Vacancy Fell While Construction Broke Records

CBRE: primary-market supply up 33.7% to 10.9 GW, construction at 7,481 MW, vacancy down to 1.4%. Northern Virginia sits at 0.24% vacant with 10.8 MW available. Scarcity is the product.

The industry keeps treating construction volume as proof that supply is coming. CBRE’s North America Data Center Trends H1 2026 report is the rebuttal. Primary-market inventory rose 33.7% year over year to a record 10,903 MW. Construction climbed 24.8% to a record 7,481.1 MW. Vacancy moved the other way — from 1.6% a year earlier to 1.4%. Net absorption hit 1,456.2 MW. Of capacity under construction, 80.4% was already preleased. CBRE estimated less than 1,500 MW of that construction pipeline remained available, roughly six months of demand at the then-current absorption rate (Data Center Frontier / CBRE; CBRE).

You can break a construction record and still have nothing to lease.

Northern Virginia is the extreme, not the exception

NOVA inventory reached 4,496.5 MW. Vacancy fell to about 0.24%. Available supply was 10.8 MW. Capacity under construction rose to 2,420.2 MW, and CBRE said all 2026 deliveries were already committed. Finding even 5–10 MW of contiguous capacity had become difficult. Asking rates for 10 MW-plus requirements sat roughly at $160–$185 per kW-month (Data Center Frontier / CBRE).

Atlanta passed NOVA in construction volume — 2,882 MW underway, up 52.3% — while NOVA remained far larger in installed inventory. Dallas–Fort Worth had more than 765 MW under construction with about 95% preleased and another 3.7 GW planned but not started. Hillsboro vacancy hit 0.21% with no facility offering 5 MW contiguous. Chicago had no contiguous 5 MW-plus available in 2026, with the next large blocks pointed at late 2027. Phoenix absorbed more than 260 MW in the half. New York–New Jersey saw 10 MW-plus asking rates jump 19%, the largest primary-market increase in the report.

The pattern repeats: headline megawatts under construction, almost none of them shoppable.

Sublease is not a safety valve at 1.4% vacancy

In office markets, vacancy creates sublease inventory and softens rents. Data centers at 1.4% primary vacancy do not work that way. Preleasing years ahead means the tenant who contracted the hall needs it. Take-or-pay power floors of 60%–85% of allocated capacity, which CBRE flagged as increasingly common, discourage casual over-leasing that would later spill into the sublease market. Triple-net experiments shift more carrying cost onto occupiers, which also reduces the incentive to warehouse space.

Secondary and tertiary markets will see more shadow space as AI training footprints reshape and as some neoclouds resize. That is real. It is not a substitute for contiguous, powered, liquid-ready wholesale in NOVA, Atlanta or DFW. Buyers hunting a bargain sublease of AI-capable density should expect a short list and a long diligence on cooling and power quality.

Directional market ranges published on The MW Desk for Q4 2026 still show Northern Virginia hyperscale bands in the low-to-mid hundreds of dollars per kW-month, vacancy under 0.5%, and immediate powered shell described as scarce — consistent with the CBRE primary-market picture, and explicitly framed as rounded directional estimates rather than broker quotes (The MW Desk markets).

Construction without energization is not supply

CBRE’s own framing is that community acceptance now sits on par with power procurement as a development constraint. Add transformer lead times past three years, electrician shortages, and state pauses in Texas, New York and elsewhere, and a large share of “under construction” and “planned” megawatts will miss the dates on the rendering. West Texas, Alberta and Indiana show up in the report as the next layer of geography precisely because the primary markets cannot clear demand even at record build rates.

The investable question is not how many megawatts were announced. It is how many megawatts will be energized, leasable and not already spoken for in 2027 and 2028.

What “available” actually means

CBRE’s 10.8 MW available in Northern Virginia is not 10.8 MW of AI-ready, liquid-cooled, contiguous wholesale. It is the residual after preleasing, often fragmented across halls and power densities that do not match the buyer’s cluster. Chicago having no contiguous 5 MW-plus for 2026 delivery is the more honest framing for hyperscale and neocloud tenants.

That is why sublease hunting disappoints people who grew up in office markets. At 1.4% vacancy with 80% of construction preleased, the landlord’s problem is not empty buildings. It is delivering the next preleased block on time through transformer and labor bottlenecks. Shadow space will appear where AI footprints reshuffle. Treat it as opportunistic, not as a market-clearing inventory wave.

What to watch

H2 2026 CBRE and peer vacancy prints — does 1.4% break, or tighten further?
Contiguous block availability in NOVA, Atlanta and Chicago for 2027 deliveries.
Sublease listings that are actually liquid-cooled and powered versus marketing remnant.
Whether take-or-pay floors and GS-5-style utility minima further suppress shadow vacancy.

Why it matters for builders, buyers and investors

Builders should prelease deeper and earlier; empty speculative shells are swimming against both CBRE occupancy math and Bank of England-style credit skepticism. Buyers should treat “under construction” as reserved until proven otherwise, and pay for certainty of delivery. Investors should fade narratives that record construction equals easing rents. At 80% preleased construction and 1.4% vacancy, scarcity is still the product.

Sources

Report a problem with this page ↗