The AI productivity J-curve says: you pay now, you profit later. Measured TFP drops first because spending on intangibles counts as a cost, not investment, in standard accounts. SpaceX in 2025–2026 is textbook J-curve territory — and the $322B Goldman Sachs AI figure is not wrong, it is simply measuring the right side of a curve whose left side has not yet been traversed.
1 The AI Productivity J-Curve: Framework
The AI productivity J-curve is a well-documented macroeconomic phenomenon: firms pay the full cost of transformation upfront, while the productivity gains arrive years later. Measured Total Factor Productivity (TFP) drops first because spending on intangible assets — software, data pipelines, organisational redesign — is expensed under GAAP rather than capitalised. The balance sheet looks worse before it looks better.[1]
Brynjolfsson, Rock and Syverson (2021) formalise this dynamic, showing that adjusting for intangibles related to computer hardware and software yields a TFP level 15.9 per cent higher than official measures by end-2017.[1] The same logic applies to any general purpose technology (GPT) — including AI — that requires large complementary intangible investments before its productivity benefits materialise.
SpaceX in 2025–2026 is textbook J-curve territory. The company is simultaneously absorbing the capital cost of Starship development, the integration of xAI and X Corp (completed February 2026), and the build-out of orbital data-centre infrastructure — all while reporting a net loss of $4.9 billion against $18.7 billion in revenue.[2] The $322 billion AI revenue figure cited by Goldman Sachs for 2030 is not wrong; it is simply measuring the right side of a curve whose left side has not yet been fully traversed.[3]
"You pay now, you profit later. The productivity paradox is not a failure of the technology — it is a failure of the accounting." — Growth accounting framework, Brynjolfsson, Rock & Syverson (2021)
2 SpaceX's Invisible Capex: The Intangible Stack
SpaceX's reported $40B/yr capex run-rate is only half the story. The other half consists of intangible investments that GAAP expenses immediately but growth accounting treats as capital formation. According to SpaceX's S-1 filing, the AI segment alone consumed $12.7 billion in capital expenditure during 2025 and runs compute infrastructure with a capacity of approximately 1.0 gigawatt.[4] Starship development has absorbed more than $15 billion in cumulative investment, with $3 billion spent in 2025 alone.[5]
| Intangible Bucket | SpaceX Examples | J-Curve Mechanism |
|---|---|---|
| Patents & IP | Raptor engine designs, Starship heat shield, laser inter-satellite links | R&D spend ($3B+ in 2025) hits P&L now; payoff only when Starship reaches 1,000 flights/yr |
| Model Weights & Code | Grok model weights, X platform code, Starlink user terminal firmware | Training costs expensed; value accrues only if enterprise adoption materialises |
| Proprietary Databases | X real-time feed for Grok, Starlink telemetry, launch reusability data | Data collection and labelling is a cost; TFP appears only when models trained on it generate revenue |
| Business Process Change | Shifting from launch-as-a-service to AI compute-as-a-service; $920M/mo Google contracts | Retooling sales, compliance, and contracting kills near-term margins |
| Organisational Capital | Integrating xAI + X + Starlink teams post-February 2026 merger | Coordination costs show up as G&A; synergy = TFP, but takes 3–5 years to materialise |
| Market Research / Pilots | Proving orbital DC latency is acceptable; Anthropic pilots at $1.25B/mo (180-day lease) | If pilots fail, cost is sunk — not an asset. 43% "No Go" probability per Morningstar |
The Anthropic compute deal — structured as a 180-day lease with a 90-day mutual cancellation clause — illustrates the speculative nature of the orbital data-centre thesis.[6] If the lease is not renewed, the associated infrastructure investment becomes a sunk cost rather than a productive asset.
3 Capital Deepening vs. TFP: 2025–2030
Standard growth accounting decomposes output growth into three sources: capital deepening (K/L), labour quality improvement, and TFP (the Solow residual). SpaceX's current phase is almost entirely capital deepening — and measured TFP is negative as a result. Goldman Sachs projects operating cash flow to plunge to negative $105 billion in 2029, only to rebound sharply to positive $72 billion by 2031[3] — a trajectory that maps precisely onto the J-curve's left-to-right transition.
Capital Deepening (2025–2028)
K rising at $40B/yr. L roughly flat. Output grows slower than K → measured TFP negative. Starship costs $15B+ cumulative; until flight rate >50/yr, $/kg is worse than Falcon 9.
TFP Inflection (2029–2030)
K flattens. Intangibles depreciate into output. Patents → 85% reusability → launch cost falls 10x. Grok trained on X + Starlink data → 70% gross margin inference. That is TFP.
4 Why AI Makes the J Deeper Than Rockets
The intangible-intensity of AI investment creates a structurally deeper J-curve than traditional aerospace capital expenditure. As Brynjolfsson, Rock and Syverson (2021) demonstrate, general purpose technologies require complementary intangible investments that are poorly measured in national accounts.[1]
| Dimension | Traditional Rockets | AI / Orbital Data Centres |
|---|---|---|
| Capital Type | Tangible K: launch pads, factories | Intangible K: model weights, data pipelines — not measured → TFP looks worse |
| Depreciation Rate | 10–20 years | 2–3 years for AI models → faster obsolescence = bigger upfront cost burden |
| Process Change | Fly more rockets | Customers must rewrite applications for orbital latency → adoption lag = productivity paradox |
| Market Validation | NASA contracts (known demand) | Convince AI labs to leave terrestrial infrastructure → speculative, 43% "No Go" probability |
Morningstar's 43% "No Go" scenario — implying an enterprise value of negative $81 billion — is precisely the statement that orbital data centres remain permanently in the intangible investment stage and never climb the right side of the J-curve.[8]
5 Three Tests for TFP Inflection
Using standard growth accounting, SpaceX's TFP turns positive only when all three of the following conditions are simultaneously satisfied:
-
1
Capex / Gross Profit < 1.0x In 2025, $40B capex against $6.58B EBITDA[2] yields a ratio of approximately 6.1x. This ratio must compress below 1.0x before capital deepening gives way to TFP generation.
-
2
Intangible Amortisation < New Intangible Investment The stock of databases and patents must be yielding revenue, not merely accumulating. When amortisation of existing intangibles exceeds new investment, the portfolio is generating rather than consuming.[1]
-
3
Revenue / Employee Rising Without K/L Rising This is the business process TFP signal. If Starlink and AI headcount remains flat while revenue grows 25x, the organisation has exited the J-curve. Goldman's $474B 2030 scenario[3] assumes all three tests pass. Morningstar's $67.5B base case[8] assumes Test 3 never passes for the AI segment.
Watch the Capex/Gross Profit ratio quarterly. A sustained move below 2.0x signals the beginning of the TFP inflection. A ratio still above 4.0x in 2028 confirms the bear case.
6 Valuation Implications: Goldman vs. Morningstar
The divergence between Goldman Sachs and Morningstar is not a disagreement about SpaceX's technology — it is a disagreement about when the J-curve inflects and whether the AI segment ever exits the left side. Goldman Sachs, acting as lead underwriter, forecasts AI segment revenue growing 388% year-on-year to $15.6B in 2026, reaching $34.5B in 2027, and $322B by 2030.[3] Morningstar, by contrast, pegs SpaceX's fair value at $780 billion — less than half the IPO target — citing unclear economics and competition from OpenAI and Anthropic.[8]
| Scenario | Source | 2030 AI Revenue | J-Curve Assumption |
|---|---|---|---|
| Bull | Goldman Sachs [3] | $474B total revenue | All 3 TFP tests pass by 2029 |
| Base | Goldman Sachs [3] | $322B AI revenue | Tests 1 & 2 pass; Test 3 partial |
| Base | Morningstar [8] | $67.5B AI revenue | Test 3 never passes for AI |
| Bear | Morningstar [8] | −$81B EV | Orbital DCs stay in intangible stage permanently |
Standard Chartered's cautious stance is methodologically sound[7]: until capex flattens and organisational integration completes, the model is measuring inputs, not TFP. The productivity paradox is visible in the hardware — Starships are flying — but not yet in the profit and loss account.
7 Bottom Line
SpaceX's $322 billion AI revenue figure is not a forecast error — it is a timing question. The J-curve framework provides a precise diagnostic for when that number becomes credible:
-
2025–2028 (Deep J-Curve): $40B/yr capex plus intangible burn equals negative TFP. Net losses persist even if revenue reaches $40B. SpaceX has posted cumulative losses of more than $37 billion since founding.[4] This is not a sign of failure — it is the cost of building the right side of the curve.
-
2029–2030 (Potential Inflection): TFP turns positive only if patents translate to reusability economics, databases translate to Grok revenue, and process change translates to enterprise contracts. Goldman projects a sharp cash flow rebound to positive $72B by 2031.[3] If any one of the three TFP tests fails, the Morningstar bear case becomes the operative scenario.
-
The Investor's Diagnostic: Track the Capex/Gross Profit ratio, intangible amortisation versus new investment, and revenue-per-employee. These three metrics will signal the J-curve inflection before it appears in headline earnings.
We see Starships. We do not yet see the profit. That gap is the J-curve — and closing it is the only question that matters for SpaceX's 2030 valuation.
References
- [1] Brynjolfsson, E., Rock, D. and Syverson, C. (2021) 'The Productivity J-Curve: How Intangibles Complement General Purpose Technologies', American Economic Journal: Macroeconomics, 13(1), pp. 333–372. Available at: https://www.aeaweb.org/doi/10.1257/mac.20180386 (Accessed: 13 June 2026).
- [2] Bradbury, R. (2026) '6 charts: SpaceX's S-1 financials', Yahoo Finance / PitchBook News, 20 May. Available at: finance.yahoo.com (Accessed: 13 June 2026).
- [3] Reuters (2026) 'Goldman Sachs expects SpaceX's AI revenue to surge 100-fold by 2030, FT reports', Yahoo Finance / Reuters, 4 June. Available at: uk.finance.yahoo.com (Accessed: 13 June 2026).
- [4] Lopopolo, A. (2026) 'SpaceX files IPO paperwork with SEC, targets Nasdaq under SPCX', Quartz, 20 May. Available at: qz.com (Accessed: 13 June 2026).
- [5] Grush, L. (2026) 'SpaceX Has Spent More Than $15 Billion on Starship Rocket', Yahoo Finance / Bloomberg, 20 May. Available at: finance.yahoo.com (Accessed: 13 June 2026).
- [6] Munene, K. (2026) 'SpaceX IPO Nears as Musk Rejects Valuation Cut Report and AI Lease Questions', Analytics Insight, 1 June. Available at: analyticsinsight.net (Accessed: 13 June 2026).
- [7] Khan, W. (2026) 'Standard Chartered says SpaceX, Anthropic, OpenAI IPOs will weigh on stock market', Invezz, 3 June. Available at: invezz.com (Accessed: 13 June 2026).
- [8] Morningstar (2026) 'SpaceX valued at just $780 billion by Morningstar, less than half its IPO target', Yahoo Finance / Morningstar. Available at: finance.yahoo.com (Accessed: 13 June 2026).
This article is prepared for informational purposes only and does not constitute investment advice. All figures sourced from publicly available filings and analyst reports as cited above.
