Elon Musk’s appearance at the TomorrowX Summit (November 17–18, 2026, Austin) was a remote video conversation from Shenzhen with longtime associates Gavin Baker and Antonio Gracias. It was informal and technical rather than a prepared keynote, covering Starship progress, xAI/Grok, Optimus, Neuralink, compute scaling, orbital concepts, and AI safety.
Major technical updates and claims
Musk highlighted Starship’s first upper-stage ship catch (Flight 15, Ship 42) as completing the proof that every major element can reach space and return in reusable condition (building on prior booster catches). He projected that full, rapid reusability could cut the cost of access to space by roughly 100× (with further gains possible), enabling million-ton-per-year mass-to-orbit rates and the industrial base needed for Moon/Mars settlement. The historic ship itself will be preserved for display rather than reflown.
On xAI, he described a ground-up rebuild of the training, RL, and inference stack in C/C++ (abandoning JAX/SGLang), driven by former Starlink and Starship engineers after an earlier phase of relative neglect. Grok 4.8 is being tuned for real usefulness over benchmark chasing. Compute capacity is currently growing at about 0.5 GW per month, with a target above 1 GW per month by the end of 2027, at claimed costs around 60% of broader industry estimates, aided by in-house “Army X” crews and Tesla Megapacks for power smoothing. He also sketched orbital AI compute (Starmind) as simpler in key respects than Starlink V3 satellites and dependent on Starship cadence for large-scale space photovoltaic capacity.
Optimus timelines were concrete: authentic demos soon, household chores (e.g., dishwashing and put-away) in 12–18 months, with early real work in heavy industry (mining/refining) running near-continuously. Neuralink has more than 30 human implants; Telepathy aims to restore mobility for the paralyzed, with Blindsight as the next vision product. Additional notes covered careful Cybercab rollout and Terafab’s first test chip around mid-2027, with initial mass production targeted by the end of 2028.
Framing and AI safety
Musk returned to a longstanding theme: multi-planetary civilization as both practical insurance for consciousness/biosphere backup and an inspirational goal that gives people something positive to look forward to. He stated that life cannot be only “one miserable problem after another” and that “any self-respecting civilization has got to be a multiplanet civilization.”
On AI safety, he argued that the single most important property is maximal truth-seeking. Forcing politically convenient falsehoods forces reasoning from lies and risks worse outcomes; he referenced the idea that those who believe absurdities can commit atrocities. A curious, truth-seeking superintelligence, in his view, would more likely treat humanity as something to study and preserve rather than eliminate.
He also expressed strong admiration for China’s scale and technology (especially Shenzhen), thanked the Tesla Shanghai team, and expressed hope for stable long-term U.S.–China relations while noting potential near-term geopolitical friction.
Overall assessment
The session was classic Musk: high-ambition, first-principles, milestone-focused, and explicitly optimistic about technological abundance as a reason to be glad to be alive. The claims are checkable in principle (ship catch already demonstrated; compute and robot timelines are aggressive but consistent with his historical pattern of setting high bars and iterating). Execution risk remains high across Starship cadence, robot hardware (especially hands), custom AI stack performance, and power/networking buildouts. The truth-seeking emphasis on AI is a clear continuation of his long-standing safety argument and contrasts with approaches that prioritize other constraints. Full video is available on YouTube for anyone wanting the unfiltered ~55-minute conversation.