-in-the-loop operation.
"Pure AI, pure robotics corporations... will far outperform any corporations that have humans in the loop." (75:34, Elon Musk)
- Fully digital corporations (AIs + robots) will quickly outcompete any human-in-the-loop operation.
5. Robotics—Optimus and Recursive Manufacturing
-
The "Infinite Money Glitch"
- Once humanoid robots can build more robots, recursive exponential growth kicks in: labor, intelligence, and production all explode.
"Optimus the infinite money glitch because you can use them to make more Optimuses... usefulness is exp(intelligence × dexterity × #robots)." (61:36, 63:00, Elon Musk)
- Once humanoid robots can build more robots, recursive exponential growth kicks in: labor, intelligence, and production all explode.
-
Manufacturing Bottlenecks, Hardware, and S-Curve
- Scaling output is an S-curve, but Optimus "gen 3" may plausibly hit 1M units/year, ramping to 10M+ after design optimizations.
"It's going to be a stretched out S curve because so much of what goes into Optimus is brand new. There is not an existing supply chain." (87:17, Elon Musk)
- Scaling output is an S-curve, but Optimus "gen 3" may plausibly hit 1M units/year, ramping to 10M+ after design optimizations.
-
US-China Competition
- The US cannot win on labor vs. China (4x population, higher work ethic), but with robots, advantage could flip—if recursive manufacturing is cracked.
"We definitely can't win with just humans... but we might have a shot at the robot front." (96:41, Elon Musk)
- The US cannot win on labor vs. China (4x population, higher work ethic), but with robots, advantage could flip—if recursive manufacturing is cracked.
6. AI Safety, Human Alignment and the Light Cone of Intelligence
-
Grok’s Mission: Understand the Universe
- Musk wants AI to have values emphasizing curiosity and expansion of consciousness, ensuring humanity's arc persists.
"You want to take the set of actions that maximize the probable light cone of consciousness and intelligence." (38:10, Elon Musk)
- Musk wants AI to have values emphasizing curiosity and expansion of consciousness, ensuring humanity's arc persists.
-
Truth-Seeking, Not Political Correctness
- He stresses rigorous truth-seeking over “politically correct” output.
"You want to make sure that the axioms are as close to true as possible... otherwise you will test that technology against reality." (43:00, Elon Musk)
- He stresses rigorous truth-seeking over “politically correct” output.
-
Governance and Guardrails
- Musk worries most about government misuse of AI and robotics—calls government “the biggest and worst corporation”—and favors limited powers.
"Probably the biggest danger of AI or maybe the biggest danger of failure for AI and robotics going wrong is government." (154:01, Elon Musk)
- Musk worries most about government misuse of AI and robotics—calls government “the biggest and worst corporation”—and favors limited powers.
7. Leadership, Corporate Culture, & Scaling Innovation
-
Maniacal Sense of Urgency
- Everything at Tesla and SpaceX is driven by Musk’s relentless focus on the limiting factor and pushing deadlines with maximum urgency—cutting through bureaucracy and chronic pain for direct acute action.
"A maniacal sense of urgency is a very big deal... figure out what the limiting factor is and help the team address that limiting factor." (133:57, Elon Musk)
- Everything at Tesla and SpaceX is driven by Musk’s relentless focus on the limiting factor and pushing deadlines with maximum urgency—cutting through bureaucracy and chronic pain for direct acute action.
-
Engineering Depth from the Top Down
- Musk personally runs deep technical reviews, skip-level meetings, and makes radical decisions (e.g., switching Starship from carbon fiber to steel).
"I have a very detailed engineering reviews weekly... skip-level meetings. There can't be advanced preparation." (134:19, 135:31, Elon Musk) "Steel at cryogenic temperature has similar strength-to-weight as carbon fiber but costs 50 times less... we should have done steel from the beginning." (118:27, 124:19, Elon Musk)
- Musk personally runs deep technical reviews, skip-level meetings, and makes radical decisions (e.g., switching Starship from carbon fiber to steel).
8. US Policy, Fraud, and Institutional Stagnation
- Government Waste and Fraud
- Musk describes how he tried (and mostly failed) to cut federal fraud as part of a hypothetical government stint, citing half a trillion dollars possibly lost to scams and a lack of competence (“it's not even a bastion of competence, it's the DMV, but worse!”).
"You really have to stand back and recalibrate your expectations for competence... payments being sent out with no appropriation code, no explanation." (146:42, Elon Musk)
- Musk describes how he tried (and mostly failed) to cut federal fraud as part of a hypothetical government stint, citing half a trillion dollars possibly lost to scams and a lack of competence (“it's not even a bastion of competence, it's the DMV, but worse!”).
9. Philosophy, Simulation Theory, and the Edge of the Future
-
Simulation Irony
- Musk muses that only the most interesting (and ironic) realities get simulated to the end—hence the succession of wild technological and narrative leaps in his career.
"The most interesting outcome is the most likely, because simulations that are not interesting will be terminated." (57:33, Elon Musk)
- Musk muses that only the most interesting (and ironic) realities get simulated to the end—hence the succession of wild technological and narrative leaps in his career.
-
Err on the Side of Optimism
- Despite “doomerish” realities, Musk urges optimism: better to err on the side of being wrong with hope than right with despair.
"Better to err on the side of optimism and be wrong than err on the side of pessimism and be right." (168:18, Elon Musk)
- Despite “doomerish” realities, Musk urges optimism: better to err on the side of being wrong with hope than right with despair.
Notable Quotes & Moments
-
On Why Space Will Be Cheapest for AI
"Mark my words, in 36 months, but probably closer to 30 months, the most economically compelling place to put AI will be space." (03:24, Elon Musk)
-
On Bottlenecks
"Basically, if something's going really well, they don't see much of me. If something's a limiting factor, they see a lot of me." (137:23, Elon Musk)
-
On Truth-Seeking in AI
"You want to make sure that the axioms are as close to true as possible... There’s no bullshitting physics." (43:00, Elon Musk)
-
On Recursive Robotics and Exponential Growth
"Optimus can be used to make more optimuses. You have a recursive multiplicative exponential. This is a supernova." (63:00, Elon Musk)
-
On Government as the Biggest Corporation
"People worry about corporations, but government is just the biggest corporation with a monopoly on violence." (154:01, Elon Musk)
-
On Limiting Factors
"I think you sort of hit the hardware wall and then whichever company can scale hardware the fastest will be the leader." (166:17, Elon Musk)
Timestamps – Important Segments
- Why AI in Space is Inevitable: 00:23–03:19
- Hardware & Infrastructure Bottlenecks: 04:51–11:58
- Solar Tariffs, Production, & China: 13:02–15:14
- Scaling Compute and Space Launches: 15:07–18:27
- AI Safety & Alignment Philosophy: 36:57–49:53
- Human Robot Recursive Manufacturing: 61:36–67:40
- US vs China — The Future of Industrial Competition: 94:16–101:36
- Leadership Philosophy – Limiting Factors: 133:57–137:23
- Starship: Steel vs Carbon Fiber Decision: 116:53–124:19
- Simulation Theory, Irony in Reality: 57:33–59:56
- Final Reflections on Optimism: 168:13–169:07
Takeaways for the Uninitiated
This episode is a deep exploration of the physical and organizational bottlenecks behind scaling AI and the once-in-a-generation technological shift that Musk envisions. From manufacturing at cosmic scale, to the shifting center of gravity toward space, through the coming disruptions in labor and energy, and the risks and responsibilities of AI, listeners get an unvarnished, technical, yet ultimately optimistic roadmap for the next decade.