Real estate, architecture & construction

Real estate, architecture & construction

Industry ID: 18 | Slug: real-estate-construction | Researched: 2026-09-15 | Analyst: agent

Research-capacity note. The shared 200-call WebSearch pool was exhausted after six searches in this sector. The remainder of the research was conducted by direct WebFetch on primary sources per the macro brief's standing instruction. Coverage of government statistics, trade-body data and brokerage research is consequently strong; coverage of company-level SEC filings and investor relations is weak, because those endpoints proved unreachable through the available tooling. See §13. This sector should be prioritised for re-run with search capacity restored.

1. Definition and boundaries

In scope. Commercial real estate across all property types (office, industrial, retail, multifamily, hospitality, data centre); residential real estate including for-sale housing, homebuilding and housing affordability; data-centre real estate as land, buildings, leases and capital structure; proptech and construction technology; construction methods, contracting and construction labour; building materials and products; architecture and engineering practice; REITs and real-estate capital markets including CMBS and CRE debt.

Explicitly out of scope, and who owns it.

• Electricity generation, transmission, interconnection queues and power procurement belong to sector 05 (energy and power). Sector 05 explicitly claims data-centre interconnection and power procurement and explicitly assigns data-centre construction, cooling and real estate to digital infrastructure. This dossier takes the land, buildings, leases, siting and capital structure; it takes power only where power has become a real-estate attribute — which, in 2026, it has.
• Mortgage origination, securitisation mechanics and REIT equity as a financial product belong to sector 07 (finance and fintech). This dossier treats mortgage rates as an input price and CMBS delinquency as a property-market indicator, not as credit products in themselves.
• Industrial and manufacturing capex as a demand story belongs to sector 09. Sector 09 already established that data-centre construction starts nearly tripled to $84.1bn year to date and is crowding out factory construction for the same crews and switchgear. This dossier does not re-derive that; it adds the real-estate and siting view.
• Semiconductors, servers and the AI workloads inside the buildings belong to sectors 01 and 03.
• Climate adaptation finance, carbon and green-building certification economics belong to sector 20.

Boundary disputes worth naming.

1. Data centres are claimed by at least four sectors — power (05), industrial construction (09), semiconductors (03) and real estate (18). Any market size that sums "data-centre market" across these sectors double-counts badly. The defensible split is: 05 owns the electrons, 09 owns the contention for industrial inputs, 03 owns the silicon, 18 owns the dirt, the building, the lease and the debt secured against them.
2. Construction labour is contested between this sector and labour-market analysis generally. We claim it, because in 2026 the decisive facts are crew availability and immigration policy, not wage-setting institutions.
3. Homebuilding sits awkwardly between construction and consumer durables. We claim it, on the grounds that land basis and entitlement — not manufacturing — determine homebuilder economics.
4. The largest definitional trap in this sector is not a boundary dispute but a statistical one: the US Census reports data-centre construction inside the Office category. Census office construction rose 21.3% year over year in July 2026 while office CRE remained distressed. Analysts who read the Census nonresidential series without disaggregating data centres will conclude that office construction is booming. It is not.

2. Subcategories

3. Market structure

Concentration varies more within this sector than in almost any other. It is not one market.

• Contracting is extremely fragmented. US construction employs 8,359,000 people (BLS, August
  2026. across hundreds of thousands of firms, most of them small. No contractor has pricing power from scale. What a few contractors now have is scarcity power in specific trades — electrical work for data centres above all.
• Homebuilding is a consolidating oligopoly at the top. The largest builders have taken persistent share through the cycle because balance sheet and in-house mortgage capability (rate buydowns) are decisive when affordability, not demand, is the constraint.
• Data-centre development is winner-take-most and getting more so. With 80.4% of under-construction capacity preleased and under 1,500 MW available across primary markets, only developers with existing hyperscaler relationships can secure the leases that unlock the financing. Average project cost rose from $352.8m to $1.06bn in a single year, which is itself a barrier — a $1bn average project size excludes most of the development industry.
• Brokerage and research are a duopoly at the top (CBRE and JLL), which matters more than it sounds: they are the only source of data-centre vacancy and capacity data, they disagree with each other by an order of magnitude, and both earn fees on the transactions they measure.
• Materials is a regional oligopoly in heavy, low-value-density products (cement, aggregates) and a tariff-exposed traded market in everything else.

Where the margin actually sits. Not in building. Contractor margins are thin, volatile and currently compressed by input costs rising 8.9% year over year against fixed-price contracts signed in 2025. Margin sits in three places: (1) land and entitlement — the value created by converting unentitled land into a site with power and permission; (2) the lease — the contracted cash-flow stream, which is what the capital markets actually buy; and (3) increasingly, the interconnection queue position, which sector 05 records is being traded as an asset in its own right and which CBRE reports is now being financed against in Dallas-Fort Worth.

Barriers to entry have risen sharply and are now mostly non-financial: interconnection (a median over five years from request to operation, per sector 05), equipment lead times (power transformers at 128 weeks), skilled-crew availability at record-low 3.1% construction unemployment, and — new in 2026 — municipal moratoria and the withdrawal of state tax incentives.

Who has pricing power. Data-centre landlords, unambiguously: asking rates for 10+ MW requirements rose 6.7% in H1 2026 nationally, 14.5% in Atlanta and 19% in New York-New Jersey, with Northern Virginia at $160-185/kW/month. Skilled electrical trades. Suppliers of long-lead electrical equipment. Nobody else — homebuilders are discounting, office landlords outside the trophy tier are not, and contractors are absorbing input inflation.

Market size. US construction put in place ran at $2,157.6bn SAAR in July 2026, down 3.8% year over year, split $859.0bn private residential and $755.2bn private nonresidential (US Census Bureau, released 2026-09-01). This is a measured government statistic, not a modelled market estimate, and is the only sizing figure in this dossier that should be treated as a fact. Data-centre construction starts of $84.1bn year to date through July 2026 across 136 projects is a ConstructConnect database measure (published 2026-08-28) and is an estimate — it counts projects entering a commercial database, not verified groundbreakings.

4. Who matters

Leading companies. Digital Realty (https://www.digitalrealty.com/) · Equinix (https://www.equinix.com/) · QTS Data Centers (https://www.qtsdatacenters.com/) · Blackstone (https://www.blackstone.com/) · Boston Properties (https://www.bxp.com/) · SL Green Realty (https://www.slgreen.com/) · D.R. Horton (https://www.drhorton.com/) · Greystar (https://www.greystar.com/) · Turner Construction (https://www.turnerconstruction.com/) · JE Dunn Construction (https://www.jedunn.com/) · Sundt Construction (https://www.sundt.com/) · Caterpillar (https://www.caterpillar.com/) · Procore Technologies (https://www.procore.com/)

Notable startups. Gravis Robotics (https://gravisrobotics.com/) — $200m led by SoftBank, 2026-08-18 · Bedrock Robotics (https://bedrockrobotics.com/) — autonomous excavators deployed with Sundt and Zachry, 2026-08-19 · Buildots (https://buildots.com/) — $130m, 2026-09-14 · DroneDeploy (https://www.dronedeploy.com/) — being acquired by Procore for $845m, 2026-07-30

Active investors. Blackstone (equity, credit and, through QTS, operator) · SoftBank (led the largest construction-robotics round of the period) · DigitalBridge · Macquarie Asset Management · Thoma Bravo · Welsh, Carson, Anderson & Stowe (owner of Green Street). Note the macro brief's standing caution: with >70% of Q2 2026 global VC going to AI companies, real-estate-specific venture is competing for a shrinking residual.

Platforms and standards bodies. International Code Council (https://www.iccsafe.org/) · ASHRAE (https://www.ashrae.org/) · Uptime Institute (https://uptimeinstitute.com/) · buildingSMART International (https://www.buildingsmart.org/) · Open Compute Project (https://www.opencompute.org/)

Regulators. US Census Bureau Construction Statistics (https://www.census.gov/construction/) · US Bureau of Labor Statistics (https://www.bls.gov/) · Federal Housing Finance Agency (https://www.fhfa.gov/) · Freddie Mac (https://www.freddiemac.com/) · Consumer Financial Protection Bureau (https://www.consumerfinance.gov/) · OSHA (https://www.osha.gov/) · state and municipal zoning and permitting authorities — which in 2026 are the binding regulator for data centres, not any federal body

Research institutions. Harvard Joint Center for Housing Studies (https://www.jchs.harvard.edu/) · MIT Center for Real Estate (https://mitcre.mit.edu/) · NBER housing and urban economics programme (https://www.nber.org/) · Urban Institute Housing Finance Policy Center (https://www.urban.org/policy-centers/housing-finance-policy-center)

Trade organisations. National Association of Realtors (https://www.nar.realtor/) · National Association of Home Builders (https://www.nahb.org/) · Associated General Contractors of America (https://www.agc.org/) · Associated Builders and Contractors (https://www.abc.org/) · Urban Land Institute (https://uli.org/) · NAREIT (https://www.reit.com/) · National Multifamily Housing Council (https://www.nmhc.org/) · American Institute of Architects (https://www.aia.org/) · Data Center Coalition (https://www.datacentercoalition.org/)

Consumer and civil-society groups. National Low Income Housing Coalition (https://nlihc.org/) · Enterprise Community Partners (https://www.enterprisecommunity.org/) · Local Initiatives Support Corporation (https://www.lisc.org/) · and — newly consequential in 2026 — the ad hoc municipal opposition groups driving data-centre moratoria in Oregon, Arizona, New Jersey and Virginia, which are not organised nationally and are therefore invisible to every commercial data provider in this sector.

5. Products, business models, technologies, customers

Major products. Leased space, priced per square foot in conventional property and per kilowatt-month in data centres — a unit change that tells you the asset is being sold as electrical capacity, not floor area. For-sale housing. General contracting and construction management services. Architectural and engineering design fees. Building materials. Mortgage and CRE debt. Increasingly: powered land, and the interconnection queue position attached to it.

How money is actually made today, and how that is changing.

1. Conventional landlording — rent minus operating cost, levered. This model is in trouble wherever the leverage was struck in 2021-22: office CMBS delinquency 11.91%, multifamily 7.69% and rising fastest of any type.
2. Development — create value through entitlement and construction, then sell or refinance. The exit leg has partly broken. Data-centre investment sales were $1.7bn in H1 2026 against roughly $27.8bn of high-yield issuance. Developers are refinancing rather than selling, because there is no equity bid to sell into.
3. The lease as a securitisable credit instrument. This is the genuine business-model change. A hyperscale lease with a minimum-take power covenant is closer to a take-or-pay energy contract than to a property lease, and it is underwritten on tenant credit. Data-centre securitisation reached $4.9bn of SASB CMBS in H1 (CBRE) or $17bn including ABS (JLL). The asset class is migrating from real estate toward corporate credit, and its valuation methods have not caught up.
4. Homebuilding — buy land, build, sell, and increasingly finance the buyer. With mortgage rates at 6.76%, the in-house mortgage subsidiary is where the competitive advantage sits: builders buy down the rate and recover it in price. This is why 63% of builders use incentives while only 35% cut list price.
5. Construction contracting — fixed-price and cost-plus. Contract form determines who eats 8.9% input inflation. Cost-plus and GMP pass it to owners; lump-sum contractors absorb it.

Technologies that matter. High-density liquid cooling and the building design it forces; prefabricated electrical rooms and skid-mounted plant (the main lever on data-centre schedule); autonomous heavy equipment (Gravis, Bedrock, Caterpillar/FieldAI); AI progress tracking from site imagery (Buildots); reality capture and digital twins (Procore/DroneDeploy); behind-the-meter generation as a siting technology; and — conspicuously — not modular construction or 3D printing, both of which absorbed a decade of capital and attention without changing the sector's cost curve.

Major customer segments and what they buy on.

• Hyperscalers buy on time-to-power and contiguous capacity, not price. This is why asking rates can rise 19% in a year without demand destruction.
• Enterprise colocation tenants buy on price and network adjacency, and are being crowded out — rate increases were steepest in the largest capacity tiers and mildest at 250-500 kW (+4.3%).
• Office tenants buy on quality and amenity where they are expanding at all; leasing above $100 per square foot hit record volumes while the commodity tier stagnated.
• Homebuyers buy on monthly payment, not price — which is why a 41bp rise in the 30-year rate did more damage than a 1.6% rise in median price.
• Renters buy on effective rent net of concessions, a figure the published advertised-rent indices do not report.

6. Geography

Where demand concentrates. Data-centre demand is concentrated and shifting simultaneously. Northern Virginia remains the largest single inventory at 4,496.5 MW with 0.24% vacancy, followed by Atlanta (1,803.2 MW), Dallas-Fort Worth (1,432.8), Phoenix (1,069.2) and Chicago (910.6). But growth has moved: Atlanta's under-construction capacity grew 52.3% year over year against Northern Virginia's 16.5%, and JLL reports 77% of North American capacity under construction now sits in frontier markets. ConstructConnect's state-level starts corroborate: Indiana, Illinois, Michigan, North Carolina ($10.4bn) and Texas ($9.9bn) each at or near $10bn year to date, with the South taking 71% of the $78.7bn near-term pipeline.

Where capital concentrates. Overwhelmingly the US, and within it the debt markets rather than equity. The macro brief records Americas at 66% of Q2 2026 global VC with Europe structurally weak at $25.6bn — which is why it is notable that the largest construction-robotics round of the period (Gravis Robotics, $200m, SoftBank-led) went to a Swiss company.

Where regulation binds. At the municipal and state level, not the federal. This is a genuine structural feature of the sector: there is no national land-use regulator, so the binding constraint on the largest infrastructure buildout in a generation is being set by county commissions and state legislatures acting independently — New York's 50 MW permitting pause, Hillsboro's land-use moratorium, Oregon's HB 4084, Arizona's three-year tax-exemption suspension.

Non-US market, with a regional source. The United Kingdom provides the cleanest test of whether the US housing freeze is a rate phenomenon or a US-specific one. UK average house prices were £272,000 in June 2026, up 2.0% annually — England £293,000 (+1.8%), Scotland £195,000 (+2.3%), Northern Ireland £202,000 (+9.2% in Q2). Average private rents were £1,393 in July 2026, up 3.7%, with London at £2,317 (+3.0%) (UK Office for National Statistics, released 2026-08-19). The ONS noted price growth was weaker this summer than last, following April 2025 tax changes.

The comparison is instructive. The UK has similar nominal price appreciation to the US (2.0% versus 1.6% on the US median) but faster rent inflation (3.7% versus a US market where JCHS reports rents have declined), and Northern Ireland at +9.2% has no US analogue. The UK also has no equivalent of the US mortgage lock-in effect, because UK mortgages are predominantly short-fixed and refinanced regularly. The conclusion: the price behaviour is broadly similar across both markets, but the volume collapse is substantially a US institutional artefact of the 30-year fixed mortgage. Anyone generalising "the global housing freeze" from US transaction data is generalising a feature of US mortgage design.

A structural note on data availability by geography. Data-centre capacity data outside North America is materially worse than inside it, and both CBRE and JLL publish companion EMEA and APAC reports with less granularity. There is no public equivalent of the US Census construction-spending series for most of the world.

7. Historical trend patterns

This sector has a long and specific record of well-funded false positives. Naming them precisely is the most useful thing this section can do.

Cycles that actually recur (10-25 years).

• The credit cycle drives this sector more than the demand cycle does. 2007-09 and 2022-26 are both stories about the cost and availability of debt, not about whether people wanted space. The tell in both cases was the same: transaction volume collapsed before appraised values moved.
• Construction employment is chronically procyclical and structurally short. The sector shed roughly 2 million jobs after 2007 and never rebuilt the training pipeline, which is why a soft market in 2026 can coexist with record-low 3.1% construction unemployment.
• Supply waves in multifamily run on a 3-4 year lag because permit-to-completion averages 18.9 months and land assembly precedes that. Every multifamily cycle over-delivers into weakness, because the decision to build was made when the market was strong.
• Appraised values lag transacted values by four to eight quarters, in both directions. This is not an occasional failure; it is the structural behaviour of an appraisal-based index.

Prior hype waves in THIS sector and how they resolved.

1. WeWork and space-as-a-service (2015-2021). Peaked at a $47bn private valuation, failed to IPO in 2019, listed via SPAC, went bankrupt. The thesis — that flexible space would re-intermediate the office lease — was not wrong about demand; it was wrong about who would capture it. Landlords now run flexible space themselves.
2. iBuying (2017-2022). Zillow Offers closed in 2021 after losses; Opendoor persists at a fraction of its peak valuation. The lesson: algorithmic pricing of heterogeneous illiquid assets works until it needs to work in a falling market.
3. Modular and offsite construction (2016-2023). Katerra raised roughly $2bn including SoftBank money and liquidated in 2021. Britain's Ilke Homes and L&G Modular both failed. The pattern is consistent and the reason is consistent: factory construction converts a variable-cost business into a fixed-cost one, which is lethal in a cyclical demand environment. This is the single most important precedent for assessing the current construction-robotics wave — and the relevant difference is that autonomy retrofits equipment a contractor already owns rather than requiring a factory.
4. 3D-printed housing (2018-2024). Sustained press coverage, near-zero units delivered at scale. A pure attention trend that never produced measurable adoption.
5. "The office apocalypse" (2020-2023). Predictions of 50%+ permanent office value destruction. Partly right on commodity stock, badly wrong on trophy — leasing above $100 per square foot is now at record volumes. The reversal of this narrative into "office has bottomed" is itself the current hype wave (see §10).
6. Bitcoin-mining data centres (2018-2022). Sector 05 records this as the nearest precedent for a large-load boom that arrived, interconnected and partly departed, leaving stranded interconnection in Texas and upstate New York. It is the precedent utilities cite privately about AI data centres, and it is the one nobody in real estate cites publicly.

The through-line. Every one of these failures shared a structure: capital arrived before the unit economics were proven, on a thesis about the sector rather than a contract with a customer. The current data-centre buildout is the opposite structure — 80.4% preleased before construction, underwritten on signed hyperscaler leases. That is a genuinely better risk profile than any prior hype wave in this sector, and it is the strongest argument that this one is different. The counter-argument is that it concentrates all the risk in a handful of correlated tenant credits.

8. What is changing now (as of 2026-09-15)

The sector has split into two economies that share a labour force.

Economy one: data centres. Primary-market vacancy of 1.4%, 80.4% of under-construction capacity preleased, under 1,500 MW available, asking rates up 6.7-19% depending on market and tier, and construction starts of $84.1bn year to date against average project costs that rose from $352.8m to $1.06bn in a year with cost per square foot up 57% to $818.20.

Economy two: everything else. Total construction spending down 3.8% year over year, private residential down 7.3%, manufacturing down 21.7%, existing-home sales at a three-decade low of 3.98 million SAAR, new-home supply at 9.6 months, office CMBS delinquency at 11.91% and multifamily delinquency rising faster than any other property type.

The link between them is labour and materials, and it is now documented rather than inferred. Construction unemployment hit a record low 3.1% in August 2026 despite the soft market, with 325,000 open positions. AGC states explicitly that shortages remain acute because data centres are straining labour supply, and that fewer than half of US metro areas added construction jobs in the year to July 2026. Construction Dive reports that data centres drove all nonresidential construction spending growth in July, with non-data-centre spending at its lowest since September 2023. This is the real-estate-side confirmation of sector 09's crowding-out finding.

Grounding in the macro brief. Three macro facts bind this sector particularly hard:

1. This is a rising-rate-risk environment, not an easing cycle. The FOMC held at 3.50-3.75% on a 9-3 vote with three dissents in favour of a hike. The 30-year mortgage at 6.76% is up 41bp year over year. Every CRE loan extended during 2023-25 on the assumption of cheaper refinancing in 2026 was extended on a false premise. The 16 September 2026 FOMC decision falls one day after this research date and is unresolved.
2. Tariff policy was rebuilt on sectoral authorities after the Supreme Court voided the IEEPA tariffs. Construction input costs are up 8.9% year over year against core CPI of 2.4% — roughly a threefold divergence. Unlike the IEEPA duties, these are not being refunded.
3. Capital concentration in AI has starved the sector's own technology layer. Proptech funding was $4.53bn in H1 2026 against $16.7bn tracked for full-year 2025, during the largest half-year of global venture funding on record.

The information problem is the story. Three separate 2026 measures of commercial property value point in different directions and none is a clean transacted price. See §10 and §13.

9. The five lists

Five most important current trends

1. Data-centre real estate as the only structurally tight major property type (T-18-01)
2. Record-low construction unemployment against a soft construction market (T-18-08)
3. Construction input costs inflating at roughly three times consumer inflation (T-18-07)
4. The US housing transaction freeze entering its fifth year (T-18-05)
5. The office loan maturity wall extended rather than resolved (T-18-03)

Five fastest-growing signals

1. Municipal moratoria and local opposition as a binding data-centre siting constraint (T-18-09)
2. Data centres financed with debt and securitisation while almost nothing trades as equity (T-18-10)
3. Frontier data-centre markets displacing established clusters (T-18-14)
4. Construction robotics and autonomy capitalising while proptech starves (T-18-13)
5. Interconnection queue position emerging as a financeable real-estate asset (T-18-11)

Five trends most likely to affect businesses

1. Record-low construction unemployment (T-18-08) — overlaps list 1; this is the constraint that binds every capital project in the sector regardless of asset class
2. Construction input cost inflation and tariff pass-through (T-18-07) — overlaps list 1
3. The office loan maturity wall (T-18-03) — overlaps list 1; the credit consequence lands on regional banks and CRE lenders, not only on landlords
4. Manufacturing and reshoring construction collapsing as data centres take the crews (T-18-18)
5. Preleasing replacing speculative development in data centres (T-18-02)

Five trends most likely to affect consumers

1. The US housing transaction freeze (T-18-05) — overlaps list 1
2. The immigration collapse as a two-sided shock to housing demand and construction labour (T-18-12)
3. New-home spec inventory overhang and builder price cuts (T-18-06) — the one trend currently working in buyers' favour: 35% of builders cutting prices by an average 6%, 63% using incentives
4. The multifamily supply wave passing while its debt fails (T-18-17) — consumers experience the supply wave as the only rent relief of the cycle, and its passing as its end
5. Municipal moratoria on data centres (T-18-09) — consumers experience this as a local land-use and electricity-price fight, which is how it will be decided

Overlaps, stated plainly. T-18-05, T-18-07, T-18-08 and T-18-03 appear on multiple lists. This is not padding: the labour and materials constraints are genuinely simultaneous business and consumer trends, because they determine both contractor margin and the price of a house.

10. Overhyped / overlooked / cooling / reversing

Most overhyped

1. "Office has bottomed" as an investment conclusion (T-18-19). The recovery data is real: record post-pandemic leasing, up 27% over the trailing twelve months; vacancy down 60bp quarter over quarter; 30 million square feet of occupancy gains; Green Street's price index up 5.0% year over year in August 2026 after +2.9% a year earlier. The hype is the leap from those facts to office credit having bottomed. The specific evidence that hype outruns substance: office CMBS delinquency was 11.91% in July 2026, barely below its January record of 12.34%; roughly $25bn sat past maturity unresolved; and per ULI's July 2026 assessment, the all-property cap rate rose just 85bp against a roughly 300bp rise in the 10-year Treasury, leaving the yield spread near zero against a century average near 230bp, with cap-rate spreads over BBB corporates at -66bp against a 2020-21 average of +160bp. Property is priced tighter to credit than at any point in the recent cycle, at the same moment as the credit underlying it is at record distress. The honest caveat: office delinquency did peak in January and has drifted down, and if the Fed cuts, refinance capacity genuinely improves. This is a probabilistic judgement, not a certainty.

2. Announced data-centre pipeline capacity treated as a demand measure (T-18-20). JLL reports more than 66 GW under construction in North America — noting it exceeds Germany's entire electricity requirement. CBRE, counting physical construction in primary markets, reports 7,481 MW. The gap is roughly ninefold. Neither firm is wrong; they count different things, and neither publishes its inclusion rule. The specific evidence that the larger figure overstates deliverable capacity comes from sector 05: ERCOT suspended its Batch Zero process against 474 GW of requests, roughly 90% data centres and more than five times its historical peak, saying its own forecast was probably too high; Exelon disclosed it expects 22% of its 65 GW pipeline to materialise. A 22% materialisation rate applied to a 66 GW pipeline describes a very different real-estate market from the headline. Meanwhile $1.7bn of data centres actually traded in H1 2026 — there is almost no arm's-length evidence validating any of it.

Most overlooked

1. The valuation information problem is the sector's defining condition, and nobody is pricing it. Three separate 2026 measures of commercial property value point different ways and none is a clean transacted price:

• Appraised (NCREIF): the capital-value component fell almost 20% between Q2 2022 and Q4 2024 without a recession — a historic anomaly, and one that arrived after transaction evidence had already moved.
• REIT-implied (Green Street CPPI): +5.0% year over year in August 2026, accelerating from +2.9% a year earlier. Derived from estimates of REIT portfolio values, not from closed sales.
• Transacted: functionally unobservable in the asset class that matters most. $1.7bn of data-centre investment sales in H1 2026 against ~$27.8bn of high-yield issuance and a single $16bn construction facility.

The gap between these is not a measurement nuisance to be resolved by picking one. It is the sector's central risk, and it is largest in exactly the asset class attracting the most capital. Lenders, appraisers and index providers are all marking data centres against almost no sale evidence, at the same time the FOMC has flagged increased leveraged financing of infrastructure buildout as a financial-stability concern. Attention has missed this because each index is published by a different firm to a different client base, and nobody is paid to reconcile them.

2. The immigration collapse is a two-sided shock and is being modelled one-sided (T-18-12). Harvard JCHS estimates net international immigration fell 54% in 2025 (2.7m to 1.3m) and projects a further 75% decline in 2026 to about 0.3m, cutting household formation from recent immigrants from 640,000 in 2024 to 143,000 in 2027 — nearly 500,000 fewer households formed by 2027. The same population decline removes construction labour, at a moment when construction unemployment is at a record-low 3.1%. A model capturing only the demand loss forecasts falling prices; one capturing only the labour loss forecasts rising costs. Both are happening, and almost nobody is running both.

3. Multifamily distress is accelerating faster than office and getting a fraction of the attention (T-18-17). Multifamily CMBS delinquency rose 46bp in July 2026 to 7.69% — the largest monthly increase of any major property type — while office fell. Multifamily is a far larger share of outstanding CRE debt than office. The attention asymmetry is itself the signal.

4. Local land-use politics is now the binding constraint on the AI buildout, and no commercial data provider tracks it (T-18-09). Every serious forecast models power and equipment lead times. A county commission vote operates on a different and much faster timetable, and the 2026 actions — New York's 50 MW permitting pause, Hillsboro's moratorium, Oregon HB 4084, Arizona's three-year tax-exemption suspension — surfaced through a single brokerage's market commentary. CBRE now rates local opposition as an obstacle on par with power procurement.

Trends that appear to be cooling

Proptech venture funding (T-18-16). The indicator that turned: $4.53bn across 231 rounds in H1 2026 against $16.7bn tracked for full-year 2025 — an annualised pace roughly 46% below the prior year, during the largest half-year of global venture funding on record. This is not a correction tracking the market; it is a decoupling from it.

Manufacturing and reshoring construction (T-18-18). The indicator: private manufacturing construction at $167.8bn SAAR, -21.7% year over year, while data centres drove all nonresidential growth and non-data-centre spending fell to its lowest since September 2023.

The multifamily supply wave (T-18-17). The indicator: six consecutive months of advertised rent increases on slowing supply, Los Angeles reaching peak supply in 2026, Texas permitting cooling.

Trends that may reverse, and the mechanism

• Data-centre tightness would reverse through a hyperscaler capex retrenchment. Mechanism: capex guidance cuts → lease cancellations and backlog erosion → the 80.4% preleasing figure falls before vacancy rises. Early indicator: preleasing share, not vacancy. Vacancy is a lagging indicator in a preleased market and will be the last thing to move.
• The housing freeze would reverse through rate relief or through capitulation on price. Early indicator: months of supply is already at 4.9 versus 4.0 pre-pandemic, and builders are cutting. Watch whether existing-home sellers follow builders into discounting — that is the volume unlock, and it does not require the Fed.
• Construction labour scarcity would reverse through an immigration policy change or through a residential downturn deep enough to release crews. It has not released them yet. Early indicator: construction job openings falling below 250,000 while employment is flat.
• Office credit distress would reverse through a genuine cutting cycle. Early indicator: delinquency falling below 10% and the inclusive measure converging toward the headline — the 176bp wedge closing is the real signal, not the headline rate falling.
• Data-centre local opposition would reverse if electricity price impacts on residential ratepayers are contained. Early indicator: state utility commission decisions on large-load cost allocation, which sector 05 notes no regulator has yet published a full incidence study on.

11. Risks and major uncertainties

Sector-specific risks.

1. Correlated tenant credit in data centres. A small number of hyperscalers underwrite most of the 80.4% preleasing and most of the securitised debt. This is not diversified real estate; it is concentrated corporate credit with a building attached.
2. Valuation without transaction evidence. Marking an asset class with $1.7bn of half-year trades against tens of billions of debt is a model-risk problem, and the FOMC has already flagged leveraged infrastructure financing as a stability concern.
3. Extension without resolution in CRE debt. The 176bp wedge between reported (7.86%) and inclusive (9.62%) delinquency is deferred loss, and it does not self-cure in a rising-rate environment.
4. Labour as an absolute ceiling. At 3.1% construction unemployment, the sector cannot build more in aggregate regardless of capital. Additional data-centre construction is substitution, not addition.
5. Tariff exposure rebuilt on durable legal authority. Unlike the refunded IEEPA duties, the sectoral tariffs driving 8.9% input inflation are not being returned.
6. Municipal veto risk on the largest projects. No federal preemption exists for data-centre siting.
7. Regional bank concentration in CRE. CMBS is only 15-20% of US CRE mortgage debt; the larger bank-held exposure is far less transparent.

Genuine unknowns — and the distinction matters.

Things we don't know but could find out with better access:

• The immigrant share of the construction workforce and the direct effect of enforcement actions on crew availability. This is measurable from CPS microdata; it is simply not published in the form the sector needs, and it is the single most important unquantified input in this dossier.
• Bank-held office loan modification rates. Regulators have this; they do not publish it at loan level.
• Actual data-centre lease terms, including how widespread minimum-utilisation covenants are. These are private contracts; only fragments surface in brokerage commentary.
• The reconciliation between CBRE's and JLL's capacity counts. Both firms have their methodologies; neither publishes them.
• Effective rents net of concessions in multifamily. The data exists inside RealPage and Yardi.

Things nobody can know:

• Whether AI compute demand in 2028 will justify capacity contracted in 2026. This is a two-year forward bet on a technology whose economics are unsettled, and no amount of research resolves it.
• Whether the September 2026 FOMC decision and the rate path beyond it will relieve or intensify the refinancing problem. Unresolved at this research date by construction.
• Whether municipal opposition to data centres is a durable political movement or a transient reaction that fades once specific projects are approved or rejected.

12. Scenarios to 2030

Base — two-speed sector persists (most likely). Data-centre construction continues at a high but decelerating rate as power and permitting bind; preleasing stays above 70%; housing volume recovers slowly toward 4.5-5 million existing-home sales as sellers capitulate on price rather than waiting for rates; office credit resolves through slow attrition with delinquency grinding down from 12% without a clearing event; construction labour stays tight and input inflation moderates to 4-6%. Falsifiable early indicator: data-centre preleasing share holding between 70% and 85% in successive CBRE semiannual reports while total US construction spending stops declining year over year.

Upside — the freeze breaks without a bust. A rate-cutting cycle from late 2026 restores refinance capacity; existing-home sales recover past 5 million; the office maturity wall resolves through refinancing rather than foreclosure as the 176bp inclusive/headline wedge closes; data-centre demand proves durable and frontier-market supply relieves pricing pressure without collapsing it; immigration policy partially reverses, easing crew scarcity. Falsifiable early indicator: the inclusive CMBS delinquency measure converging toward the headline measure — that is the specific signal that extensions are curing rather than deferring.

Downside — AI capex retrenchment meets the CRE debt overhang. Hyperscaler capex guidance cuts trigger lease cancellations; the 80.4% preleasing figure falls first, then backlog, then vacancy; data-centre securitisations reprice with no equity bid to clear into; the labour that was crowded out does not return to housing fast enough to matter; regional banks with CRE concentration take losses simultaneously. Falsifiable early indicator: preleasing share falling below 70%, or data-centre ABS/CMBS new issuance stalling for two consecutive quarters. Watch preleasing, not vacancy — vacancy moves last.

Disruption — the lease stops being a real-estate instrument. Minimum-utilisation covenants, behind-the-meter generation and financeable queue positions generalise to the point where data-centre assets are underwritten, rated and traded as infrastructure credit rather than property. REIT structures become the wrong wrapper; infrastructure funds and private credit displace real-estate capital; appraisal-based indices stop describing the asset class entirely. Falsifiable early indicator: a major data-centre platform financing or listing structured explicitly as infrastructure rather than as a REIT, or rating agencies publishing a distinct data-centre methodology that abandons property comparables.

Regulatory — land use becomes the binding national constraint. Municipal and state moratoria spread from the four to six jurisdictions identified in 2026 to a majority of high-growth markets; tax-incentive withdrawal becomes the norm rather than the exception; federal preemption is attempted and contested. Capacity growth slows sharply for reasons unrelated to demand or power. Falsifiable early indicator: count of US states with active data-centre permitting restrictions or incentive suspensions exceeding ten, or a federal preemption bill receiving a committee vote.

Failure — the information problem produces a credit event. The gap between appraised, REIT-implied and (absent) transacted values persists until a forced sale establishes a genuine clearing price far below carrying values; the repricing cascades through data-centre ABS, the CMBS stack and bank CRE books simultaneously; the FOMC's leveraged-infrastructure-financing concern is validated. Falsifiable early indicator: the first large, genuinely arm's-length data-centre portfolio sale — whatever price it clears at will be the most informative single datapoint this sector produces between now and 2030. There is currently no scheduled event that would generate one, which is precisely the problem.

13. Data gaps and limitations

Research-capacity limitation (material, and disclosed per the standing instruction). The shared 200-call WebSearch budget was exhausted after six searches in this sector. All subsequent research used direct WebFetch on primary sources. Consequences:

• Strong coverage: US government statistics (Census C30, New Residential Construction, New Residential Sales, BLS employment and PPI), Freddie Mac PMMS, NAR, NAHB, UK ONS, Harvard JCHS, and brokerage/trade research reachable by direct URL.
• Weak coverage: company-level financials. Digital Realty, Equinix and every homebuilder's quarterly 2026 results could not be retrieved. IR pages returned 404 or robots-disallowed responses; SEC EDGAR was unreachable through both WebFetch (robots) and curl (proxy CONNECT rejected, 403). Every entity record affected is marked unverified rather than estimated.
• Not attempted: non-English-language sources beyond the UK; APAC and EMEA data-centre markets; Chinese and Indian construction and housing data.

Specific unresolved conflicts.

1. Data-centre capacity under construction: CBRE 7,481 MW versus JLL 66+ GW — roughly ninefold, definitional, and unreconcilable from public information because neither firm publishes its inclusion rule. Recorded as a contradiction on T-18-01 and T-18-20.
2. Data-centre securitisation: CBRE $4.9bn SASB CMBS versus JLL $17bn SASB/CMBS and ABS — a ~$12bn scope gap larger than the entire equity transaction market for the asset class.
3. Data-centre vacancy: CBRE 1.4% versus JLL 1.0%. Immaterial to the conclusion, but it demonstrates there is no authoritative vacancy figure for this asset class.
4. Housing starts versus permits, July 2026: starts -13.5% year over year while permits +3.1% — same agency, same release, opposite signals, unresolved.
5. Multifamily: rents rising six straight months versus delinquency up 46bp to a multi-year high. Both true; they measure different layers of the same asset.
6. Inflation indices: construction inputs +8.9% year over year against core CPI 2.4% and core PCE ~3.3%. The macro brief addendum already flags the CPI/PCE gap as unresolved; construction inputs are a third series diverging from both.
7. Office delinquency: 12.34% (January 2026 record) versus 11.91% (July 2026) — same provider, different months, routinely cited as though it were a source conflict.

Figures I could not verify at all.

• FHFA House Price Index 2026Q2 values (page reachable, data not extractable).
• S&P CoreLogic Case-Shiller 2026 readings.
• AIA/Deltek Architecture Billings Index 2026 readings — four separate URL attempts failed. Architecture practice is consequently the thinnest-covered subindustry in this dossier, and the ABI is a leading indicator for construction spending that Phase 2 should prioritise.
• National office vacancy level (JLL gives the quarterly change, not the absolute rate).
• Office-to-residential conversion volumes. Frequently discussed as a solution to both office vacancy and housing shortage; no verifiable 2026 figures were obtained, so no trend record was created for it. This is a deliberate omission, not an oversight.
• US multifamily absorption and completions in units, and effective rents net of concessions. Yardi and RealPage gate the numbers behind registration.
• The immigrant share of the construction workforce, and any direct measurement of immigration enforcement effects on crew availability. The demand-side effect is well modelled by JCHS; the labour-supply side of T-18-12 is inferred from record-low construction unemployment and is not evidenced. This is the most important single gap in the dossier.
• Data-centre REIT operating metrics for 2026 (see above).
• Lead investor and valuation for the Buildots $130m round; total funding for Bedrock Robotics.

Known weaknesses in what I did use.

• CBRE and JLL are brokerages reporting on markets they earn fees in. They are also the only source.
• NAR, NAHB, AGC and ABC are trade bodies with explicit lobbying positions; their data is generally sound and their commentary is advocacy.
• CRETI's proptech totals use a self-defined, shifting category boundary and are single-source. Treated as an estimate throughout; T-18-16 is scored accordingly and is the lowest-confidence cooling trend.
• Census monthly changes frequently fall within their own confidence intervals. Where a change is not statistically significant (new-home sales -10.5% ±14.0%; multifamily starts -7.1% ±31.8%) the dossier and trend records say so explicitly.
• Green Street's CPPI is REIT-derived, levered to public-market sentiment, and can overshoot in both directions. It is not transaction evidence and is never described as such here.

14. Ranking scorecard

15. Sources

1. "North American Data Center Demand Continues to Outpace Supply Despite Record Construction Activity" / North America Data Center Trends H1 2026, CBRE, https://www.cbre.com/press-releases/north-american-data-center-demand-continues-to-outpace-supply-despite-record-construction-activity, 2026-08-27, Tier B
2. "North America Data Center Report Midyear 2026", JLL, https://www.jll.com/en-us/insights/market-dynamics/north-america-data-centers, 2026-08-11, Tier B
3. "September 2026 Data Center Report: Year-to-Date Spending Nearly Three Times a Year Ago", ConstructConnect (Michael Guckes, Devin Bell), https://news.constructconnect.com/september-2026-data-center-report-year-to-date-spending-nearly-three-times-a-year-ago, 2026-08-28, Tier B
4. "Construction Spending, July 2026 (Value of Construction Put in Place)", US Census Bureau, https://www.census.gov/construction/c30/pdf/release.pdf, 2026-09-01, Tier A
5. "CBRE: Record Data Center Construction Fails to Ease Capacity Crunch", Data Center Frontier, https://www.datacenterfrontier.com/site-selection/news/55401462/cbre-record-data-center-construction-fails-to-ease-capacity-crunch, 2026-08-28, Tier B
6. "Trepp: CMBS Delinquency Rate Up 51 Basis Points in July", Trepp via MBA NewsLink, https://newslink.mba.org/mba-newslinks/2026/august/mba-newslink-thursday-august-6-2026/trepp-cmbs-delinquency-rate-up-51-basis-points-in-july/, 2026-08-06, Tier B
7. "CMBS Delinquencies Hit Record With $25B Past Maturity", The Real Deal (Trepp data), https://therealdeal.com/national/2026/02/17/cmbs-delinquencies-hit-record-with-25b-past-maturity/, 2026-02-17, Tier B
8. "United States Office Market Dynamics, Q2 2026", JLL, https://www.jll.com/en-us/insights/market-dynamics/us-office, 2026-07-27, Tier B
9. "Commercial Property Price Index, August 2026", Green Street, https://www.greenstreet.com/resources/pricing-index/, 2026-09-04, Tier B
10. "Economist Snapshot: Navigating Commercial Real Estate Valuation Blind Spots", Urban Land Magazine / ULI (Beth Mattson-Teig), https://urbanland.uli.org/capital-markets-and-finance/economist-snapshot-navigating-commercial-real-estate-valuation-blind-spots, 2026-07-20, Tier B
11. "Primary Mortgage Market Survey, week ending 2026-09-10", Freddie Mac, https://www.freddiemac.com/pmms, 2026-09-10, Tier A
12. "NAR Existing-Home Sales Report Shows 2.0% Decrease in August", National Association of Realtors, https://www.nar.realtor/newsroom/nar-existing-home-sales-report-shows-2-0-decrease-in-august, 2026-09-10, Tier A
13. "New Residential Construction, July 2026", US Census Bureau / HUD, https://www.census.gov/construction/nrc/pdf/newresconst.pdf, 2026-08-18, Tier A
14. "New Residential Sales, July 2026", US Census Bureau / HUD, https://www.census.gov/construction/nrs/pdf/newressales.pdf, 2026-08-25, Tier A
15. "NAHB/Wells Fargo Housing Market Index, August 2026", National Association of Home Builders, https://www.nahb.org/news-and-economics/housing-economics/indices/housing-market-index, 2026-08-18, Tier A
16. AGC news releases, September 2026 (construction employment, workforce shortages, input costs), Associated General Contractors of America, https://news.agc.org/, 2026-09-10, Tier B
17. "Employment Situation Summary, August 2026", US Bureau of Labor Statistics, https://www.bls.gov/news.release/empsit.nr0.htm, 2026-09-04, Tier A
18. "Industries at a Glance: Construction (NAICS 23)", US Bureau of Labor Statistics, https://www.bls.gov/iag/tgs/iag23.htm, 2026-09-04, Tier A
19. "Household Growth to Increasingly Feel the Impacts of the Immigration Drop", Harvard Joint Center for Housing Studies (Daniel McCue), https://www.jchs.harvard.edu/blog/household-growth-increasingly-feel-impacts-immigration-drop, 2026-08-11, Tier A
20. "Private rent and house prices, UK: August 2026", UK Office for National Statistics, https://www.ons.gov.uk/economy/inflationandpriceindices/bulletins/privaterentandhousepricesuk/latest, 2026-08-19, Tier A (regional source)
21. Proptech venture funding tracker, H1 2026, Center for Real Estate Technology and Innovation, https://www.creti.org/, 2026-09-15, Tier B
22. Construction Dive technology and workforce coverage, July-September 2026, Industry Dive, https://www.constructiondive.com/topic/technology/, 2026-09-14, Tier B
23. ABC Construction Backlog Indicator and construction economics releases, 2026, Associated Builders and Contractors, https://www.abc.org/News-Media/News-Releases, 2026-07-14, Tier B
24. "Eye on Housing", September 2026 posts, National Association of Home Builders, https://eyeonhousing.org/, 2026-09-14, Tier A
25. "The State of the Nation's Housing 2026", Harvard Joint Center for Housing Studies, https://www.jchs.harvard.edu/state-nations-housing-2026, 2026-06-01, Tier A
26. "Matrix Multifamily National Report, August 2026", Yardi Matrix, https://www.yardimatrix.com/multifamily-national/matrix-multifamily-national-report-august-2026/, 2026-09-04, Tier B
27. "Annual Rent Growth Strengthens as Apartment Market Recovery Continues", RealPage Analytics, https://www.realpage.com/analytics/, 2026-09-03, Tier B
28. "NAR: Existing-Home Sales Decreased to 3.98 million SAAR in August", Calculated Risk (Bill McBride), https://calculatedrisk.substack.com/p/nar-existing-home-sales-decreased-ee1, 2026-09-10, Tier B
29. Equinix press releases (atNorth acquisition completion, Fabric One, AI inference), Equinix, https://newsroom.equinix.com/press-releases-global, 2026-09-02, Tier A
30. Construction Dive labor topic coverage, September 2026, Industry Dive, https://www.constructiondive.com/topic/labor/, 2026-09-04, Tier B

Cross-sector sources relied on but not re-verified here (traceable to their own dossiers): sector 09 (industrial supply chain) for the data-centre construction crowding-out finding and the $84.1bn starts figure; sector 05 (energy and power) for interconnection queue dynamics, the ERCOT Batch Zero suspension against 474 GW of requests, Exelon's 22% materialisation disclosure, and transformer and turbine lead times; and the macro context brief for FOMC policy, tariff architecture and venture-capital concentration.

More research in this sector · Explore trend records