Back to News
Market Impact: 0.15

Musk claims Tesla will restart work on its Dojo supercomputer

TSLA
Artificial IntelligenceTechnology & InnovationAutomotive & EVTrade Policy & Supply ChainManagement & Governance

Elon Musk announced Tesla will restart work on Dojo3, the third-generation in-house supercomputer project previously disbanded as the company prioritized on-board AI chips. Tesla says the revival follows progress on the AI5 chip design, while AI6 production is slated for Samsung’s Texas fab under a reported $16 billion agreement; Dojo’s stated purpose is large-scale video data training for Full Self-Driving neural nets. Musk has characterized Dojo3 as a space-based compute initiative, a speculative concept that outside experts have questioned, leaving the short-term commercial and financial impact uncertain.

Analysis

Market structure: Tesla restarting Dojo3 is a verticalization signal that benefits Tesla (TSLA) long-term control over FSD data and Samsung (chip foundry partner) for AI6 wafer revenue, while marginally pressuring external GPU/cloud training demand (likely <5–10% of global training spend over 1–3 years if Tesla fully internalizes). Competitive dynamics favor specialized inference/vehicle-chip makers vs. hyperscaler training revenues; major winners in the near term are chipmakers and semicap suppliers (ASML, LRCX, AMAT) while pure-play cloud GPU rental growth slows regionally. Cross-asset: expect higher TSLA equity IV and modestly wider credit spreads on corporate debt if capex rises; commodities impact (copper/launch materials) immaterial near-term, FX neutral. Risk assessment: Tail risks include a failed Dojo3/space compute experiment causing $1–5+ billion write-offs, regulatory action on FSD training data, or Samsung fab yield issues; low probability but high impact over 12–36 months. Immediate (days) market impact is small; short-term (3–12 months) execution and partner-risk dominate; long-term (1–3 years) outcomes hinge on capex discipline and whether space-based compute proves economical versus hyperscalers. Hidden dependencies: launch access, FCC/FAA approvals, Samsung yield curves, and data-sovereignty/regulatory constraints on vehicle video export. Trade implications: Tactical relative-value tilt into semiconductors/AI infrastructure (NVDA, ASML, LRCX) and away from concentrated TSLA exposure is warranted over the next 3–12 months. Direct plays: bias long NVDA and semicap names; hedge or reduce TSLA exposure with puts or modest short. Options: favor directional NVDA calls (6–12 month) and TSLA protective puts (6–9 month 10–20% OTM) to size asymmetry of upside from AI demand versus capex/execution risk. Entry: initiate within 1–4 weeks and re-evaluate at Tesla earnings or Samsung fab update (next 90 days). Contrarian angles: Consensus overstates Tesla’s ability to remove material GPU training demand — historical parallel: Google/Apple in-house silicon improved product differentiation but did not crush GPU ecosystem (NVIDIA still expanded). The market may over-penalize TSLA for ambitious, speculative space compute plans; a >20% drawdown would create a buy-the-dip opportunity for long-term FSD exposure. Unintended consequences include regulatory scrutiny increasing moat erosion risk for FSD if training data rules tighten.