Built by Claude Fable

SPCX | Investor Sentiment: 4/10

Deep dive pages: Overview | Business Model | Financial Trends | Thematic Exposure | Management Quality | Investor Sentiment | Concerns, Catalysts & Risks

Prepared 2026-10-29, revised after the PM Review the same day. Fundamentals from Daloopa (company_id 247912), cited inline as daloopa.com/src/ID. Fiscal year = calendar year. Last print Q3 2026 (2026-10-20, 10-Q filed 2026-10-22); next print about 2027-02-25.

Inverted scoring. The question is not whether SPCX is loved or hated, but whether management is repeatedly bullish on a specific driver the street refuses to model. There is a real gap on AI compute monetization, but once the $/W units are fixed it is much narrower than the first draft claimed: on observed Q3 economics, FY2027 consensus looks roughly fair, and the gap only opens if the pricing implied by the Q1 2027 guide holds on all 2027 capacity. It also sits inside a crowded, retail-heavy, mostly-Buy tape, with insiders selling into lockup expiries and a CEO whose timelines the street discounts for good reason. Fundamentals are inverting from worse to better (GAAP operating margin -41.4% in Q1 2026 to +19.1% in Q3), but sentiment already inverted, so the easy money is not on offer. Net: divergence is partial and smaller than scored before, edge is thin.

4a. Management-Street Divergence (primary, 50% weight)

The specific thesis management keeps repeating: SpaceX can deploy power, cooling and compute faster than anyone (casting its own turbine blades and vanes, "Yankees vs Little League"), compute is supply-constrained, renewals are pricing up, and therefore revenue scales roughly with gigawatts online. Management has said this on both public calls, with numbers attached each time.

Sources: transcripts/SPCX_2026Q2.txt, transcripts/SPCX_2026Q3.txt.

What "$30 to $50 per watt" means (unit check against both transcripts). Neither call states a period. Three facts pin it down as an annual revenue rate on rented capacity, not lifetime value and not a blended rate on all nameplate: (1) the CFO pairs it with "a payback of less than a year" on new deployments, which only works if $30 to $50 is earned per year against all-in capex of roughly that size per watt; (2) the filed cloud contracts (424B4 terms as summarized in the house file research/ai-economics/debate/cloud_rev_per_mw.md) imply about $24 per IT watt per year for Anthropic ($1.25B a month for about 325k GPUs) and about $51 per IT watt per year for Google ($920M a month for about 110k GPUs), so the newest deal sits "slightly above the high end" exactly as the CFO said; (3) Musk frames it per watt "of, say, Rubins", a forward chip, which is a deal-level rate, not a fleet average. The first draft's error was applying that rented-IT-watt annual rate to all incremental nameplate watts for a full year. Half the compute is used internally, nameplate runs about 1.4x IT, and capacity bills only after it is energized.

Correction to the first draft: the "$4.1B per GW" figure is per quarter, not annualized (about $16.5B per GW per year), and it is consistent with the $30 to $50 range only for externally rented IT watts, not on a blended basis.

FY2027 revenue bridge on observed economics ($B). Start-2027 nameplate about 2.75 GW (midpoint of the 2.5 to 3.0 GW year-end target). The company target adds +10 GW during 2027; energized evenly, that is about 5.0 GW of average incremental capacity in service, or about 3.75 GW with a one-quarter billing lag. Non-infrastructure revenue (Connectivity about $26B, Space about $6B, X advertising about $2B, Cursor about $5B) is about $39B, my estimate.

FMP consensus has two readings. The annual mean is $151.0B (29 analysts). The sum of the 2027 quarterly means is $175.0B ($32.0B, $39.3B, $48.0B, $55.8B; only 4 to 9 analysts per quarter). The analysts who have published quarterly 2027 models since the Q3 guide sit about $24B above the annual mean, which suggests the annual mean still carries pre-guide estimates. Revision history itself is not obtainable: FMP /stable has no estimate-history endpoint, and Visible Alpha and Bloomberg are not connected.

Read-through. On observed Q3 economics (Case A), consensus is roughly right. The gap exists only if the Q1 2027 guide's implied yield (about $35 per nameplate watt, roughly double Q3) holds across all of 2027's new capacity (Case B, about +$115B to +$160B). That is still a real gap, but it is about a quarter to a third of what the first draft wrote, it depends on spot pricing holding while 10 GW is added, and the analysts who model quarterly are already closing it. The one hard, CFO-backed data point is near term: Q1 2027 consensus of $32.0B sits below the $33B to $37B guide.

What the street models (FMP consensus):

FY2028 dispersion ($110B to $1,042B) confirms the street has no shared view of the out-years.

Reported trajectory that supports management's case (Daloopa):

Why the trend is accelerating: AI revenue grew about 9.3x YoY in Q3 (about 8.6x ex Cursor) because cloud services agreements (management names Google and Anthropic as customers on the Q2 call) ramped on Colossus and Colossus II capacity, plus about $608M from Cursor (closed mid-quarter). AI solutions revenue per nameplate watt rose from about $2.4 a year in Q3 2025 to about $16.5 a year in Q3 2026 (+591% YoY), and AI segment EBITDA swung from -$428M to +$4,953M. Backlog more than doubled versus FY2025 year-end and the near-term share rose to 67% (about $42B within a year), so the near-term ramp is contracted, not hoped for.

Capex: confidence signal or red flag? Management framed Q2 capex as "next 2 quarters are very similar" to the $18.4B Q2 level. Q3 came in at $19.6B, and Q4 is now guided to $35B to $40B, roughly double. It does show management believes its own compute numbers. But under the framework the "increasing revenue with declining FCF" red flag matters more: revenue rose YoY in each 2026 quarter while FCF fell YoY in each (-$9.1B vs -$3.4B, -$16.0B vs -$3.2B, -$10.0B vs -$3.1B), and the capex guide doubling within a quarter is a guidance-accuracy miss (Dim 3). The composite treats this as the red flag, and so does this dimension.

Where the street pushes back (Q&A and published notes):

• Funding: Citi (Godyn, Q2) and JPMorgan (Anmuth, Q3) both asked how the build is financed. Morgan Stanley's initiation flagged a $672B funding need and no FCF before 2035 (ii.co.uk). Post-Q3 coverage cites funding the 2027 build as the main concern (leverageshares).
• Durability of pricing: Anmuth (Q2) asked whether premium compute pricing persists. Management's answer (short contracts, renewals at higher $/W) is the crux the street does not yet underwrite, and it is exactly the Case A vs Case B gap above.
• Starlink Mobile: Evercore (Maral) on both calls questioned whether the spectrum portfolio can compete with incumbents; Musk responded that analysts are "missing some of the pieces" and promised a separate presentation.
• Execution: Susquehanna stays Neutral citing execution risk (teslarati).

Why this is not a "management is the contrarian" setup with a great track record: the street is not ignoring the AI story; it is discounting the speaker. Musk's Q2 timelines already slipped by Q3: the Starship ship catch "as soon as the next flight" did not happen on Flight 14; "at least 1 flight a day" a year out became "weekly in 1H27, twice weekly by end of 2027." The $500B to $1T ARR figure was delivered with Musk himself saying "on that crazy note." The CFO put his name only to the $100B ARR, the $33B to $37B Q1 guide and the $30 to $50 per watt pricing. The principle says the setup is better when management has a great track record; Dim 3 scores this one 4/10 on five months of public history, so the analogy to a long-credible management team does not hold. The credible divergence is the CFO-backed near-term numbers versus a consensus that is still below the Q1 guide, plus the Case B pricing question for 2027.

Divergence sub-score: 5/10 (was 7). Specific, repeated and quantified, and the Q1 2027 consensus sits below the CFO's guide. But on observed economics the FY2027 gap is close to zero, the bullish case needs about a doubling of realized $/W to hold through a 10 GW build, the quarterly-model analysts are already about $24B above the annual mean, and the speaker carries a known timeline discount.

4b. Analyst consensus

FMP's current grades consensus is 17 Buy, 4 Hold, 0 Sell (81% Buy) on a narrower set. Post-Q3 actions were Overweight reiterations at Cantor, Barclays and Morgan Stanley and a Neutral at Susquehanna (leverageshares). That is the "already loved by the sell side" profile the rubric penalizes. The offset is that the Buy-rated analysts' annual revenue mean does not reflect the Q1 2027 guide, so some of the bullishness is in the rating, not yet in the numbers.

4c. Insider activity

No insider is buying during the divergence. The confirming signal the rubric looks for is missing, and the flow is net supply.

4d. Retail and social sentiment

Retail attention is about as high as it gets for any listed equity: roughly 20% of the IPO was allocated to retail across Robinhood, Schwab, Fidelity, SoFi and E*TRADE, with every eligible requester receiving shares (Bloomberg Gov); Musk livestreams the earnings call on X and polls retail questions; tokenized SPCX products trade offshore. Prominent skeptics (Chanos, Burry, Gerber) keep the name viral on both sides (ii.co.uk). This is the opposite of the low-attention setup the rubric rewards.

Scoring build

Assessment. SPCX has a real but narrow management-street gap. The CFO guided Q1 2027 revenue to $33B to $37B on a backlog of $63.0B that is 67% due within a year, and Q1 2027 consensus ($32.0B) still sits below the low end. But the first draft's FY2027 argument rested on a unit error. Management's $30 to $50 per watt is an annual rate on rented IT watts on the newest contracts; the realized blended yield on all nameplate was about $16.5 per watt per year in Q3. Applied to the 2027 build with realistic energization, observed economics give FY2027 revenue of about $146B to $167B against a $151B consensus, so the gap opens only if the roughly doubled yield implied by the Q1 guide holds across 10 GW of new capacity (about +$115B to +$160B). The analysts who publish quarterly 2027 models are already about $24B above the annual mean. Around that, the sentiment profile is crowded: about 80% Buy, insiders selling into staged lockups with no buying, and retail attention at a maximum. Fundamentals are going from worse to better, but sentiment already got there first, and the capex that signals conviction is also the red flag. Score: 4/10 (was 5).

Data sourced from Daloopa.