What does the 2026 digital-infrastructure capital wave actually look like, and where does it create real estate opportunity versus narrative noise?

Synthesized the current source set across hyperscaler announcements, Stargate program expansion, platform acquisitions, financed metro and exurban projects, and one watch-list mega-campus announcement. Read against Texas Digital Infrastructure Corridors, Digital Infrastructure Real Estate, and Powered Land and Grid Advantage so this page could stay a national capital map rather than a corridor page.

The useful 2026 point is not that AI infrastructure is large. It is that the money is arriving through different lanes that should not be underwritten as if they were the same thing. This is the first-order AI real estate map: compute, power, data centers, powered land, utility partnerships, and supply-chain spillover. Office demand is a narrower second-order read-through handled in AI Infrastructure and Office Demand 2026 and AI Office Demand Engine 2026.

The Source Collection: CBRE Insights Market Reports Public Crawl 2026 adds a market-wide operating constraint layer underneath the project-announcement map. CBRE's North America Data Center Trends H1 and H2 2025 public pages show the same pattern as the project sources: primary-market inventory kept expanding, but vacancy stayed at record-low levels, net absorption hit record scale, and under-construction capacity remained governed by power procurement and delivery timing rather than ordinary real estate supply.

The underwriting read-through is narrower than "data centers are hot." In CBRE's H2 2025 framing, rent growth, site pricing, and development timing are being driven by limited powered land, booked grid capacity, and the need for power access within roughly 18 to 36 months. That reinforces this page's existing lane discipline: power queue position, utility structure, fiber, zoning / community acceptance, and customer commitment decide whether a data-center or powered-land thesis is underwritable. The row-level data-center observations in the CBRE package remain dry-run or review-only until source locators and value normalization are complete.

Source: JLL 2026 Global Data Center Outlook adds a global sizing and cost layer beneath the project map. JLL forecasts 97 GW of new global data-center capacity between 2026 and 2030, potentially taking global capacity to 200 GW by 2030, with a 14% base-case global supply CAGR. The source also frames the Americas as the fastest-growing region at 17% supply CAGR, with 2030 average 1-20 MW lease-rate forecasts of $222/kW/month in the Americas, $207/kW/month in EMEA, and $179/kW/month in APAC.

That does not turn every powered-land headline into underwritable capacity. JLL's delivery-risk rows are the useful discipline: 57% of projects experienced construction delays of three months or more in 2025, global equipment lead times averaged 33 weeks, and 2026 shell-and-core construction costs are forecast at $11.3M per MW before land and active IT equipment. The right read-through is that AI infrastructure is large enough to be durable, but power, equipment, liquid cooling, construction execution, local approvals, and tenant/capital-stack proof still decide which sites convert into investable real estate.

Source: Cushman & Wakefield Asia Pacific Data Centre Construction Cost Guide 2026 adds an APAC-specific cost-inflation check to that global frame. C&W reports average APAC data-centre construction costs rose 10% year-over-year in 2025 and identifies Japan, Singapore, South Korea, Australia, and Hong Kong as the five most expensive mid-spec build markets. For the capital-deployment map, this is a reminder to separate demand durability from project feasibility: high-cost APAC markets need stronger sponsor basis, contingency, power-cost assumptions, and lending reserves before growth headlines become investable delivery.

Source: Cushman & Wakefield Asia Pacific Data Centre Investment Landscape 2025 adds the investor-screen complement. The page frames APAC data centres as an institutional asset class with yield-stability and inflation-hedging appeal, but it also says the report evaluates 14 key markets through population per megawatt, rent revenue, CapEx requirements, and yield-on-cost. For this capital map, that means APAC capital should be sorted by demand density, revenue capture, capex intensity, and development spread rather than by country growth narrative alone.

Source: Cushman & Wakefield APAC Data Centre Update H2 2025 adds the current APAC delivery scale beneath that investor screen. The public page reports 13,763 MW of operational capacity, a 19,371 MW development pipeline, 3,677 MW under construction, and 15,694 MW in planning as of H2 2025, with Malaysia and India accounting for 58% of new operational capacity. For capital deployment, that shifts APAC from only a growth narrative to an execution market, while still requiring power, regulation, geopolitical, operator, and project-level diligence before planned capacity is capitalized.

Source: Cushman & Wakefield H2 2025 EMEA Data Centre Market Update adds an EMEA delivery-capacity screen. C&W reports EMEA pipeline capacity approaching 15 GW and a 33-market maturity index, but its more important capital-allocation point is that power availability, grid access, regulatory complexity, sustainability requirements, scalable land, and faster grid connection timelines determine where that pipeline can become real capacity. That keeps EMEA capital deployment in the same discipline as North America and APAC: demand is necessary, but executable power and approvals decide investability.

Source: JLL North America Data Center Report Year-end 2025 narrows that global frame to the North American delivery map. JLL reports 1% vacancy for a second consecutive year, more than 35 GW under construction, 92% of under-construction capacity precommitted, and 64% of capacity under construction in frontier markets. The capital-deployment implication is that the opportunity set is moving outward from mature hubs, but only precommitted, power-secured, utility-executable projects deserve underwritable treatment. Texas may overtake Northern Virginia by 2030 if the state is viewed as one market, but that is a power / land / utility-execution thesis, not a generic Texas land thesis.

Source: JLL Power Progress in Your Global Data Center Expansion adds the occupier-portfolio version of the same constraint. The source is useful because it separates AI training near power from inference closer to users, keeps traditional cloud / storage workloads in the demand mix, and ties capacity reservation to multiyear utility lead times. It also shows why power partnerships and tenant-credit economics belong in the capital-deployment map: the same demand headline can price differently depending on energization timing, rack density, counterparty credit, escalation floors, and pass-through exposure.

Source: Avison Young Q1 2026 U.S. Data Center Market Overview adds a current U.S. operating check beneath the capital map. Avison Young reports 29.0 GW of U.S. colocation inventory at Q1 2026, up 22% quarter over quarter and 48% year over year, while net absorption reached 5.29 GW and vacancy declined to 1.2%. That strengthens the capital-deployment read but also narrows it: capital is chasing a market where available capacity is scarce and new supply is largely pre-leased, so the underwritable opportunity is still tied to power delivery, preleasing, construction execution, skilled labor, and local approvals.

Source: CBRE Data Centers - U.S. Real Estate Market Outlook 2026 adds the 2026 CBRE forward-looking layer. The source verifies the previously review-only outlook rows: 300 MW-plus delivery inside 36 months is now a site-selection threshold, 500 MW-plus AI campuses push schedules into multi-year territory, interconnection can stretch to 24, 36, or 48-plus months, and primary-market preleasing is expected to remain in the mid-70% range versus a 40% to 50% historical norm. Capital deployment should therefore favor proved power paths and preleased / customer-backed capacity over generic land-control stories.

Source: Colliers 2026 Data Center Marketplace Report adds the Colliers capital-intensity counterpart. The applied rows (market_observations.id=33900-33914) show the scale of the lane: $580B-plus global data-center investment in 2025, $120B-plus hyperscaler debt issued for AI infrastructure, 47% year-over-year build-cost growth, 15.6 GW of North America absorption, and 90%-plus preleasing of new capacity. The capital-deployment read is selective, not euphoric: private credit and utility deposits are moving earlier in the cycle, but $64B-plus of delayed or canceled U.S. projects since 2023 shows that local process and power deliverability can still kill nominal demand.

There are five distinct capital lanes in the current source set:

1. hyperscaler and AI-program commitments
2. platform M&A and infrastructure-fund consolidation
3. financed metro and exurban delivery
4. supply-chain and manufacturing spillover
5. watch-list announcements that are still too early to equate with committed buildout

Batch 72 adds a sixth adjacent lane: power-system capital. DOE nuclear financing and the reported NextEra / Dominion merger are not real-estate projects, but they matter because they show the capital stack moving upstream into generation, utility scale, and ratepayer politics. The "whale hunting" source also adds a counterweight to mega-campus thinking by showing inference demand can favor smaller, lower-latency sites closer to users.

Batch 75 adds two boundary markers. BXDC shows institutional capital trying to package newly built hyperscaler data centers into a dedicated vehicle, while Google's reported orbital-compute work with SpaceX shows the extreme end of power-constrained compute siting. One is investable real estate capital formation; the other is a frontier watchlist item.

The June 15 policy / mega-campus batch adds a useful counterweight to capex headlines. Local and state data-center bans show that community acceptance can remove capacity from the pipeline, while the reported OpenAI federal-land discussion shows how AI users may seek public-land, power, and federal-site solutions when ordinary sites cannot scale. Keep both as watchlist and underwriting-context evidence until official policy texts, site-control documents, utility records, and executed campus commitments are preserved. See Source: Data Center Local Ban Wave 2026 and Source: OpenAI Federal Land Data Center Campus 2026.

The subagent data-center article batch adds a cross-market capital-deployment trio: Prime's Metro Phoenix groundbreaking belongs in the growth-market campus lane, Newmark's $975M Northern Virginia financing belongs in the mature-core debt lane, and Google's New Florence, Missouri investment belongs in the noncore infrastructure-expansion lane. Together they reinforce that 2026 data-center capital is not one market; it is a set of power, utility, financing, and local-acceptance gates by geography. See Source: Prime Data Centers Metro Phoenix Campus Groundbreaking 2026, Source: Newmark 975M Northern Virginia Data Center Financing 2026, and Source: Google New Florence Missouri Infrastructure Investment 2026.

The later June 15 digital-infrastructure tranche adds three more lanes to the map. DigitalBridge / ArcLight and Helix show capital platforms combining compute, power, connectivity, and generation expertise; Switch's nearly $10B credit-facility package shows balance-sheet liquidity being tied to contracted pipeline, campus growth, and transmission / generation support; and the NDA / public-opposition source shows community acceptance becoming an entitlement variable rather than a soft public-relations issue. Use the batch as capital-formation and approval-risk evidence, not as delivered MW inventory. See Source: DigitalBridge ArcLight Power Infrastructure Acquisition 2026, Source: KKR NVIDIA Kuwait Helix AI Infrastructure Platform 2026, Source: Switch Credit Facilities Nearly 10B 2026, and Source: Data Center NDA Public Opposition 2026.

Broadcom / Apollo / Blackstone adds a sixth lane inside capital formation: compute-platform financing tied to hardware, networking, and frontier-lab capacity commitments. The reported $35B initial tranche and 20+ GW design target should be treated as source-scoped platform claims, but the direction is durable: private credit and insurance capital are moving directly into AI compute infrastructure, not only into real-estate shells. See Source: Broadcom Apollo Blackstone AI Infrastructure Platform 2026.

The underwriting mistake is collapsing all five into one giant capex number. The better move is to ask which lane actually creates a real estate opening that outside capital can access.

The May 2026 podcast synthesis pass adds a softer but useful signal around the same distinction. source-podcast-396-inside-trepp-connect-capital-deployment-the-lending-reset-cre-s-emerging-div-28a5dd47d4a9d56b38bf7bb1|TreppWire 396 pairs lending reset with AI and data-center growth, while source-podcast-392-gigawatts-green-shoots-data-center-101-blue-owl-s-healthcare-grab-nyc-s-pens-45eefeb6cee2903c94ee603d|TreppWire 392 frames data centers as a gigawatt-scale capital topic. Separately, AI-workflow episodes across family-office, BOMA, and Best Ever feeds show that enterprise AI adoption is broadening as an operating theme even where it does not yet create a direct office or data-center lease claim. Use these notes to track market attention and candidate source leads, not to size megawatts, power costs, or data-center rent.

The transcript / rich-episode pass gives the BOMA Suburban Chicago AI episode a clearer role: it points to AI in building operations, predictive maintenance, energy optimization, and smart-building technology. That belongs in the operating-infrastructure lane, not the hyperscaler-capex lane. It can support diligence questions about controls, sensors, vendor systems, and operating-data readiness, but it should not be used to infer data-center capacity, utility load, or AI-office absorption.

This is the largest headline bucket and the hardest for ordinary real estate capital to access directly.

Representative signals include:

• Amazon's $15B northern Indiana program with 2.4 GW
• Google's roughly $40B Texas buildout through 2027
• Stargate's $450B+ program-scale commitment and 8+ GW plan
• Amazon's government AI infrastructure lane

The core lesson is not the raw dollar amount. It is that hyperscalers route wherever power, transmission, land, and utility structure clear. Prestige geography matters far less than infrastructure readiness.

This lane is different from new campus development. It is infrastructure capital buying operating platforms.

This is the valuation lane, not the land-acquisition lane. It affects what institutional scale looks like and what private portfolios are measured against.

This lane sits upstream from real estate but increasingly governs which sites become real. DOE nuclear-financing comments point to public-sector intervention in long-lead generation procurement, while the reported NextEra / Dominion transaction shows regulated utility platforms being reframed around data-center load growth. This is not directly acquirable real estate, but it changes the underwriting context for powered land, utility partnership, and ratepayer-risk exposure.

This is the most actionable middle layer because it contains actual projects, debt, and delivery signals.

This is the lane where outside investors can often act indirectly through powered land, utility relationships, adjacent industrial, or operator and lender exposure. It should not be converted into a broad office-demand claim unless a separate office leasing source proves that specific read-through.

The Fisk University Quantum Leap source belongs outside this delivery lane even though the headline uses data-center language. The preserved article body supports an institutional Innovation Center / campus-technology read, not commercial colocation or hyperscale inventory. That distinction is useful because AI-infrastructure RSS items often blend real data-center capacity, campus technology projects, and marketing language. See Source: $1B Fisk Upgrade Includes On-Campus Data Center and Source: Prime Data Centers Breaks Ground on Second Sacramento Facility.

Crusoe's Brighton AI factory is the cleanest signal in this stack that digital infrastructure creates adjacent demand beyond the data center box itself.

The implication is that logistics, modular assembly, cooling, electrical equipment, and utility-served industrial land near confirmed compute clusters may offer cleaner and more understandable exposure than chasing every campus headline.

MacroValor's hydrogen-powered campus belongs here, not in the same bucket as financed, permitted, or clearly advancing projects.

This distinction matters because the digital-infrastructure branch now produces a lot of spectacular announcements. Some deserve underwriting attention. Some only deserve monitoring.

Northern Indiana, Haskell County, Castroville, Parker, Phoenix, and now Northern Virginia Digital Infrastructure Corridor all reinforce the same point: the winning real estate is where power, transmission, and delivery structure clear. Prestige branding and generic office identity are secondary to utility and network readiness.

The cleaner tradable edges are often:

• powered land before full campus announcement
• utility-enabled industrial parcels
• supply-chain manufacturing and modular assembly sites
• mature campuses with proven tenants and financing liquidity

Texas keeps recurring in this source set because it wins across multiple forms at once:

• metro-integrated enterprise campuses
• outer-ring hyperscale buildout
• exurban and rural power-first siting
• platform scale and operating history

That does not mean every Texas announcement deserves the same weight. It means Texas has the broadest state-level stack in the current material.

• Investors with real utility, powered-land, or infrastructure relationships
• Institutional capital targeting proven operators, mature campuses, or platform-scale exposure
• Industrial investors near confirmed compute and equipment clusters who understand spillover demand

• Treating every AI infrastructure headline as equivalent to financed or delivered demand
• Assuming metro prestige matters more than grid headroom and entitlement clarity
• Mixing platform M&A, campus announcements, and supply-chain manufacturing into one blended underwriting story

• Site-level Texas detail for Stargate rather than state-level mention only
• More Castroville and Medina County disclosure
• Whether Phoenix, Parker, and Indiana-type secondary wins keep compounding
• Which watch-list projects move into entitlement, financing, and actual construction

• Stargate remains much clearer at the program level than at the site-by-site level.
• The federal AI lane now has an initial geography node in Northern Virginia and Washington DC, but it is still thinner than the Texas and Phoenix branches.
• Some of the biggest headlines are commitments rather than fully disclosed project pipelines.
• MacroValor remains a watch-list signal, not a de-risked project.

• Source: Amazon Plans to Invest $15B to Build New Data Center Campuses in Northern Indiana
• Source: Amazon to Invest Up to $50B to Expand AI Infrastructure for U.S. Government
• Source: Google to Invest $40B in Texas Data Centers Through 2027
• Source: Reports: Microsoft Plans $400M Data Center in Castroville, Texas
• Source: OpenAI Plans to Build Data Center Near Ann Arbor, Invest $450B Over Next Three Years
• Source: SoftBank Agrees to Acquire DigitalBridge in $4B Deal
• Source: BlackRock-Led Consortium Agrees to Acquire Aligned Data Centers for $40B
• Source: Crusoe Building $200M Denver-Area AI Factory
• Source: Flexential Building 249K SF Denver-Area Data Center in Parker, CO — $192M, 22.5 MW
• Source: Edged U.S. to Open New Data Center Facility at Irving Campus
• Source: QTS Tapping Two Phoenix Data Centers for $510M Refinancing
• Source: MacroValor, Favis Will Build Hydrogen-Powered AI Mega Campus
• Source: Related Digital Secures Financing for $16B Oracle Data Center Project in Michigan
• Source: Vantage Data Centers to Invest $2B for New Campus in Northern Virginia
• Source: Prime Data Centers Breaks Ground on Second Sacramento Facility
• Source: $1B Fisk Upgrade Includes On-Campus Data Center
• Source: DOE To Finance Parts for 10 Nuclear Reactors to Power Data Centers
• Source: NextEra Dominion Utility Merger and Data Center Power Demand 2026
• Source: The Whale Hunting Era Of Data Center Development May Be Ending
• Source Collection: CBRE Insights Market Reports Public Crawl 2026
• Source: JLL 2026 Global Data Center Outlook
• Source: JLL North America Data Center Report Year-end 2025

• Texas Digital Infrastructure Corridors
• Digital Infrastructure Real Estate
• Powered Land and Grid Advantage
• Northern Virginia and Washington DC
• AI Infrastructure and Office Demand 2026
• Industrial Hub
• Analyses Hub
• United States

• Treats Blackstone's reported $5B AI-cloud venture with Google as digital-infrastructure capital-stack evidence, not a site-level data-center supply metric. See source-blackstone-google-ai-cloud-venture-2026. Caveat: Preserve as strategic capital-allocation context; verify venture terms and site-level exposure before structured import.