Elon Musk – "In 36 months, the cheapest place to put AI will be space”
Speakers
Elon Musk
— CEO, Tesla / SpaceX / xAI
Summary
Elon Musk joins Dwarkesh Patel for a wide-ranging discussion on AI scaling bottlenecks, orbital data centers, SpaceX, Tesla, Optimus, xAI alignment, and government. Musk argues power and chips are the main constraints, with memory and fab capacity especially tight, and predicts space will become the cheapest place for AI within 30-36 months. He also outlines Tesla's AI5/AI6, TeraFab, Optimus, self-driving, solar, and refining plans, and warns China is dominant in manufacturing and refining. Investment implications center on AI semiconductor supply chain, Tesla's robotics and autonomy stack, solar and power, and Chinese EV competition.
- Musk sees energy and chip supply as limiting factors for AI scaling.
- He predicts orbital data centers become economically compelling in 30-36 months.
- Memory, advanced fab capacity, and turbine blades are cited as bottlenecks.
- Tesla targets AI5/AI6, TeraFab, Optimus mass production, self-driving, solar, and refining.
- xAI's plan uses a Tesla self-driving-like path for digital human emulators.
- Musk warns China dominates manufacturing, refining, and electricity capacity.
- He discusses government, DOGE, fraud, and AI and robotics risks.
- He remains optimistic about AI, robotics, and space expansion.
Ideas
Solar scalable; Tesla builds capacity.
Musk argues solar is the scalable power source for AI: solar cells are already extremely cheap, space solar is about five times more effective than ground solar and avoids batteries, and scaling solar on Earth is a good path though tariffs and permitting slow it. Tesla and SpaceX are both targeting 100 GW per year of solar cell production, giving Tesla a company-specific role in the buildout.
ASML is China's key chip bottleneck.
Musk says China has not replicated TSMC because it has not replicated ASML; ASML is the limiting factor for China's leading-edge chip efforts and export bans on ASML equipment are biting. This highlights ASML's critical supply-chain position.
Memory bottleneck; Micron asked to expand.
Musk says his biggest concern for scaling AI chips is memory, not logic: the path to sufficient memory is less obvious, DDR prices are going ballistic, and TeraFab must include memory. He has asked Micron to build more fabs faster and offered to guarantee purchases of output.
Foundry capacity tight; Tesla guarantees demand.
Musk says leading-edge fab capacity is a constraint: Tesla has secured all available TSMC and Samsung capacity, using TSMC Taiwan/Arizona and Samsung Korea/Texas, and he has asked suppliers to build more fabs faster while guaranteeing purchases. This is a strong demand signal for leading-edge foundries.
Chinese EV flood favors BYD.
Musk says China is extremely competitive in manufacturing, BYD is reaching Tesla's production and sales scale, and a massive flood of Chinese vehicles and manufactured goods will hit global markets. This is a strong relative demand thesis for Chinese EV leaders like BYD, even as it pressures incumbents.
