AI, Labor Displacement & US Office Demand: A Quantitative Analysis
AI, Labor Displacement & US Office Demand: A Quantitative Analysis
Data-driven assessment of how AI automation reshapes the US workforce and translates into office space obsolescence — with historical context
Executive Summary
Artificial intelligence is poised to reshape US office demand on a scale not seen since the Industrial Revolution, but we are only three years into a multi-decade transformation. This analysis synthesizes labor market data, office inventory figures, and historical parallels to quantify the displacement pipeline and its real estate implications.
The Scale of Exposure: Of the 163.7 million employed Americans, approximately 58–60 million are office-using workers. Goldman Sachs estimates AI can automate tasks accounting for 25% of all US work hours—equivalent to 300 million jobs globally. Applied to the US, this translates to a projected displacement of 6–7% of workers (9.8–11.5 million) over the next decade, or roughly 980,000–1,150,000 workers per year.
Office Space at Risk: The US has 5.5 billion square feet of office inventory, with 4.48 billion currently occupied (18.6% vacancy). Through two mechanisms—productivity-driven space efficiency gains (15–25% reduction) and direct worker displacement—AI threatens 1.2–1.85 billion square feet of demand over 10 years, representing 21–34% of total inventory. After netting AI job creation (~142,000 roles/year requiring ~25M sq ft), the analysis projects net demand destruction of 950 million to 1.6 billion square feet, or 95–160 million sq ft annually.
Current vs. Projected Impact: As of 2025, confirmed AI-driven job losses total only ~12,700 (versus 119,900 AI roles added), and workers aged 22–25 in high-exposure occupations have seen employment fall just 6%. But adoption remains early: 49% of workers never use AI at work, and only 16% report AI doing some of their tasks. The gap between today's modest displacement and the projected pipeline is enormous.
Historical Context Suggests a Long Horizon: E-commerce displaced retail workers for at least 22 years (1995–2017) before stabilizing, and net losses continue today. The Industrial Revolution imposed a 60-year "Engels' Pause" (1780–1840) during which productivity surged but real wages stagnated. AI has been a mainstream economic force for only 3 years since ChatGPT's launch in November 2022. If historical patterns hold, the real displacement is still ahead, not behind.
The Geographic Mismatch: AI job creation is hyper-concentrated in five metros (San Francisco, New York, Seattle, Boston, Austin), which account for an estimated 60–70% of new roles. Meanwhile, AI-driven displacement is distributed broadly across every metro with customer service centers, back-office operations, and administrative workforces. This geographic and skills mismatch means even AI hub cities will see net negative office demand due to productivity compression, while non-hub metros face job losses with no offsetting AI hiring.
The Creation-Displacement Gap: AI is creating ~120,000–142,000 new jobs per year, representing just 1.7% of total US job openings and covering only 12–15% of the projected annual displacement. Even if AI job creation triples, it would offset less than half of the displacement pipeline. Moreover, displaced administrative and customer service workers are "less suited" to AI engineering and data science roles, amplifying the skills mismatch.
Investment Implication: Office markets face a structural demand shock of 95–160 million sq ft per year—2 to 4.5 times the current annual absorption loss of 66.5M sq ft. Vacancy, currently at 18.6%, could reach 22–25% nationally by 2029 without aggressive supply shrinkage through demolitions, conversions, and adaptive reuse. Quality bifurcation will intensify: premium tech-enabled assets in AI hubs will stabilize as AI firms lease aggressively, while commodity office faces permanent obsolescence. The paradox of San Francisco—positive absorption from AI leasing yet 25–35% CBD vacancy—previews the national trajectory.
Bottom Line: We are in the opening chapter of a decades-long labor market transformation. In hindsight, this may prove an era of extraordinary growth—but history shows the pain comes first, and it lasts longer than real-time participants expect. For US office markets, the quantitative evidence points to 1+ billion square feet of structural demand at risk, bifurcated by quality and geography, with displacement risks front-loaded over the next 5–10 years.
Part 1: Worker Automation Displacement
1.1 How Many US Workers? How Many Are AI-Exposed?
Total US workforce: 163.7 million employed as of November 2025 [1][2]. Of these, approximately 93 million (~57%) are white-collar professionals [3].
Office-using employment — the subset that physically occupies office space — includes:
| Occupational Group | Employment | AI Task Exposure |
|---|---|---|
| Management | 20.9M | Moderate (strategic tasks shielded) |
| Business & Financial Operations | 9.6M | High (analysis, reporting) |
| Computer & Mathematical | ~5.5M | Very High (75% task coverage for programmers) |
| Office & Administrative Support | 18.5M | Very High (data entry, scheduling, filing) |
| Legal | ~1.3M | High (research, document review) |
| Architecture & Engineering | ~2.8M | Moderate-High (design, modeling) |
| Total office-using (est.) | ~58–60M | — |
Sources: BLS Occupational Employment Statistics [4][3]
AI exposure scale:
- Goldman Sachs estimates AI can automate tasks accounting for 25% of all US work hours — equivalent to ~300 million jobs globally [5][6][7]
- Two-thirds of US occupations have some level of AI task exposure [6]
- The most-exposed roles: computer programmers (75% task coverage), customer service representatives (~70%), data entry keyers (67%), financial analysts (significant) [8][9]
- Only 16% of workers currently say AI does some of their work; 49% say they never use AI at work [10][11]. Adoption is early.
- Goldman Sachs base case: 6–7% of workers displaced over the next decade = ~9.8–11.5 million workers if applied to the current 163.7M employed [5][7]
- Annualized: ~980,000–1,150,000 displaced workers per year over 10 years
Who is already feeling it:
- Workers aged 22–25 in high-AI-exposure occupations saw employment fall 6% between late 2022 and July 2025 — while overall employment in those same fields grew [8]
- BLS projects customer service representatives to decline 5.0% through 2033
- In 2024, confirmed AI-driven job losses: ~12,700 (vs. ~119,900 AI roles added) [13] — a 9.4:1 creation-to-destruction ratio, but early days
- White-collar job openings in professional services hit a decade low in January 2025 — down 20% year-over-year [14]
1.2 How Much US Office Space Exists?
| Metric | Figure | Source |
|---|---|---|
| Total US office inventory | ~5.5 billion sq ft (peaked Q4 2024) | Cushman & Wakefield [15] |
| National vacancy rate | 18.6% (Q1 2026) | CBRE [16] |
| Occupied office space | ~4.48 billion sq ft (5.5B × 81.4%) | Derived |
| Avg. sq ft per worker | 150–175 sq ft | Industry standard [17][18] |
| Implied workers housed | ~25.6–29.9 million (occupied space ÷ sq ft/worker) | Derived |
| Actual utilization rate | 55–65% on any given day (hybrid effect) | WorkInSync [19] |
Key insight: The US has 5.5B sq ft of office space but only ~26–30M workers actively using it at allocated density. Hybrid work already means that on any given day, only 55–65% of desks are occupied. This is the baseline from which AI-driven displacement compounds.
Top Metro Office Inventories:
| Metro | Office Inventory | Vacancy Rate |
|---|---|---|
| Manhattan (NYC) | ~420M sq ft | 2.9% (prime), higher overall [20] |
| Chicago | ~215–220M sq ft | Elevated [20] |
| Washington, D.C. | ~113M sq ft | Moderate [20] |
| San Francisco CBD | ~55M sq ft | ~25–35% [21][20] |
| Seattle | Large metro inventory | ~26% [21] |
1.3 How Much Office Space Becomes Obsolete?
Step-by-step derivation:
Step 1 — Task-level AI productivity gains:
Research shows 30–33% productivity gains per AI-assisted hour on tasks like coding, customer support, and document writing. However, only ~16% of workers currently use AI, and economy-wide productivity has risen just 1.1% when including non-users. We temper raw claims to 15–25% effective space reduction over 5–10 years. [5]
Step 2 — Apply to office-using workforce:
| Scenario | Space Reduction | Sq Ft Impact | Annual Impact |
|---|---|---|---|
| Conservative (15%) | 15% of 4.48B occupied | 672 million sq ft | ~67–134M sq ft/yr (over 5–10 yrs) |
| Moderate (20%) | 20% of 4.48B occupied | 896 million sq ft | ~90–179M sq ft/yr |
| Aggressive (25%) | 25% of 4.48B occupied | 1.12 billion sq ft | ~112–224M sq ft/yr |
Step 3 — Add displacement effect:
Goldman Sachs projects 6–7% of workers displaced over 10 years = ~3.5–4.2M office-using workers (applying 6–7% to ~60M office workers). At 150–175 sq ft each:
- 525M–735M additional sq ft of demand destruction over 10 years
- Annualized: 52.5–73.5M sq ft/yr
Step 4 — Combined impact:
| Impact Channel | 10-Year Sq Ft Reduction | Annualized |
|---|---|---|
| Productivity/efficiency gains | 672M–1.12B sq ft | 67–224M sq ft/yr |
| Worker displacement | 525–735M sq ft | 52.5–73.5M sq ft/yr |
| Total potential | 1.2–1.85 billion sq ft | 120–298M sq ft/yr |
| As % of total inventory | 21–34% of 5.5B sq ft | — |
Reality check: Not all of this will materialize. Adoption is uneven, new AI firms lease premium space, and conversions/demolitions shrink supply. The existing document models 12–22% net demand decline as the most likely range. But the theoretical ceiling of 1.2–1.85B sq ft sets the outer bound of obsolescence risk.
For context: in the 12 months ending February 2024, the US office market recorded -66.5M sq ft of net absorption [22]. AI-driven demand erosion of 120–298M sq ft/yr would be 2–4.5× the current annual absorption loss — if it arrives in full.
1.4 City-Level Breakdown
AI's impact varies dramatically by metro, driven by (a) concentration of AI-exposed occupations, (b) existing vacancy, and (c) whether the city is a net creator or net loser of AI jobs.
| Metro | Office Inventory | Current Vacancy | AI Exposure Profile | Projected Sq Ft at Risk (est.) |
|---|---|---|---|---|
| Manhattan | 420M sq ft | 2.9% prime; higher overall | Finance, legal, media — high exposure; but AI firm leasing offsets | 60–85M sq ft (15–20%) |
| San Francisco | 55M sq ft (CBD) | 25–35% | Epicenter of AI creation and displacement; highest bifurcation | Already 14–19M sq ft vacant; further 5–10M at risk |
| Chicago | 215–220M sq ft | Elevated | Heavy admin/financial services exposure; limited AI job creation | 35–55M sq ft (16–25%) |
| Washington, D.C. | 113M sq ft | Moderate | Government/contractor roles; AI adoption lags private sector | 12–20M sq ft (11–18%) |
| Seattle | Large metro | ~26% | Dual pressure: AI HQ expansions (Microsoft, Amazon) vs. productivity displacement | Vacancy already extreme; modest net risk |
| Austin | Growing | Improving (100bps decline Q1 2026) | Net AI job creator but oversupply from building boom | Supply-driven risk > AI displacement risk |
The paradox of AI hubs: San Francisco, Seattle, and Austin are simultaneously the largest creators and largest destroyers of office demand. AI firms lease premium space aggressively — but the tools they build enable every other tenant in the city to need less. The net effect is severe quality bifurcation: Class A+ thrives, everything else hollows out.
San Francisco illustrates this perfectly: AI leasing drove +896,000 sq ft of positive absorption in Q1 2025 [23], yet overall CBD vacancy remains 25–35% — the highest of any major US metro [21].
Part 2: Historical Parallels
2.1 The E-Commerce/Internet Boom (1995–Present)
Timeline: Amazon launched in 1995. E-commerce began meaningfully displacing retail jobs by the mid-2000s. The displacement cycle has now run ~30 years and retail still hasn't recovered.
The numbers:
| Metric | Figure | Period |
|---|---|---|
| Retail jobs created by e-commerce FCs | ~400,000 | 2007–2017 [24] |
| Retail jobs destroyed per FC per county per quarter | 938 lost vs. 256 gained in logistics | Ongoing [25][26] |
| Net retail job losses projected | ~600,000 | 2020–2030 [27] |
| E-commerce share of retail sales | 7.4% → 16.4% | 2015–2025 [28] |
| Displaced retail worker income reduction | -2.4% | Per FC rollout [26] |
| FC wage premium over brick-and-mortar | +31% | But displaced workers don't get these jobs [24] |
Key lesson: E-commerce displaced retail for at least 22 years (1995–2017) before any stabilization appeared, and even then, net losses continue . The logistics/warehouse jobs that emerged paid better but employed different people in different places. Displaced retail workers "did not find a substitute, at least not in the short run" [25].
Office analogy: Just as e-commerce didn't eliminate all retail — it bifurcated it into experiential/prime (thriving) and commodity (dying) — AI will bifurcate office into premium tech-enabled (thriving) and commodity (structurally obsolete).
2.2 The Industrial Revolution (1760–1840+)
The Engels Pause: Between 1780 and 1840, the British economy experienced what historians call "Engels' Pause" — a ~60-year period where real wages remained stagnant despite surging productivity [30][31].
| Metric | Figure | Period |
|---|---|---|
| Productivity growth | 0.63–0.69% per year | 1800–1830 [30] |
| Real wage growth | ~0% (flat) | 1780–1840 [30] |
| Capital's share of income | Doubled | 1780–1840 [30] |
| Duration before workers benefited | ~60 years | 1780–1840 [30][32] |
| Iron/steel workforce growth | +1,200% | 1870–1910 [33] |
Key lesson: The Industrial Revolution did ultimately create massive net employment gains — iron workers grew 1,200% from 1870–1910 [33] — but it took decades of pain before workers, not just capital owners, benefited. During the pause, profits doubled while labor's share collapsed [30].
2.3 The Hypothesis: In Hindsight, These Were Growth Periods — But It Took a Long Time
We are now mapping a consistent pattern:
| Technology Disruption | Duration of Net Pain | When Growth Became Visible | Time to Net Positive |
|---|---|---|---|
| Industrial Revolution | ~60 years (1780–1840) | Post-1840 wage growth | 60+ years |
| E-Commerce | ~22+ years (1995–2017+) | Logistics jobs emerged ~2010s | Still net negative in retail |
| AI (so far) | ~3 years (Nov 2022–present) | TBD | TBD |
AI is approximately 3 years old as a mainstream economic force (dating from ChatGPT's launch in November 2022). By comparison:
- At the 3-year mark of e-commerce (~1998), Amazon had just barely turned a profit, and retail displacement was negligible
- At the 3-year mark of industrialization (~1763), the spinning jenny hadn't been invented yet
- At the 3-year mark of AI, we already see: 6% youth employment decline in exposed roles, 20% drop in white-collar job openings, and 25% of work hours theoretically automatable
AI appears to be moving faster than prior disruptions. The 3-year comparison is instructive:
| Metric at ~3 Years In | E-Commerce (1998) | AI (2025) |
|---|---|---|
| Technology penetration | ~1% of retail sales | 16% of workers use AI; 4.2% of job postings mention AI [11][34] |
| Measurable job displacement | Negligible | 6% decline in 22–25 yr old employment in exposed roles; ~12,700 confirmed losses [8][13] |
| Employer adoption | Early movers only | 49% of workers still never use AI — but adoption doubling yearly [10] |
| Projected total displacement | Not yet modeled | 300M globally / ~10M+ in US over decade [6] |
The implication for office demand: If historical patterns hold, we are in the very early innings of displacement. E-commerce's real damage to retail came 10–15 years in (2005–2010). The Industrial Revolution's labor share collapse took decades to play out. AI's current 12,700 confirmed job losses and 18.6% office vacancy may be just the opening chapter.
But the critical difference: AI's adoption curve is steeper. E-commerce required physical infrastructure (warehouses, logistics networks). AI requires only software deployment. Goldman Sachs warns that "a faster wave of adoption could lead to much larger economic disruptions" [6].
Part 3: AI Job Creation — The Other Side of the Ledger
3.1 How Many Actual New AI Job Openings?
Absolute numbers:
| Metric | Figure | Period | Source |
|---|---|---|---|
| AI-related positions (quarterly) | 35,445 | Q1 2025 | Veritone [35] |
| Annualized AI positions | ~142,000 | 2025 run rate | Derived (35,445 × 4) |
| AI roles added (net of losses) | ~119,900 created vs. 12,700 lost | 2024 | [13] |
| Gen AI-specific postings | ~66,000 | 2024 full year | Lightcast/Stanford [36] |
| Gen AI postings growth | 55 → ~10,000/month | Jan 2021 → May 2025 | Lightcast [37] |
| AI as share of all job postings | 4.2% | Dec 2025 | Indeed [34] |
| YoY growth in AI postings | +25.2% | Q1 2024 → Q1 2025 | Veritone [35] |
| AI mentions in job listings | +56.1% YoY | 2025 | Autodesk [38] |
Fastest-growing AI roles (Q1 2025): [35]
- AI/ML Engineer — 2,951 postings (+41.8% YoY)
- Data Scientist — 3,301 postings
- Big Data Engineer — growing
Putting it in context:
| Comparison | Annual Volume |
|---|---|
| AI jobs created (2024) | ~120,000 |
| AI jobs created (2025 run rate) | ~142,000 |
| Total US job openings (monthly avg.) | ~8.0–8.5 million |
| AI as % of all openings | ~1.7% of total |
| Goldman Sachs projected displacement | ~980,000–1,150,000/yr |
| Creation-to-projected-displacement ratio | ~0.12–0.15:1 |
The gap is stark. AI is creating ~120,000–142,000 new roles per year while the projected displacement pipeline is 980,000–1,150,000 per year. Even if AI job creation triples to ~400,000/yr, it would still cover only ~35–40% of projected displacement.
Additionally, the types of jobs differ. AI creates roles for ML engineers, data scientists, and AI researchers — highly specialized positions requiring advanced degrees. Displaced office workers (administrative assistants, customer service reps, data entry clerks) are "less suited to the kinds of labor that is most needed." This is the skills mismatch problem.
3.2 AI Job Creation by City
| Metro | AI Job Postings (Current Listings) | AI Share of Tech Openings | Key Employers | Concentration Notes |
|---|---|---|---|---|
| San Francisco Bay Area | ~7,900+ | 42% of all tech openings (up from 20% in mid-2022) | OpenAI, Google AI, Meta, Anthropic, Salesforce | Innovation epicenter; highest postings per capita [39][40] |
| New York City | ~8,200+ | High; 2nd largest metro for AI | Finance AI, healthcare AI, advertising AI | Diverse industry applications [41] |
| Seattle | 3rd largest metro | Large (Microsoft, Amazon) | AWS, Microsoft, Zillow | Cloud/enterprise focus [42] |
| Austin | Growing rapidly | Emerging | Google AI, Tesla | Highest cost-adjusted salaries ($252K) [42] |
| Boston | Top 3 hub | High | Research-focused; universities | AI/ML research concentration [39] |
| Washington, D.C. | Top 5 per capita | Moderate | Government/defense AI | Postings per 100K residents among highest |
The geographic concentration problem:
AI engineers cluster overwhelmingly in San Francisco, New York, and Boston [39]. AI consultants: same three cities. AI/ML researchers: same three cities. The top 5 metros likely account for 60–70% of all AI job creation, while the remaining ~380 US metros share the rest.
Meanwhile, AI-driven displacement is distributed broadly — every metro with customer service centers, back-office operations, or administrative workforces feels the impact. This geographic mismatch means:
- AI hub metros (SF, NYC, Seattle, Boston, Austin): Net positive job creation, but still net negative on total office space due to productivity compression
- Non-hub metros (most of America): Net negative on jobs AND space, with no offsetting AI hiring
Part 4: Synthesis — What This Means for US Office Demand
The Math, Summarized
| Input | Figure |
|---|---|
| Total US office space | 5.5B sq ft |
| Currently occupied | ~4.48B sq ft (81.4% occupancy) |
| Already functionally underutilized (hybrid) | ~1.6–2.0B sq ft (55–65% daily utilization of occupied space) |
| AI productivity-driven space reduction (10-yr) | 672M–1.12B sq ft |
| AI displacement-driven space reduction (10-yr) | 525–735M sq ft |
| Total AI-driven demand at risk (10-yr) | 1.2–1.85B sq ft |
| As share of total inventory | 21–34% |
| AI job creation offset (10-yr, generous) | 142K jobs/yr × 175 sq ft × 10 yrs = ~249M sq ft |
| Net demand destruction | ~950M–1.6B sq ft over 10 years |
| Annualized | ~95–160M sq ft/yr |
The Timeline Question
If historical parallels are any guide:
- Years 1–3 (2022–2025): We are here. Measurable but modest displacement. 18.6% vacancy already elevated from pre-AI trends (remote work). AI job creation outpaces confirmed losses 9:1. The narrative is "AI creates more than it destroys."
- Years 3–7 (2025–2029): Adoption accelerates. The 49% of workers who "never use AI" shrinks rapidly. Productivity gains compound. Entry-level hiring contraction deepens. Office vacancy could reach 22–25% nationally.
- Years 7–15 (2029–2037): Full displacement cycle plays out. Goldman Sachs' 6–7% displacement scenario materializes. Office inventory must shrink by 1B+ sq ft via demolitions, conversions, and adaptive reuse — or vacancy becomes permanent.
- Years 15+ (2037+): If history rhymes with the Industrial Revolution and e-commerce, this is when net positive effects emerge clearly — new industries, new job categories, new office demand drivers that don't yet exist.
The uncomfortable truth: In hindsight, every major technological disruption looks like a period of extraordinary growth. In real time, it felt like destruction. AI has been a mainstream force for just 3 years. The e-commerce parallel suggests the real pain is still ahead, not behind.
Sources: Bureau of Labor Statistics [1][2][4], Goldman Sachs Research [5][6][7], Cushman & Wakefield [15], CBRE [46][16][49], Anthropic Research [8], Pew Research Center [11], Gallup [10], Veritone [35], Lightcast/Stanford [37][36], Indeed Hiring Lab [34], Newmark [51], JLL [52][53], Allen (2009) "Engels' Pause" [30], NBER e-commerce research [26], Progressive Policy Institute [24], BLS Retail Trade Analysis [27], WorkInSync [19], YAROOMS [17], OMB [18], DPE-AFL-CIO [3], Business Insider [54], HeroHunt.ai [39], Youngstown State University [42]