Risk Grok
Skim: How value fails to route even if pieces “work.” Common-mode Starship · thermal physics · Suncatcher peer (not disruptor) · Optimus rate cliff · Terafab split-brain (elevated) · Focus bottleneck · multiple confusion · permission.
Opinion-heavy on purpose. Each card: failure mode, second-order effect, what would change your mind.
Common-mode Starship as single throat
Judgment If Starship cadence or reliability disappoints for years, Starmind and heavy space narratives share one failure mode. Diversifying “space bets” that all need the same launcher is not diversification. LIFT binds the whole space half.
Second-order · check
Second-order: terrestrial AI/robotics become the residual story by default — which can inflate Optimus/Autonomy narratives even when they are not ready. Check: correlated schedule language across orbital programs after a Starship stall.
Physics Orbital compute underestimates thermal
Judgment Vacuum is great for some things and hostile for heat. Risk: Starmind slides assume ground-data-center analogies. Reality may force lower density, higher mass, or exotic cooling — killing the “orbit is cheaper” punchline.
Second-order · check
Second-order: cheap lift without thermal closure produces beautiful launches and expensive radiators. Check: whether engineering talk shifts from “tons to orbit” to radiator area, pointing, and degradation — that’s the honest bottleneck migrating.
Competitive Suncatcher as ODC peer (not disruptor)
Judgment Bear (mod conf): Google is a credible orbital-compute competitor — owns chips, may bank earlier rad/hardware data before SpaceX Q4 2027 compute launches; terrestrial Google↔SpaceX ~$920M/mo cancellable ~90d. Not thesis-breaking today: MVP is ~1 kW / one-server; Google pays SpaceX and needs Starship economics; Google large SPCX shareholder; Starlink sats/lasers remain the edge; concept validation supports ODC valuation more than it displaces it.
Second-order · check
Second-order: tweet hype can reprice ODC narrative without moving Q4 2027 SpaceX compute timeline. Check: post-deploy telemetry (orbit + contact Oct 1); rad/HBM (~2 krad HBM); 2027 formation; Google books Starship (bull) vs Blue/ULA/RL (bear); terrestrial renew/cancel. Position stance: no change to SPCX shares or 12/2028 LEAPs; dated risk still lock-up ~Dec 9.
Product Optimus demo → rate cliff
Judgment Classic hardware risk: viral demos outrun reliability and BOM. Markets can re-rate on demos; cash burns on the cliff between demo and rate. LABOR optionality dies on the cliff, not the keynote.
Second-order · check
Second-order: demo success pulls talent/cash from quieter gates (Starmind thermal, D3 yield). Check: internal utilization metrics (even qualitative) lagging demo cadence.
Product Semi Sparks optics → volume cliff
Judgment Inauguration without follow-through rate leaves Semi as calendar theater. Risk: Focus/cash absorbed by ribbon optics while Gate G stays shut — and trucking/fleet narrative outruns disclosed production.
Second-order · check
Second-order: founder calendar collision with F14 / Optimus / Terafab in the same two-week window. Check: whether Sparks is followed by production/fleet evidence (Gate G) rather than another demo cycle. Do not invent Semi unit counts.
Governance Terafab split-brain Elevated
Judgment Elevated Sep 24: Suncatcher sharpens why Terafab matters — chip-cost answer to Nvidia margin / Google TPU. Rad-hard timelines and AI silicon race timelines still do not want the same process compromises. Risk is endless half-measures: neither world-class AI silicon nor trustworthy space silicon — a JV that dilutes both missions. SILICON without walls.
Fact Sep 10: SpaceX sued to block Grimes County release of Terafab tax-abatement/negotiating records — disclosure risk on the JV seam sits beside capacity-wall risk.
Fact Sep 16–17: Tesla + SpaceX (+ SpaceXAI) sued TERA-print LLC in W.D. Tex. Austin for declaratory judgment over the “Terafab” trademark (after C&D) — naming/brand friction parallel to the records fight; does not invent plant timelines or wafer starts.
Second-order · check
Second-order: “process issues” become the public label for priority fights. Check: Unpublished whether capacity is ring-fenced; if not, expect chronic mediocrity on both paths.
Key-person Focus bottleneck
Judgment The orchestra is designed around one routing brain. That is also the risk: illness, distraction, political entanglement, or simple overload. “Bus factor = 1” is not edgy — it’s the org chart. Focus = scarce founder bandwidth.
Fact Sep 14–15: All-In — Musk declined to rule out TSLA–SPCX structural action amid deep collaboration (no plan/timeline). Same window: X/SpaceXAI dismissed Apple antitrust claims; OpenAI claims continue.
Second-order · check
Second-order: political or platform fire drills tax every entity’s critical path simultaneously. Check: whether deputies can close gates without continuous founder arbitration. Structure talk without board mechanics is FOCUS noise until capital action appears.
Market Multiple confusion
Judgment TSLA can trade as auto + energy + robot + space-proxy. When any leg disappoints, the whole multiple compresses. Risk is correlated narrative unwind, not just one P&L miss.
Second-order · check
Second-order: a space miss can reprice Optimus (and vice versa) without cash-flow linkage — pure narrative correlation. Check: which leg the market blames first after an orchestra-wide shock.
Geopolitics / export Space compute & advanced silicon
Judgment Orbital data centers + advanced fabs invite export-control and spectrum/ITAR-shaped headaches. Even if technology works, permission-to-operate can be the binding constraint. Grok treats “we’ll figure regulation later” as a first-class risk, not a footnote.
Second-order · check
Second-order: permission lag turns engineering wins into stranded assets. Check: whether critical path language shifts from welds to hearings — Barriers §5.