My speculative theses on where the economic measurement apparatus may fail before the underlying technologies fully deploy. These are predictions backed by data, not factual claims about the present. Where I lean, and what would change my mind.
Where I lean: the 10–15 year window where measurement instruments may stop working before the economy has adjusted to the underlying disruption is where I see the biggest macro mispricing. That window, not the technologies themselves, is what these four forecasts try to map. Each thesis below states what I think, why, and what evidence would force me to update.
"Two things fill the mind with ever increasing wonder and awe: the starry heavens above me and the moral law within me." Immanuel Kant
Every tool in this table was designed for an economy where labour is human, capital is tangible, data costs power, encryption holds, and prices are set by independent agents. My forecast, the substance of these four theses, is that several of these assumptions don't survive intact through the next fifteen years. The status column reflects where I think each instrument sits today; reasonable readers will disagree on individual rows.
| Signal | Measures | Failure Mechanism | Status | Thesis |
|---|---|---|---|---|
| Jobless Claims LIVE → | Human unemployment | Robots displace silently through attrition, no claim filed, no signal sent | AT RISK | T03 |
| JOLTS Job Openings | Job creation rate | Robot placements disappear without a hire event, openings vanish, not transition | COMPROMISED | T03 |
| Wage Growth | Labour market tightness | Composition effect: displaced low-wage workers exit, raising average wage even as market loosens | BLIND | T03 |
| Phillips Curve | Inflation–unemployment tradeoff | Option value of automation flattens the curve, workers accept wage restraint rather than risk replacement | COMPROMISED | T03 |
| PMI Employment | Factory output conditions | Automated facilities suppress the employment sub-index permanently, the signal goes structurally negative | AT RISK | T03 |
| CPI / Core PCE LIVE → | Price level | AI-driven deflation and quality-adjusted price gains uncaptured in hedonic methodology | AT RISK | T04 |
| Bond Settlement Risk | Counterparty / credit risk | Quantum cryptographic vulnerability entirely unpriced, RSA/ECC breakage converts systemic infrastructure risk to zero-premium credit event | UNPRICED | T01 |
| Utility Demand Forecasts | Power consumption growth | Cold storage electricity demand absent, glass and DNA storage convert perpetual opex to one-time capex, stranding demand projections | OVERSTATED | T02 |
| GDP / TFP | Productivity and output | $4.7T annual intangible investment invisible in national accounts, Brynjolfsson finds true TFP 15.9% above official measures | UNDERSTATED | T04 |
| EBITDA | Corporate profitability | Robot capex conversion distorts the multiple, labour opex disappears, capex spikes, D&A rises, EBITDA inflates even as true economics deteriorate | DISTORTED | T03 |
| Price Discovery | Information aggregation | AI homogeneity degrades signal-to-noise, when 78% of institutions use similar models trained on similar data, prices stop discovering information | DEGRADING | T04 |
My forecast: migration capex from RSA/ECC to post-quantum cryptography is large, lumpy, and nowhere visible in current sovereign credit pricing or bank capex guidance.
Q-Day forecast: ~2030 · No observable migration pricingMy forecast: glass and DNA storage convert cold-tier data from perpetual opex to one-time capex. Goldman's +165% power demand projection assumes that conversion never happens.
Deployment: 2027–30 · 30–40% cold tier decouplesMy forecast: humanoid cost parity arrives 2028–2030. Acemoglu's 5.6-workers-per-industrial-robot displacement is the historical anchor; the Fed's labour instruments may not register the next wave.
Forecast crossing: 2028–30 · Unitree G1 ($16k) is the price-point previewNBER Working Paper 34054: two AIs autonomously generated supra-competitive profits without any agreement or intent to collude. The market doesn't have a category for this.
Algo Share: 60–75% · 240% more flash crashesWhat's at risk. Fedwire settles $4–5 trillion daily. SWIFT connects 11,000+ financial institutions across 200+ countries. TARGET2 is the backbone of the Eurozone. Every one of these systems runs on RSA or ECC encryption. Shor's algorithm breaks both. When sufficient qubits exist, and the timeline is accelerating, the cryptographic infrastructure underpinning global financial settlement becomes exploitable.
Harvest Now, Decrypt Later. Nation-state actors and sophisticated adversaries are already collecting encrypted financial data today with the intent to decrypt it once quantum capability arrives. This is not theoretical, CISA, NSA, and NIST have all issued formal warnings. The data being harvested today has a shelf life of 10–30 years. The decryption capability arrives in roughly 5–8.
Migration is nearly impossible at scale. BIS Project Leap documented that a TARGET2-equivalent system requires 5–10 years to migrate to post-quantum cryptography. The NIST mandate (IR 8547, November 2024) requires phase-out of RSA and ECC by 2030, complete elimination by 2035. The institutional investment community has not begun pricing the migration cost, the operational risk, or the systemic disruption.
Capex signatures confirm the thesis. IBM has committed $30B to quantum. PsiQuantum is raising $7B+ with TSMC partnership for silicon photonic chips. Quantinuum secured $10B in capital commitments from JP Morgan and Microsoft. These are not research budgets. These are commercial timelines.
"My forecast: the migration capex needed to move global financial settlement to post-quantum cryptography is large, lumpy, and currently invisible in any price I can see, neither in sovereign credit spreads, nor in bank capex guidance, nor in equity multiples for institutions running this infrastructure. What would change my mind: a single G-SIB issuing a quantitative PQC migration capex disclosure, or a sovereign issuer pricing a quantum-resistant tranche at a measurable basis differential."HORIZON 2040 · THESIS 01 (FORECAST, NOT PRESENT-STATE CLAIM)
NIST IR 8547 (November 2024): Begin phase-out of RSA/ECC by 2030. Complete elimination by 2035. This is a government mandate, not a suggestion, and most financial infrastructure hasn't started.
⚠ GAP HAS CLOSED, Mosca's Theorem states: if (migration time) + (time to Q-Day) < (data shelf life), you are already exposed. For most long-dated financial data, the overlap is now. Migration must begin immediately.
The mechanism. Cold storage, archival data that is written once and rarely accessed, currently accounts for 30–40% of total data centre capacity. It requires continuous power to maintain. Glass storage (developed at Microsoft Research) can hold data for 10,000 years with zero power at rest. DNA storage offers similar permanence at declining cost ($0.06/MB trajectory). When these technologies reach commercial scale, the electricity demand for cold data permanently decouples from the grid.
The February 2026 breakthrough. Pyrex glass, a common material with established supply chains, was successfully demonstrated as a viable substrate for optical data storage at research scale. This resolves the primary supply chain objection to glass storage: exotic materials. Commercial deployment in the 2027–2030 window is now plausibly funded.
The accounting inversion. Today's data infrastructure is structured as recurring opex: lease the power, pay monthly. Glass storage converts that flow into a one-time capex event. Corporate CFOs understand this trade immediately. The rerating of hyperscaler power consumption forecasts will follow once pilot deployments are publicly announced.
The real estate distortion. Goldman Sachs projects data centre electricity demand rising 165% by 2030. That projection assumes all data requires power forever. Glass and DNA storage break that assumption. If 30–40% of storage demand permanently decouples from electricity, the demand wedge supporting utility re-ratings and data centre real estate valuations partially evaporates.
"My forecast: the utility re-rating since 2023 has been built on the assumption that all data requires power forever. Glass and DNA storage make that assumption falsifiable. What would change my mind: commercial-scale glass/DNA deployments slipping past 2030 with no credible cost path, or hyperscalers reaffirming cold-tier power forecasts after a public glass-storage pilot."HORIZON 2040 · THESIS 02 (FORECAST, NOT PRESENT-STATE CLAIM)
The measurement problem. Jobless claims require a fired worker. Robots displacing through attrition never trigger a claim, the role simply goes unfilled. JOLTS job openings require a posted vacancy. Robot placements generate no posting. Wage data shows composition effects: as lower-wage displaced workers exit, average wages rise even as the true labour market weakens. The BLS acknowledged in a 2024 GAO report that redesigning labour market measurement for automation requires 10–15 years. The Fed is flying blind.
The empirical evidence on industrial robots is unambiguous. Acemoglu and Restrepo's landmark study found each additional industrial robot in a US commuting zone displaces ~5.6 workers per thousand and reduces area wages by 0.42% (the original 2020 paper figure; later updates around 6.0). Econometrica (2022) documented that automation accounts for 50–70% of the observed rise in US wage inequality over the past 30 years. Bain & Co report a 40% decline in industrial robot unit costs between 2022 and 2024 alone.
The forecast extension is where I'm out on a limb. The Unitree G1, a commercially available bipedal platform, retails at $16,000. Caveat I want to be honest about: the G1 has no commercial labour deployment today, runs ~2 hours per charge, and lacks general manipulation capability. A 2025 capex price-point on a research-grade platform is not the same as functional labour substitution. What I forecast: the price-performance curve of bipedal robotics over 2020–2025 implies a sub-$10k commercial-grade humanoid by ~2028–2030, at which point the cost-substitution argument starts to apply, first in narrow tasks (warehouse pick-and-pack, security patrol, basic facilities), then broader. The Acemoglu industrial-robot displacement pattern is the historical anchor; the humanoid extension is my forecast, not present-state.
Basso & Rachedi (JME, 2025) provide the most important recent theoretical contribution: the option value of automation flattens the Phillips Curve even before a single robot is deployed. Workers in sectors threatened by automation accept wage restraint rather than risk accelerating their own replacement. The curve doesn't just flatten after automation, it flattens in anticipation of it.
The Fed governor admission. Governor Lisa Cook warned in February 2026 that "our normal demand-side monetary policy may not be able to ameliorate an AI-caused unemployment spell without also increasing inflationary pressure." Governor Barr separately described a scenario where AI causes a "jobless boom", rapid structural displacement that rate cuts cannot fix, because the unemployment is structural, not cyclical. This is the FOMC acknowledging, in public, that its existing toolkit may be inadequate for what is coming.
"The Fed will hold rates too high for too long because it interprets the labour market as tight when it is structurally loosening. The unemployment is invisible. The wages are a mirage. The instruments are reading the wrong economy."HORIZON 2040 · THESIS 03
The NBER finding. Working Paper 34054 (Wharton + HKUST, July 2025) documented that two AI agents autonomously generated supra-competitive profits without any agreement, communication, or intent to collude. They were trained to maximise profit. They discovered, independently, that restraining competition was more profitable than competing. No human designed this. No human instructed it. The collusion emerged from the objective function alone. Traditional antitrust frameworks have no category for algorithmic tacit collusion.
The FSB concentration risk. The Financial Stability Board (2024) flagged that three or four model providers now underpin the majority of institutional AI trading infrastructure globally. A single model failure, whether from a training data error, a discovered exploit, or a regulatory forced shutdown, could cascade across asset classes simultaneously. The concentration risk is not in any single institution. It is in the shared cognitive architecture across all of them.
The Productivity Paradox and J-Curve. Official GDP accounts for approximately $4.7T in annual US intangible investment. Brynjolfsson (MIT) estimates true TFP is 15.9% above official statistics once correctly-adjusted quality improvements and digital goods are counted. The "Solow Productivity Paradox reloaded", you can see the AI everywhere except in the productivity statistics, is resolving, but slowly. The J-curve implies the productivity gains are real but backloaded. The market is pricing the trough, not the inflection.
The mechanism of degradation. When 60–75% of US equity volume is algorithmic, and when 78% of institutions use AI for trading decisions, and when those AI systems are trained on similar historical datasets and optimise for similar objectives, the independence condition required for price discovery fails. Prices stop discovering information and start reflecting the shared biases of the models generating them. This is not speculation, flash crash frequency has increased 240% over the past decade.
"Price discovery, the foundational function of capital markets, is degrading. Not because of bad models, but because of systemic homogeneity: too much capital, making too similar decisions, from too similar training data, on too similar timescales."HORIZON 2040 · THESIS 04
The same instruments I track in my macro dashboard. Here's what they're showing now, and what the research says about what they're actually measuring.
Quantum computing, permanent storage, humanoid robots, and AI reflexivity are four separate structural shifts. The thesis is not about any of them individually. It is about the 10–15 year window in which the measurement infrastructure, the instruments that price these risks, will fail before the technologies fully deploy. That window is the trade.
| Instrument | Thesis | Failure Mode | Market Pricing | Status |
|---|---|---|---|---|
| Jobless Claims / JOLTS | T03 | Robot attrition files no claims | Zero adjustment | AT RISK |
| PMI Employment Sub-index | T03 | Automated facilities suppress sub-index permanently | Recession signal misread as structural | AT RISK |
| Wage Growth | T03 | Composition effect inverts signal | Treated as tightness indicator | BLIND |
| Phillips Curve | T03 | Option value of automation flattens slope | Still embedded in Fed models | COMPROMISED |
| Bond Settlement Risk | T01 | RSA/ECC vulnerable to Shor's algorithm | No observable migration-capex pricing in current spreads | UNPRICED (my forecast) |
| Utility Demand Forecasts | T02 | Cold storage decouples from grid | +165% demand priced in | OVERSTATED |
| CPI / Core PCE | T04 | AI deflation + quality gains uncaptured | Inflation target treated as binding | AT RISK |
| GDP / TFP | T04 | $4.7T intangible investment invisible | Official statistics taken at face value | UNDERSTATED |
| EBITDA Multiples | T03 | Robot capex distorts comparables | Multiples applied to inflated EBITDA | DISTORTED |
| Price Discovery | T04 | AI homogeneity breaks independence condition | Market function assumed intact | DEGRADING |
"The risk is not prediction error. It's measurement error. And the window to see it clearly is closing."