Register

Risks

Each trench that can stall the scoreboard. These are the files that kill a window — not a probability of merger.

• 

  China: Giga Shanghai vs Starshield identity

  existentialbothSilent killer

  CFIUS-class and Beijing-class reviews can stall a merger for years. Do not bury this.

  Tesla’s largest factory and a large slice of deliveries sit in Shanghai, under Beijing. SpaceX is a US national-security launch and comms company. Combining them is not a Delaware filing. It is a CFIUS-class file in Washington and a Party-class file in Beijing. Either capital can stall a merger for years, or bless it with conditions that make the synergy slide a carcass. The merge model defaults chinaApproval to false for a reason.

  Heat

  96

  Window

  If merge

  Likelihood

  80

  File

  geo

  Trigger

  Beijing signal, a CFIUS letter, or a forced Starshield / Shanghai structural demand.

  Merge hit

  Regulatory gauge collapses. Base window becomes “not this decade” unless a carve-out is accepted.

• 

  Unsupervised FSD fatality

  existentialteslaLive mileage

  The first no-supervisor death is not a press cycle. It is a freeze.

  NHTSA is the machinery. This is the event. Unsupervised robotaxi is a geofenced paid pilot — low-million cumulative miles, dozens of Cybercabs. A fatality in a no-steering-wheel vehicle puts a supervisor back in every paid mile, kills the 1M-commercial-robotaxi tranche, and turns the merge model’s unit-economics toggle off. Demos are demos. Miles are the product NHTSA counts.

  Heat

  94

  Window

  2026 Q4

  Likelihood

  38

  File

  regulatory

  Trigger

  Unsupervised fatality, or a broad recall that reimposes a driver in every paid mile.

  Merge hit

  Robotaxi-economics gauge collapses. Window slides later. Pay tranche becomes political poison.

• 

  CFIUS / national-security review of a combined co

  highbothIf merge

  Starshield + Tesla China is the file the staffers actually read.

  Separate from Beijing. A combined Tesla–SpaceX is a US company that owns a China auto factory and a national-security launch/comm business. Staffers read Starshield + Tesla China, not the synergy slide. Mitigation can mean a proxy board, a forced carve-out, or a no. Texas shareholder law does not bind CFIUS.

  Heat

  90

  Window

  If merge

  Likelihood

  75

  File

  geo

  Trigger

  Formal CFIUS notice, or a political demand to split Starshield before any vote.

  Merge hit

  Same file as China risk, Washington side. Regulatory gauge, not engineering.

• 

  Independent Tesla committee absence

  highbothHas not shipped

  Venue moved to Texas. The only collar that holds up in court has not.

  Merge relevance 96. Completion ~15%: Texas reincorporation is done; the committee is not. Elon is buyer and seller. Related-party traffic is already the fact pattern (framework, $1B MTM, Grok in cars). A 2/3 vote is not a committee. A synergy slide is not a committee. Without it a Tesla–SpaceX deal is indefensible in any court that still has a minority-holder docket.

  Heat

  89

  Window

  If merge

  Likelihood

  72

  File

  valuation

  Trigger

  A ratio is published without a truly independent Tesla committee, or the committee is staffed by friends.

  Merge hit

  The deal is a lawsuit on day one. Regulatory gauge is not the problem — fiduciary law is.

• 

  NHTSA FSD / robotaxi audits

  highteslaOpen

  A single unsupervised fatality can freeze the robotaxi trench and the merge clock.

  FSD Supervised is a paid driver-responsibility product (~1.48M subs in Q2 2026). Unsupervised robotaxi is a geofenced paid pilot — low-million cumulative miles, dozens of Cybercabs, not Waymo-scale. NHTSA already has recall and investigation machinery on Autopilot/FSD. A fatality in a no-steering-wheel vehicle is not a press cycle. It freezes the robotaxi trench, the 1M-commercial-robotaxi pay tranche, and the merge model that assumes unit economics in Tesla segment reporting.

  Heat

  88

  Window

  2026 Q4

  Likelihood

  62

  File

  regulatory

  Trigger

  Unsupervised fatality, or a broad FSD recall that puts a supervisor back in every paid mile.

  Merge hit

  Robotaxi-economics gauge drops. Window slides later. The 1M-robotaxi pay tranche becomes political poison.

• 

  Terafab $55B chip moonshot without a fab

  highbothLOI

  Highest merge-relevance node. A cited bill is not wafer starts. CHIPS/CFIUS and talent are the real gates.

  Highest merge-relevance node in the set (98) at ~8% complete. Cited ~$55B-class captive node. Framework exists; projects still “subject to separate negotiations.” No public yield, no wafer starts as a Tesla/SpaceX plant. The conflict: Tesla holders funding a fab whose customers are Tesla inference and SpaceX AI sats. CHIPS, CFIUS, talent, and transfer pricing are the gates — not the name in a bio.

  Heat

  86

  Window

  If merge

  Likelihood

  60

  File

  valuation

  Trigger

  LOI dies, or a real commitment that Tesla funds without a collar.

  Merge hit

  If it dies, the strongest merge argument dies with it. If it commits without a merger, Tesla holders fund SpaceX silicon.

• 

  Terafab transfer-pricing collar

  highboth2%

  TF8, merge relevance 94. The invoice a Tesla minority can take to Texas court. Has not shipped.

  Tesla holders funding a fab whose customers are Tesla inference and SpaceX AI sats. Framework agreement is not an arm’s-length offtake. A collar — audited transfer prices, a floor, a ceiling, a walk-away — is the only document that makes the $55B-class citation defensible. It is 2% complete. A synergy slide is not a collar.

  Heat

  85

  Window

  If merge

  Likelihood

  68

  File

  valuation

  Trigger

  A real commitment that Tesla funds without a collar, or the LOI dies.

  Merge hit

  If it commits without a collar, Tesla holders fund SpaceX silicon. If it dies, the strongest merge argument dies with it.

• 

  FAA Starship licenses / mishap clocks

  highspacexOpen

  Every anomaly inserts months. Cadence is a regulatory product as much as an engineering one.

  Starship is licensed flight-by-flight and mishap-by-mishap. Every anomaly opens an investigation before the next license. Flight 14 (NET 15 Sep 2026) is the first true orbital attempt with ~20 V3; ship catch is delayed. A beautiful stack on the pad is not a cadence. Pad explosions at Starbase feed this same clock. Until orbital reuse is boring, “Starship operational” stays a scenario toggle, not a fact.

  Heat

  84

  Window

  2026 Q3

  Likelihood

  70

  File

  regulatory

  Trigger

  Mishap that opens a long investigation, or a license narrower than the flight manifesto.

  Merge hit

  Engineering gauge stalls. Starship-operational toggle stays off. Cash-flow polarity remains Tesla-funds-Starship.

• 

  Related-party Tesla–SpaceX–xAI traffic

  highbothStructural

  Framework agreement, $1B Tesla MTM on SpaceX, Grok in cars. Merger math is not arm’s length.

  Framework agreement. Tesla’s Q2 2026 $1B mark-to-market on SpaceX holdings. Grok in cars after xAI collapsed into SpaceX pre-IPO. Overlapping shops and compute. Related-party traffic is already the fact pattern. A merger either collapses it into one P&L or cements a transfer-pricing swamp. Fairness opinions that ignore this are not fairness opinions.

  Heat

  82

  Window

  Structural

  Likelihood

  70

  File

  valuation

  Trigger

  SEC or shareholder suit on related-party, or a disclosed offtake that prices Tesla as the captive customer.

  Merge hit

  Independent-committee toggle becomes mandatory. Without it the deal is indefensible.

• 

  AI hosting 90-day cancellation

  highspacexStructural

  $100B ARR talk sits on Cloud Services Agreements that either side can exit in ~90 days.

  Anthropic ~$1.25B/mo, Google ~$920M/mo, plus a Sep 2026 ~$1.11B/mo mystery deal from December. CFO $100B company-wide ARR run-rate is a December exit rate, not FY revenue. S-1-class CSA language: after a short initial period, 90-day termination. SpaceX is being valued as an AI neocloud on contracts that are not backlog. A tenant walking is not a pad explosion — it is a multiple compression. Tesla holders asked to buy that multiple need the independent committee, not a conference line.

  Heat

  80

  Window

  2026 Q4

  Likelihood

  48

  File

  valuation

  Trigger

  A named tenant exercises the 90-day out, or December load is not energized.

  Merge hit

  Valuation-fairness gauge. SpaceX print as AI co becomes a 90-day option, not a DCF.

• 

  Raptor yield

  highspacexGate

  Mars-window cadence is an engine factory problem wearing a rocket costume.

  Flight rate, booster reuse, and ship-catch theater all sit on Raptor yield. A stacked engine photograph is not a production system. If engines are the constraint, Starship cannot be “operational” no matter what the pad looks like. Mars-window cadence is a factory problem wearing a rocket costume. Treat engine serial output as the gate, not the livestream. The Raptor 3 factory card (W) is the production trench; this is the failure mode.

  Heat

  78

  Window

  2026 Q4

  Likelihood

  58

  File

  tech

  Trigger

  Yield stall, a design change that resets the line, or a flight that implicates the engine.

  Merge hit

  Engineering gauge and Starship-success scenario both freeze.

• 

  SpaceX IPO lockup / valuation

  highspacexLive tape

  Tape printed $135 / ~$1.8T class. Morningstar-style fair value sat ~half that.

  An IPO print is not a DCF. Lockup expiry is a supply event. AI-hosting ARR with 90-day outs is being capitalized as if it were backlog. Tesla’s $1B mark-to-market sits on that print. A combined ratio that uses the tape as gospel overpays; a ratio that uses half the tape starts a war with SpaceX employees. Either way the lockup is a clock.

  Heat

  77

  Window

  2026 Q4

  Likelihood

  54

  File

  valuation

  Trigger

  Lockup expiry with a gap-down, or a tenant walking that compresses the AI multiple.

  Merge hit

  Valuation-fairness gauge. The exchange ratio becomes indefensible on one of the two sides.

• 

  ITAR / export on Starship, Raptor, Starshield

  highspacexStructural

  Makes a combined Tesla-in-China / SpaceX-as-defense-prime company a jurisdictional pretzel.

  Starship, Raptor, and Starshield are defense-adjacent, export-controlled articles. Tesla’s supply chain, over-the-air software, and Giga Shanghai sit in a different export universe. Combining them is not paperwork — it is a pretzel: ITAR on the rocket, dual-use AI in the car, China volume in the factory. Staffers who own Starshield will not share a parent with Shanghai without a carve-out, a proxy board, or a no.

  Heat

  76

  Window

  Structural

  Likelihood

  85

  File

  geo

  Trigger

  Export determination that taints Tesla software/AI, or a Starshield program that cannot live inside a China-exposed parent.

  Merge hit

  Forces a carve-out that kills the “one company” slide.

• 

  Optimus useful labor still ~0

  highteslaAcademy

  Fremont Academy units are self-use. Useful factory labor is still approximately zero.

  Optimus is 14% of a labor product (10–18). Design capacity 1M is a slide. 2026 output is imitation-learning units in Fremont, not customer sales, not a closed labor file, not the 1M-bot pay tranche. Merge relevance 92 as a Mars labor story. A walking demo is not a wage. A hand close-up is not a takt time. Until a bot displaces a shift, this is theater with a capex bill.

  Heat

  74

  Window

  2027

  Likelihood

  66

  File

  labor

  Trigger

  Another year of Academy-only output, or a tranche that counts internal units as delivered.

  Merge hit

  The Mars-labor merge story stays a story. Pay-package optics get worse, not better.

• 

  2025 Tesla pay package vs minority holders

  highteslaLive

  1M Optimus + 1M commercial robotaxis can pull capital away from auto margin.

  The live Tesla pay package pays Elon for printing 1M Optimus and 1M commercial robotaxis, among other tranches. Those numbers pull capex and attention away from auto margin. Delaware already voided a prior package. Texas law is friendlier; minority holders are not. A CEO maximizing his Tesla package is not maximizing a fair SpaceX exchange ratio. Academy units counted as “delivered” would be the tell.

  Heat

  72

  Window

  2026 Q4

  Likelihood

  50

  File

  valuation

  Trigger

  Lawsuit, withhold vote, or a tranche that is gamed (Academy units counted as delivered).

  Merge hit

  Makes the independent Tesla committee politically necessary. Package incentives ≠ minority-holder incentives.

• 

  Related-party Grok in cars

  highbothShipped as product

  In-dash AI after xAI collapsed into SpaceX. Merge relevance 89. Not arm’s length.

  Grok in the car is a Tesla SKU whose vendor is now inside SpaceX. Tesla cannot fire the vendor without firing the controller. It is not FSD. It is not the driving stack. It is related-party traffic sitting in the cabin, in the 10-Q, and in every fairness opinion that pretends the companies already behave as one. A merger either collapses the invoice or cements it.

  Heat

  71

  Window

  Structural

  Likelihood

  64

  File

  valuation

  Trigger

  A disclosed offtake that prices Tesla as the captive customer, or an SEC comment on related-party.

  Merge hit

  Independent-committee toggle becomes mandatory. Cabin AI is the exhibit, not the theory.

• 

  Memphis Colossus emissions / turbines

  highspacexOpen

  Air permits cap megawatts. Megawatts cap the $100B ARR story.

  Colossus is a GW-class Memphis cluster on gas turbines and a grid neighbors can see. TDEC/EPA paperwork is the clock, not a keynote. Community air fights do not care about Anthropic’s CSA. A permit stall is a rack that never energizes. Tesla holders asked to buy an AI-neocloud multiple are buying this docket.

  Heat

  70

  Window

  2026 Q4

  Likelihood

  52

  File

  ops

  Trigger

  Denied or narrowed air permit, or a federal notice that caps turbine count.

  Merge hit

  Valuation-fairness gauge. SpaceX-as-AI-co print compresses if Colossus cannot power the CSA load.

• 

  Starship heat shield

  highspacexGate

  Tiles are the product that makes orbit a round trip. A stack on the pad does not prove reentry.

  Flight 13 V3s reentered. That is suborbital theater plus a heat-shield exam, not airline-like reuse. Missing tiles, relight, and ship catch are the same family. Raptor yield gets you up. The shield gets you down in one piece. Until the shield is boring, Starship-operational stays a toggle. A beautiful stack is not a thermal protection system.

  Heat

  68

  Window

  2026 Q4

  Likelihood

  50

  File

  tech

  Trigger

  Reentry anomaly that opens a long investigation, or a design change that resets the tile line.

  Merge hit

  Engineering gauge and Starship-success scenario both freeze.

• 

  Elon attention split

  highbothStructural

  Capital follows attention. The calendar is not a telemetry feed, but the proxy is ugly.

  Tesla, SpaceX, xAI-in-SpaceX, X, Neuralink, Boring, a pay package, a possible merge, and a President-adjacent political load. Attention scores on this board are forensic proxies from earnings mentions, X, and capex guidance — not his calendar. They still show a split. A CEO maximizing a Tesla package is not maximizing a fair SpaceX exchange ratio. A CEO livestreaming a catch is not in a Chancery deposition.

  Heat

  66

  Window

  Structural

  Likelihood

  78

  File

  labor

  Trigger

  A quarter where Tesla ops slip while SpaceX/xAI absorb the public clock, or the reverse.

  Merge hit

  Makes the independent committee mandatory. The controller’s time is a related-party input.

• 

  Pad explosions / environmental at Starbase

  medspacexRecurring

  Physical risk that also feeds the FAA clock.

  Starbase is a factory, a pad, and a town. Catching Super Heavy is real; so are explosions, environmental challenges, and the license docket they feed. Physical risk in Cameron County is political risk in the FAA file. A catch photo is not airline-like reuse. Ship catch is delayed. The orbital site has to survive its own test program.

  Heat

  64

  Window

  2026 Q4

  Likelihood

  55

  File

  ops

  Trigger

  Pad-destroying anomaly, or an environmental injunction that idles the orbital site.

  Merge hit

  Same as FAA — cadence dies, operational-orbit toggle stays off.

• 

  Starship ship-catch delay

  medspacexDelayed

  Booster catch is real. Ship catch is not. Airline-like reuse needs both.

  Chopsticks catching Super Heavy is a photograph that shipped. Ship catch is delayed. Flight 14 is not a catch attempt. Orbital reuse that is boring — the thing the merge model calls “Starship operational” — sits on catching the upper stage, not on a booster highlight reel. A delay here is a cadence delay wearing a livestream costume.

  Heat

  61

  Window

  2026 Q4

  Likelihood

  64

  File

  ops

  Trigger

  Further slip past 2026, or a catch attempt that writes a long mishap clock.

  Merge hit

  Engineering gauge. Starship-operational toggle stays off. Cash-flow polarity remains Tesla-funds-Starship.

• 

  FCC Starlink shells + V3 constellation

  medspacexOpen

  Working product ≠ authorized mega-constellation. V3 100k request is a separate proceeding.

  ~12.9k launched / ~11.1k working (Sep 2026) is a live network. It is not an authorized mega-constellation. V3 and the 100k-class request sit in a separate FCC proceeding from today’s shells. Astronomers, debris dockets, and Kuiper all file in the same building. A dish on a roof proves service; it does not prove spectrum forever. Starlink-in-cab and AI-sat offtake for Terafab both assume the network keeps growing.

  Heat

  58

  Window

  2027

  Likelihood

  40

  File

  regulatory

  Trigger

  Denial or severe condition on V3 / Gen2 shells; sharing rules that cap capacity.

  Merge hit

  Starlink-in-cab and the AI-sat demand story for Terafab both weaken.

• 

  Lithium refinery / cathode ramp

  medteslaRamp

  Vertical materials 30–50%. A miss keeps Tesla a pack assembler.

  The Corpus Christi-class lithium refinery is the vertical-materials trench. Public commentary puts independence in the 30–50% band. Construction is not nameplate; nameplate is not spec lithium in a qualified cathode. A miss keeps Tesla a pack assembler buying lithium and cathode at whoever’s price. Energy and robotaxi both sit on that bill.

  Heat

  52

  Window

  2027

  Likelihood

  45

  File

  ops

  Trigger

  Ramp miss, environmental/permit stall, or cathode that does not qualify.

  Merge hit

  Tesla stays a car company with a materials bill. Weakens the industrial-champion merge story, not the legal one.

• 

  Direct-to-cell / T-Mobile

  medspacexLive SKU, thin spectrum

  T-Satellite is a product. A fourth nationwide 5G network is not.

  Direct-to-cell talks to an ordinary phone. T-Mobile is the live partner. EchoStar spectrum is the next bet, not a grant that already shipped. Capacity per cell is physics plus spectrum, not a keynote. A text that works in a dead zone is not a replacement for a dish and not Starlink-in-cab. Partner concentration is a single-thread risk.

  Heat

  52

  Window

  2027

  Likelihood

  42

  File

  tech

  Trigger

  Partner unwind, EchoStar path dies, or FCC conditions that cap DTC capacity.

  Merge hit

  Soft on merger law. Hard on the Starlink growth story inside a combined multiple.

• 

  4680 COGS still not at cost

  medteslaRamp

  Unbundled ~48%. A cell photo is not a cost curve.

  4680 was supposed to take cell cost out of the vehicle and the pack. Public commentary still reads as a ramp, not a nameplate that beats cylindrical commodity. Energy GWh can grow on purchased cells. Robotaxi unit economics cannot, not at the slide price. Dirt in Corpus and a Nevada building are not spec cells in a qualified pack.

  Heat

  50

  Window

  2027

  Likelihood

  58

  File

  ops

  Trigger

  Another cost miss, or a quiet return to 2170/externals as the volume cell.

  Merge hit

  Tesla stays a car company with a materials bill. Weakens the industrial-champion merge story.

• 

  Labor / UAW / Fremont conversion optics

  medteslaWatch

  S/X line converted to Optimus Academy. A factory doing robots is a labor story.

  Fremont S/X converted toward Optimus Academy (~late Aug 2026). A factory that printed halo cars now prints humanoid R&D. “Design capacity 1M Optimus” is a slide. A labor product is headcount, safety, and a wage story. UAW and California optics travel with that conversion. Academy units are not delivered bots, and they are not a closed labor file.

  Heat

  48

  Window

  2027

  Likelihood

  35

  File

  labor

  Trigger

  Organizing win, a fatal factory incident, or a conversion that reads as a jobs wipe.

  Merge hit

  Soft. Hurts Tesla narrative and the Optimus pay tranche more than the legal merge path.

• 

  Cybertruck demand / price cuts

  medteslaWatch

  A halo truck is not a volume franchise. Price cuts are the tell.

  Cybertruck was the demand-is-infinite exhibit. Production exists. Residual demand at the original price does not. Price cuts, inventory, and a line that can be turned into something else (Optimus Academy logic) are the forensic tells. A stainless truck is not the robotaxi, and it is not the 1M-bot tranche. It is a cash and attention drain if demand stays thin.

  Heat

  48

  Window

  2026 Q4

  Likelihood

  60

  File

  ops

  Trigger

  Sustained inventory, another cut, or a line reassignment that reads as a demand confession.

  Merge hit

  Tesla cash-flow gauge. A weaker auto P&L is a weaker buyer of SpaceX.

• 

  Dojo / Cortex independence

  medteslaSlip

  Tesla still sits in someone else’s training queue. A cluster photo is not a process.

  Dojo was the independence story. Cortex is the cluster that actually trains. Public artifacts show spend and racks, not a closed training stack that no longer needs Nvidia. FSD and Optimus both sit on that bill. Related-party Grok/xAI compute after the SpaceX collapse makes the independence claim harder, not easier. A demo board is not a training run.

  Heat

  46

  Window

  2027

  Likelihood

  55

  File

  tech

  Trigger

  Another architecture reset, or a disclosed spend that shows Tesla still a captive Nvidia tenant.

  Merge hit

  Soft on the legal path. Hurts the “Tesla is an AI company” multiple that is supposed to pay for SpaceX.

• 

  Kuiper spectrum / sharing fights

  medspacexOpen

  Amazon files in the same FCC building. A dish on a roof is not spectrum forever.

  Starlink is the network (~12.9k launched / ~11.1k working). Kuiper is a funding story with rockets on order and a law firm in the FCC docket. Sharing rules, NGSO interference, and V3 capacity all sit in one proceeding. A condition that caps low-elevation or V-band is a capacity haircut, not a press war. Starlink-in-cab and AI-sat offtake both assume the network keeps growing.

  Heat

  44

  Window

  2027

  Likelihood

  36

  File

  regulatory

  Trigger

  Sharing rule or condition that caps Starlink V3 capacity in Kuiper’s favor.

  Merge hit

  Soft on the legal merge. Hard on the Starlink cash-flow story Tesla holders are asked to buy.

• 

  Florida LC-39A Starship pad

  medspacexPermit stack

  A second pad is not a second license. NASA landlord plus Florida environment plus FAA.

  LC-39A already flies Falcon. Turning it into a Starship orbital site is a different stack: NASA as landlord, Florida DEP, FAA, wildlife. Starbase remains the factory. Cadence that assumes two orbital sites assumes both dockets stay green. A Cape delay does not kill Starship. It keeps “operational” a scenario toggle for longer.

  Heat

  42

  Window

  2027

  Likelihood

  40

  File

  ops

  Trigger

  Environmental injunction, NASA lease friction, or a pad mod that slips past the next window.

  Merge hit

  Same family as FAA — cadence, not CFIUS. Engineering gauge stalls if Starbase is the only clock.

• 

  Solar Roof slip

  medteslaSlipped

  ~22% of the story it was sold as. Storage leads GWh, not tiles.

  Solar Roof is photogenic and late. NEM 3.0 cut the California export. Energy beachhead completion sits on Megapack, not on roofs. A tile that leaks is a warranty, not a climate product. This trench can stay slipped forever without killing a merge — and that is the forensic point. It is a credibility haircut, not a CFIUS file.

  Heat

  38

  Window

  Structural

  Likelihood

  70

  File

  ops

  Trigger

  Another quiet de-emphasis, or a warranty wave that makes the SKU a carcass.

  Merge hit

  Soft. Hurts Tesla narrative more than the legal merge path.