Executive Summary
Artificial intelligence is repricing the economy in real time. U.S. labor share of GDP fell to 53.8% in Q3 2025, the lowest reading since 1947, while nonfarm productivity growth surged to 4.9% annualized. Corporate profits hit record highs as hiring slowed sharply. The central question is whether abundant machine intelligence will produce a deflationary boom that raises living standards, or a displacement spiral that collapses consumer demand faster than markets can adapt.
Core finding:
Current data favors a gradual transition (5–15 years) over an acute crisis (2–3 years). But markets are already imposing crisis-speed repricing on software and intermediation businesses, creating real winners and losers regardless of which macro scenario materializes.
Three dynamics are occurring simultaneously:
- A structural repricing of SaaS and white-collar-dependent business models that is already well advanced, with $285 billion in single-day market cap destruction following AI product launches.
- A genuine productivity acceleration that is real but concentrated in a narrow set of industries, with gains running roughly 2.5 percentage points above the pre-pandemic trend.
- A speculative narrative about macro collapse that has limited supporting data: unemployment remains 4.3%, initial claims are 212K/week, and adoption intensity shows no inflection.
53.8%
Labor Share of GDP
Record low since 1947. Down from 55.6% decade average. Productivity gains accruing to capital, not labor.
38%
AI Adoption (Flat QoQ)
Gallup Q4 2025. Half of U.S. workers have never used AI at work. Adoption is linear, not exponential.
4.3%
Unemployment
Initial claims 212K/week. Cooling but not cracking. Displacement is a risk, not a current reality.
What the Data Actually Shows
The White-Collar Labor Market
The white-collar recession is real and predates current AI capabilities. Finance, insurance, information, and professional services have cut jobs on net for three years despite solid GDP growth. If pre-pandemic hiring trends had continued, these sectors would employ 2.3 million more workers.
- White-collar job postings fell 35.8% between Q1 2023 and Q1 2025 (Revelio Labs), with software developers declining at roughly double the overall rate.
- LinkedIn reports a 32% drop in hiring above $125K. Junior tech hiring is down approximately 40% from pre-2022 levels.
- Employers added just 584,000 jobs in all of 2025, down from 2 million in 2024. January 2026 revisions showed average monthly gains of only 15,000.
- However, headline unemployment remains 4.3% and blue-collar sectors (healthcare, construction, trades) continue adding jobs. Weakness is concentrated in back-office roles, recruiting, HR, and mid-level management.
Productivity
U.S. labor productivity is accelerating. Nonfarm business productivity grew 4.9% annualized in Q3 2025, the strongest pace in years. The Kansas City Fed finds the post-2022 pickup runs approximately 2.5 percentage points above the pre-pandemic trend.
- The productivity-pay divergence is widening: Q3 2025 productivity growth was 4.9% annualized while real wage growth remained modest. Fortune 500 profits hit a record $1.87T in 2024.
- The Kansas City Fed finds these gains are not yet broad-based. A small set of industries (computer systems design, online retail, data processing, management of companies) accounts for most of the pickup.
Ghost GDP: Output that never circulates through consumer paychecks. The embryonic form of this dynamic is already visible in the productivity-pay gap. If it deepens, it becomes a drag on consumption and a structural headwind for consumer-facing businesses.
AI Adoption: Linear, Not Exponential
AI adoption is widespread but shallow. The gap between adoption rates and realized economic effects is the single most important data point in this debate.
- Gallup Q4 2025: only 38% of employees report organizational AI integration, essentially unchanged quarter-over-quarter.
- OECD: only 20.2% of firms globally use AI. U.S. Census BTOS: just 5.4% of businesses used AI in the previous two weeks.
- Morgan Stanley survey of 935 executives: 11.5% average productivity gains and 4% net headcount reduction, but over 80% of firms report no measurable impact on employment or productivity over the past three years.
Recursive technology capability does not equal recursive economic deployment. Every source in this analysis — bull, bear, and neutral — agrees that AI capability will continue advancing rapidly. The disagreement is entirely about how quickly that capability translates into real economic displacement.
SaaS and Software Markets
The SaaS repricing is severe and well-documented.
- The IGV ETF fell over 23% year-to-date by mid-February, its worst stretch since 2008.
- Anthropic product launches triggered approximately $285 billion in single-day market cap destruction. Salesforce is down roughly 40% from 2025 highs.
- Forward P/E ratios have compressed from roughly 35x to 20x, levels not seen since 2014.
- Apollo cut its direct lending funds' software exposure nearly in half during 2025.
However, SaaS growth had decelerated every quarter since the 2021 peak, well before AI fears catalyzed the selloff. Bain reports gross retention still around 90% or better. The selloff reflects both real structural concern and a repricing of long-overdue valuation excesses from the ZIRP era.
Business Formation and Demographics
- New business applications remain elevated at approximately 532,000/month (January 2026), well above pre-pandemic levels, with projected formations up 4.5% month-over-month.
- BLS projects 17% software engineering growth through 2033 (327,900 new U.S. jobs). AI/ML and cybersecurity postings surged 163% and 124% YoY respectively.
- U.S. fertility sits at 1.62 (well below 2.1 replacement), population over 65 rising from 17.9% to 21.2% by 2035, and the U.S.-born labor force is growing just 0.3%/year.
AI may be offsetting a structural labor shortage, not creating a surplus. The labor market is transforming, not vanishing.
Scenario Framework
The debate maps onto four scenarios with distinct probability weightings and portfolio implications. None is certain; portfolio construction should account for all four.
| Scenario | Prob. | Key Mechanisms | Market Implications |
|---|---|---|---|
| Gradual Transformation | 50–65% | AI diffusion follows S-curve; institutional friction slows adoption; deflation raises purchasing power gradually; labor market restructures over 5–15 years | Software repricing continues but moderates; broad equity market grinds higher on productivity; value rotation favors physical-world assets |
| Abundance Boom | 20–30% | AI deflation creates massive consumer surplus; business formation surges; new industries emerge from science breakthroughs; demographic cliff makes AI complementary | Equity bull market broadens; infrastructure and energy outperform; compute layer captures outsized returns; consumer discretionary benefits |
| Rapid Displacement | 10–20% | White-collar displacement accelerates into negative feedback loop: layoffs → reduced spending → more AI adoption → more layoffs. Private credit contagion. Ghost GDP dynamics dominate | Broad equity drawdown of 25–40%; software destroyed further; Treasuries rally; gold outperforms; compute partially insulated |
| Stalled Diffusion | 5–15% | Regulatory pushback, compute bottlenecks, trust barriers, or capability plateaus slow deployment. AI capex generates disappointing ROI | Resembles prior tech hype cycles. Modest productivity gains. AI infrastructure retains some value |
The gradual transformation is the base case because: (1) adoption intensity data shows no inflection, (2) productivity gains remain concentrated in a narrow industry set consistent with early diffusion, (3) headline labor markets have not broken, and (4) two centuries of technology transitions have followed gradual S-curves rather than step-function shocks.
Sector-by-Sector Implications
High-Conviction Overweights
AI Infrastructure and Compute
OverweightCompute demand exceeds supply across all scenarios except the most extreme stalled-diffusion case. Hyperscalers are on track to spend $660–690 billion on infrastructure in 2026, nearly doubling 2025. The real bottlenecks are power, land, data center shells, and advanced memory. Even in the bear case, AI infrastructure continues posting record revenues as companies shift spend from labor to compute. Semiconductors, power generation/transmission, data center REITs, and cooling technology are all structurally supported.
Energy and Power Infrastructure
OverweightData center electricity demand is an observable, near-term tailwind for utilities, nuclear, natural gas, and grid infrastructure. This sector benefits under all macro scenarios and is the closest thing to a consensus overweight across the entire debate. The Jevons Paradox applies: AI efficiency creates more demand for physical infrastructure, not less. The energy transition is a multi-decade tailwind.
Physical Economy and Skilled Trades
OverweightHealthcare, construction, advanced manufacturing, and skilled trades retain structural demand because AI struggles with physical-world dexterity. Blue-collar job openings have remained relatively stable even as white-collar postings collapsed. The demographic cliff reinforces demand for human-performed physical work. AI-driven design and optimization accelerates demand for physical buildout rather than substituting for it.
Healthcare and Life Sciences
OverweightAI-accelerated drug discovery is compressing 10-year pipelines into months. Each breakthrough creates manufacturing, distribution, and care delivery work. Demographic aging guarantees demand growth (healthcare AI adoption growing at 36.8% CAGR). Physical delivery of care resists automation. This sector benefits from both the technology tailwind and the demographic tailwind simultaneously.
Sectors Requiring Nuance
Traditional SaaS and Enterprise Software
UnderweightSeat-based models in horizontal SaaS face structural headwinds as enterprises consolidate vendors and compress licenses. However, deeply embedded systems of record (CRM, ERP) with high switching costs and proprietary data will survive at lower multiples rather than go to zero. Bain reports gross retention still around 90%. The Jevons Paradox also applies: demand for better software may be near-infinite. The opportunity is in identifying companies that transition to consumption-based or outcome-based pricing, own irreplaceable data assets, or serve as the mediation layer between AI models and enterprise workflows. Avoid long-tail horizontal SaaS with narrow functionality that AI can replicate.
Consulting and IT Services
UnderweightAI directly substitutes for the core value proposition: navigating complexity humans find tedious. India IT services ($200B+ in exports) face a structural headwind as the marginal cost of AI-generated analysis approaches the cost of electricity. Already visible in slowing hiring at major firms. Geographic concentration risk (India, Philippines) is underappreciated.
Consumer Discretionary
NeutralThis sector is the most sensitive to which scenario materializes. The top 20% of earners drive approximately 65% of consumer spending. Even modest displacement of this cohort has outsized consumption impact, with a 2–3 quarter lag as savings buffers deplete. If AI deflation raises purchasing power (bull case), consumer discretionary benefits enormously. If white-collar displacement outpaces price declines (bear case), it suffers first. Position neutrally and watch white-collar jobless claims data closely.
Financial Intermediaries
Underweight SelectivelyCompanies whose moats depend on consumer friction, habit, or information asymmetry face genuine long-term pressure from AI agents. This includes elements of payments (interchange fees), insurance distribution, real estate brokerage, travel aggregation, and gig-economy platforms. However, the timeline is longer than bears suggest. As of early 2026, there is no evidence of meaningful agent-led payment disintermediation at scale. Reduce exposure to pure friction-based moats but recognize this is a 5–10 year structural risk, not an immediate catalyst.
Tail Risk Hedging
While the rapid-displacement scenario is low probability, a potential GFC-scale drawdown warrants dedicated hedging. Key instruments and transmission mechanisms:
Treasuries and Duration
In the bear case, bonds rally as the Fed cuts aggressively into a deflationary displacement spiral. Technology-driven disinflation should keep real rates contained even outside the bear case. Modest duration exposure serves as portfolio insurance against deflationary demand collapse.
Gold
If displacement reaches crisis proportions, governments will respond with massive fiscal stimulus, UBI proposals, or aggressive money-printing. Gold hedges the tail scenario where governments attempt to inflate their way out of an AI-driven deflationary crisis, and performs well if geopolitical instability rises during the transition.
Private Credit Exposure (Reduce)
PE-backed software deals from the 2021–2023 vintage with seat-based revenue models represent concentrated risk. Apollo's decision to cut software exposure in half is a signal. Monitor credit spreads on PE-backed software and technology paper; Zendesk-style defaults are the canary. Audit software-sector concentration and vintage exposure immediately.
Prime Mortgage Deterioration (Monitor)
Monitor early-stage delinquency in tech-heavy ZIP codes (SF, Seattle, Austin, NYC). HELOC draws, 401(k) withdrawals, and credit card balances spiking while mortgages remain current are leading indicators. The 30-year fixed-rate structure provides substantial cushion, making this a watch item rather than an immediate position.
Forward Monitoring Framework
The debate will be resolved by data over the next 6–18 months. Rather than making binary bets, the optimal approach is to monitor leading indicators and adjust positioning as evidence accumulates.
| Indicator | Current | Bear Trigger | Source / Freq. |
|---|---|---|---|
| Labor share of GDP | 53.8% (Q3 2025) | Sustained below 52% | BLS / Quarterly |
| AI adoption (daily use) | ~19% of workers | Step-function jump above 35% | Gallup, St. Louis Fed / Quarterly |
| Org. AI integration | 38% (flat QoQ) | Jump above 50–60% | Gallup / Quarterly |
| Initial jobless claims | 212K/week | Sustained above 350K | DOL / Weekly |
| White-collar JOLTS | Declining YoY | Collapse below 4M total | BLS / Monthly |
| Software credit spreads | Elevated but contained | PE software defaults >3% | Moody’s / Ongoing |
| SaaS net revenue retention | Gross ~90% | Gross drops below 80% | Quarterly Earnings |
| New business applications | ~532K/month | Sustained YoY decline | Census Bureau / Monthly |
| Productivity breadth | Narrow (top 5–6 industries) | Remains narrow 12+ months | KC Fed QILP / Quarterly |
| Inference cost / token | Declining ~40% YoY | Decline accelerates >70% YoY | AI Lab Pricing / Ongoing |
Portfolio Construction Principles
The AI economic transition creates a portfolio construction challenge that is unusual because the range of outcomes is genuinely wide, the dominant scenario requires patience, and the tail scenarios would reward very different positioning. Six principles emerge from the evidence:
Do not position the entire portfolio for one scenario.
The honest assessment is that nobody knows whether AI will produce a 2028 crisis, a 2028 boom, or (most likely) a muddled middle. Maintain core equity exposure for the base case while using satellite positions and hedges to capture tail outcomes.
The software repricing is a sector story, not a macro story — yet.
The SaaS selloff is well-advanced and real. But the leap from "software margins are compressing" to "the consumer economy is collapsing" requires several additional links that have not materialized. Be precise about which thesis you are betting on.
Overweight what benefits under all scenarios.
Energy infrastructure, compute supply chain, and physical-world-intensive sectors benefit whether AI produces a boom or a crisis. These are the highest-conviction allocations because they do not require predicting which macro scenario wins.
The speed of adoption is the critical variable.
If Gallup’s organizational AI integration metric jumps from 38% to 60%+ in the next two quarters, the bear case timeline becomes much more credible. If it continues its gradual drift, the base case holds. This single indicator is more informative than any earnings call or market commentary.
Deflation is the underappreciated macro force.
Technology-driven disinflation in services (70% of consumer spending) is coming regardless of pace. A 2–3% sustained decline in service costs functions as a tax cut for every household in the economy. Position for a world where nominal growth may be modest but real growth is strong.
Audit private credit exposure immediately.
PE-backed horizontal SaaS from the 2021–2023 vintage is where concentrated risk is already identifiable. This is not speculative; it is reflected in Apollo’s portfolio actions. Understand your software concentration and vintage profile.
Key Risks to This Framework
This report's base case assumes gradual transformation. The following developments would invalidate that assumption and require rapid portfolio repositioning:
- Non-linear adoption breakpoint. If a consumer AI agent achieves mass adoption (analogous to the iPhone moment), diffusion curves could steepen dramatically. Watch for any single AI product reaching 500M+ daily active users within a 12-month period.
- Recursive AI self-improvement. If AI systems begin meaningfully accelerating their own capability development — not just training efficiency, but qualitative capability jumps — all timeline assumptions collapse. This remains speculative but cannot be dismissed.
- Energy or compute breakthrough. Fusion or other step-function energy cost reductions would remove the natural cost floor on AI deployment that currently constrains substitution speed.
- Geopolitical AI race dynamics. U.S.–China competition could override domestic economic caution, pushing both nations toward aggressive AI deployment regardless of labor market consequences.
- Regulatory failure or capture. If policy response is absent during rapid displacement or counterproductive through incumbent entrenchment, the transition risk magnifies dramatically.
Conclusion
AI is producing a real and measurable shift in the distribution of economic returns from labor to capital. This is confirmed by record-low labor share, surging productivity, and slowing job creation. However, the pace of actual workplace deployment remains firmly linear, not exponential, and the overwhelming weight of evidence favors a 5–15 year transformation over a 2–3 year crisis.
The optimal portfolio stance is to position for gradual capital-over-labor rotation — overweight AI infrastructure, physical economy, energy, and healthcare; underweight labor-intensive intermediation and undifferentiated software — while maintaining dedicated tail-risk hedges against the displacement spiral scenario.
Position for the base case. Hedge for the tail. Monitor relentlessly.
Sources
- Citrini Research & Alap Shah
- Citadel Securities (Frank Flight)
- Colin McNamara
- Seb Krier
- Loeber
- Galois
- Dwarkesh Patel & Brian Albrecht
- Sleepysol
- Kobeissi
- Michael Bloch
- Jason Lemkin & Roy O'Driscoll
- Morgan Stanley
- Bain & Company
- Goldman Sachs
- Apollo Global Management
Data: BLS Employment Situation (Feb 2026), BLS Productivity & Costs (Q3 2025), Gallup Workplace AI Survey (Q4 2025), OECD AI Adoption (Jan 2026), Indeed/FRED Job Postings, Kansas City Fed QILP, St. Louis Fed RTPS, U.S. Census Bureau Business Applications, Revelio Labs, LinkedIn Economic Graph.
This report is for informational and educational purposes only. It does not constitute investment advice, a recommendation, or an offer to buy or sell any security. Past performance does not guarantee future results. Consult a qualified financial advisor before making investment decisions.