Tesla's Q3 is the capex receipt Musk has to honor — and the SpaceX merger is the fallback he's keeping warm

Q3 lands as the first earnings call since Reuters (April 23) and again on Sept 29 confirmed Musk has lifted 2026 capital spending to more than $25B — almost triple the $8.53B Tesla spent in 2025. The ramp is concentrated in the second half: H1 2026 capex of $8.28B (per the Q2 10-Q filed Jul 23) implies $16.7B+ of capex has to land in H2 alone to hit the floor. That is roughly 1.6× H1, and the AI/robotics/Optimus/Terafab lines aren't yet generating material revenue.

Tesla paid $2.00B on Mar 12, 2026 for [SpaceX] Class A shares, disclosed in the Q1 10-Q. By Jun 30 the position was carried at ~$4.2B (Basenor, citing Tesla's filing). The Q2 10-Q cash flow report then booked a $1.005B pre-tax unrealized gain — $763M after tax — into Other Income, single-handedly supporting GAAP profit. That same accounting is now a sword pointed back at Q3: [SPCX] closed at $146.03 on Oct 16, down ~25% from the $192 post-IPO peak and within a few dollars of the 52-week low of $104.83. A modest 30% retracement in Tesla's SpaceX position implies a ~$900M–$1.1B unrealized loss booked in Q3.

Two structural changes make the SpaceX line item stickier, not smaller. First, SpaceX completed its acquisition of [xAI] on Feb 2, 2026 at a $1.25T valuation, now operating as \"SpaceXAI\" (rebranded \"SpaceXSI\" in October). Second, Tesla confirmed on the Q2 call it now has an \"investment and framework agreement\" with SpaceX tied to Terafab — meaning the stake sits alongside a deepening operating relationship, not just a passive mark.

Musk started the merger rumor cycle on Tesla's Jul 22, 2026 Q2 call when he cited \"more and more overlap\" between the two companies — naming Terafab, Grok in cars, and Starlink in Cybercab. Tesla general counsel Brandon Ehrhart confirmed \"an investment and a framework agreement\" with SpaceX and said the companies \"can't talk about combining\" on the call. The story then went quiet as [SPCX] sold off — until Oct 14, when Barclays said a TSLA-SPCX merger could be announced before the Dec 6 AGM, citing vertical AI integration as the main value driver.

On Mar 21, 2026 Musk announced Terafab, a planned semiconductor megafab jointly developed by Tesla, [SpaceX], and Intel, with an initial cost band of $20–25B and analyst estimates pushing to $35–45B. Tesla's own role splits: AI4 chips at Samsung, AI5 at TSMC, AI6 (a $16.5B deal) back at Samsung — Terafab, on **Intel](intc) 14A process, sits across both Tesla and SpaceX compute demand, including [Grok] workloads that already power Tesla's in-car assistant. This is the capex line item that survives any thesis change — the AI pivot still happens, the question is whether it shows up in one balance sheet or two.

For NVIDIA, the Tesla Terafab move is a slow-burn margin risk: Tesla is a top-five NVIDIA customer, and an in-house fab means the H100/Blackwell cycle on Tesla workloads has a built-in expiry date. For Samsung Electronics, Terafab partially cannibalizes the AI6 deal but also adds a high-volume customer — and Tesla, per Korean trade press, is already pressuring Samsung on AI5/AI6 pricing. For LG Energy Solution and CATL, the Terafab story is irrelevant to their lithium-ion volume, which depends on Cybercab ramp and storage deployments — the 13.7 GWh Q3 miss is the warning shot.

Three windows matter more than the headline EPS beat. Days (Oct 21 print): free cash flow guidance or commentary is the swing item — anything north of -$2B for Q3 FCF forces a fresh valuation reset. Weeks (Oct 28–Nov 6): Barclays's \"before AGM\" window for an SPCX/TSLA merger announcement; if it slips past early November, the deal pathway narrows materially against the Dec 6 AGM vote. Quarters (Q4 + 2027 ramp): Cybercab ridership growth (Austin-launched Sept 3, 2026) and the first quarterly Optimus V3 units shipped into non-Tesla factories — these are the cash that converts the capex story.