Topic: Memory supply constraints and Tesla's AI memory efficiency
Key points:
Tesla AI is "very compute efficient and very memory efficient"; metric is "intelligence per gigabyte" (especially when RAM-constrained).
Tesla's AI intelligence density is "more than an order of magnitude better" than large models like Grok.
Tesla has a solution for logic and memory for roughly the next three years; beyond that, it will be supplier-limited unless more fab capacity (especially in the US) is built.
Without an AI chip, Optimus is "completely useless" (cannot even sit); cars can be retrofitted with steering wheels/pedals, but Optimus cannot.
Mgmt stance: Bullish near-term (3-year solution in place) but cautious beyond due to geopolitical risk and fab supply uncertainty; stresses need for US fab capacity.
Q12 — George Gianarikas
Topic: Competitive landscape for humanoid robots vs. Chinese startups
Key points:
Biggest competition will come from China, which is "incredibly good at scaling manufacturing" and "very good at AI"; no significant competitors outside China.
Optimus is expected to be much more capable than any robot under development in China, particularly in:
Real-world intelligence
Electromechanical dexterity (especially hand design – "by far the hardest thing")
Scaling production
Tesla is the only company that has all three of those components (hand dexterity, real-world AI, production scale).
Mgmt stance: Bullish on Tesla's competitive advantage (hand design, AI, scale); acknowledges China as "tough competition" and warns outsiders often underestimate China.