Elon Musk teases expectations for Tesla’s AI6 self-driving chip

Tesla says it may tape-out its AI6 self-driving chip as soon as December while accelerating 9‑month chip cycles (AI5/AI6) and planning Samsung/TSMC production to reduce third‑party GPU dependence and power Optimus, Robotaxi and data‑center workloads. Rolls‑Royce has scrapped its 2030 all‑EV target, will continue V12 production and cited a 47% drop in Spectre sales to 1,002 units in 2025, signaling a pullback from aggressive electrification. The U.S. Space Force moved a GPS III launch from ULA’s Vulcan to SpaceX Falcon 9 after a Vulcan solid‑rocket anomaly (the fourth consecutive GPS III reallocation to SpaceX); separately Tesla submitted UN R‑171/Article 39 documentation to Dutch RDW after 1.6M km of FSD testing (13,000 ride‑alongs, 4,500+ track scenarios), with Netherlands approval expected April 10 and potential EU rollout by summer 2026.

Analysis

Tesla's push to internalize high-performance inference and edge training materially changes addressable TAM for third-party datacenter GPUs over a multi-year window. If Tesla reaches cadence of ~2–3 major SoC iterations per product cycle, it converts what used to be recurring GPU capex into episodic foundry + packaging spend; that shifts margin pools toward foundries, HBM vendors and advanced packaging specialists while compressing CPU/GPU OEM growth rates by an observable single-digit percentage annually across Tesla's fleet and robot deployments.

The primary near-term risks are non-linear: a single yield or system-level regression can erase the optionality embedded in future generations, while regulatory/safety shocks can force feature rollbacks that remove monetization levers. Time horizons differ — execution and yield story plays out in quarters (3–9 months), market-share and TAM reallocation play out over years (2–5 years); competitive countermeasures (price cuts, bundled stacks from hyperscalers or GPU vendors) are the fastest way to blunt disruption within 1–2 quarters.

Consensus underestimates two second-order effects: (1) strategic foundry share gains are sticky because moving advanced-node production is multi-year and capacity-constrained, and (2) software-first hardware can forcibly reprice data-center economics by reducing $/TOPS and $/Watt for specific autonomous workloads. That makes a capital-light, option-like long on Tesla software/stack optionality attractive, while selectively shorting pure-play datacenter hardware exposure that lacks vertical integration is a hedge against asymmetric downside.

Key Decisions for Investors

• Long TSLA equity (1.0–1.5% NAV) with 6–12 month horizon; add via a dollar-cost ladder. Rationale: optionality on software monetization and verticalized silicon execution. Risk management: trailing stop 20% or hedge with long-dated puts sized to 25% of position notional.
• Pair trade: Long TSM (1.0% NAV) / Short NVDA (1.0% NAV) for 3–9 months. Mechanic: buy TSM shares or Jan-2027 call LEAPs; fund by shorting NVDA or selling NVDA calls. Risk/reward: capture foundry share reallocation if Tesla shifts volume to Samsung/TSMC while limiting directional market risk; max loss capped by sizing and covered-call leg.
• Small tactical short on legacy OEM optionality (F and GM, 0.5% NAV each) with 6–18 month horizon. Rationale: margin and LTV compression vs vertically integrated competitors. Keep positions small and use single-name CDS or equity puts to limit blow-ups from cyclical rebounds.
• Tail hedge: Buy NVDA 9–12 month out-of-the-money puts sized at 0.25% NAV to protect against rapid secular data-center demand repricing from edge SoC adoption. This is asymmetric insurance — limited premium, high payoff if hardware displacement accelerates.