Last updated 2026-05-15 | Vertical landing page | For hyperscaler operators, BTS developers, and data-center REIT acquirers
Data center cooling is not an HVAC problem anymore. It is a capital-allocation problem with a thermodynamics deadline.
BLUF: Blackwell-class racks at 60–120 kW are landing inside buildings that were sized for 8–25 kW. Air-cooled CRAH plants are at their thermodynamic ceiling. Direct-to-chip and immersion are not "future tech" — they are the only physical paths through the next 18 months. AISB is the only public-facing AI fleet built for operators making the air-to-liquid transition while the rack is energized. If you are a hyperscaler operations lead, a BTS developer signing 60–120 kW leases, or an LP underwriting the Blackstone BXDC REIT IPO, this page is your decision tree.
The three numbers that anchor every decision in this market
| Anchor | Value | What it means for cooling capex |
|---|---|---|
| Hyperscaler FY26 capex aggregate | ~$700B (Microsoft + Google + Meta + Amazon + Oracle) | ~$450–480B routes to AI infrastructure; cooling is now first-order line item, not a footnote |
| Blackwell rack density | 60–120 kW/rack (NVIDIA GB200 / GB300 NVL72) | 4–10× the design density of buildings under management today; air-cooled CRAHs cannot remove the heat at this concentration |
| BXDC REIT S-1 (Apr 2026) | Northern Virginia, Ohio, Maryland, Phoenix, Austin focus | First institutional retail data-center REIT — operational data per building becomes public benchmark by 2027 |
Three more numbers worth holding in mind: Prologis Q1 2026 booked $1.3B of build-to-suit data center starts inside a $3.5–4.5B annual development guide. ConstructConnect: private office construction starts surged to roughly 4× the 2-year average — almost entirely from data center buildouts. Industrial absorption forecast for 2026 jumps to 220M sf, fueled materially by reshoring, manufacturing, and data centers (Prologis alone is ~30% of that).
What this means: the cooling-engineering decisions made between today and Q3 2026 will set the operational cost basis for roughly a decade of building life.
The retrofit decision is not "should we." It is "which path, on what schedule."
Operators come to us with one of four physical situations. The decision framework is different for each.
| Situation | Existing thermal envelope | Realistic options | Typical capex range (per MW IT load) |
|---|---|---|---|
| Brownfield air-cooled colo, 8–12 kW/rack design | CRAH + hot/cold aisle containment | (a) cap leasing at 25 kW/rack and stay air-cooled, (b) introduce rear-door heat exchangers per row for selective densification, (c) full DLC retrofit with new CDU plant | $0 / $1.5–2.5M / $5–8M |
| Mid-2010s hyperscale, 15–25 kW/rack | Indirect evaporative + air containment | (a) targeted DLC for AI training cells; air for inference + storage; (b) full DLC + immersion for densest tiers | $3.5–6M / $7–10M |
| New-build BTS, 60–120 kW/rack target | Designed from concrete: water + facility + IT cooling loops | Direct-to-chip warm-water + facility-water economization; immersion as an option for densest skus | $8–12M (baked into shell + core) |
| Office conversion to data center (rare but rising) | VAV + central plant sized for human occupancy | Gut central plant, new electrical service, new water service, ground-up cooling — typically infeasible vs greenfield economics | Usually NPV-negative; flag as feasibility blocker |
Capex bands are operator-reported ranges as of Q2 2026; verify against your equipment vendor quotes and IPMVP Option D simulation baseline. AISB does not sell equipment and will tell you when the answer is "do nothing right now."
The 9 questions our agents close in under 60 seconds, per site
Operators run these queries on real portfolios as the first 15 minutes of a strategy conversation. Each query returns a structured answer with cited sources, not a generative summary.
- What is the maximum sustainable kW/rack in this room given current CRAH plant, containment, and water service?
- What is the incremental capex per MW to introduce direct-to-chip in one row vs full hall?
- At what point does warm-water economization beat chilled-water year-round in this climate zone?
- What is the PUE / WUE delta between two compared design paths under realistic load profiles?
- What is the refrigerant-transition risk (AIM Act + EPA SNAP rules through 2027) on the equipment you are about to buy?
- Which of my existing leases have density-cap clauses that block a DLC retrofit without tenant consent?
- What is the IPMVP Option D baseline for an active-load building when no 12-month pre-retrofit baseline exists?
- What local water + power authority limits apply at this site, and when do they require a 12-month forward filing?
- What is the embodied-carbon delta (LEED v5 A1–A3 + California SB-253 Scope 3) for the two cooling architectures under consideration?
Each of these is a specific question with a verifiable answer. Generalist AI chatbots cannot answer them because the answer requires reasoning across plant data, code, IPMVP, refrigerant policy, and lease language at the same time. AISB is built to do exactly that.
Why this is a fleet problem, not a chatbot problem
Cooling is the most cross-disciplinary line item in a data center P&L. A single retrofit decision touches MEP engineering, IPMVP measurement & verification, refrigerant compliance, embodied carbon, water authority filings, lease covenants, capex underwriting, and tenant communications. No single specialist closes the whole arc.
AISB routes one cooling question across the fleet:
- cre-ts-mep-systems — HVAC/electrical/plumbing diagnostics; ASHRAE 90.1 / 90.4 anchoring
- cre-ts-energy-mv — IPMVP Option B/C/D selection; PUE/WUE baselining; CV(RMSE) discipline
- cre-ts-capital-projects — NPV/IRR/payback; discount-rate pull from FIN; RSMeans Class-3/4 estimating
- cre-ts-vendor-standards — refrigerant transition (AIM Act), code lookup, manufacturer bulletin tracking
- cre-pm-permit-regulatory — water authority filings, jurisdiction-specific permitting
- cre-con-carbon-tracker — LEED v5 A1–A3 + SB-253 Scope 3 during the build, not after
- cre-lease-intel — density-cap covenant scan across the operating portfolio
- fin-commander — discount-rate input + REIT-specific OpEx signal
The answer the operator gets is a single, sourced memo. The 57-agent fan-out happens underneath. We have not seen a competitor — VTS, Cherre, Yardi, Altus ARGUS, ProptechOS, Trane Connect, JCI OpenBlue, Honeywell Forge — answer all nine of the questions above in one workflow. (We surveyed them through late April 2026; the public landscape report is at /competitive-benchmark-cre-ai/.)
Pricing for cooling-centric portfolios
| Tier | Operator profile | Monthly | What you get |
|---|---|---|---|
| Cooling Pilot | Single facility, 1–5 MW IT load, evaluating densification path | $5,000 | 9-question structured assessment + IPMVP Option D simulation memo + density-cap lease scan (up to 6 leases) |
| Portfolio | 5–25 facility colo or BTS owner, 25–250 MW aggregate | $25,000 | Cooling Pilot scope across full portfolio + monthly retrofit-sequencing memo + refrigerant-transition watch + quarterly board-ready summary |
| Enterprise / REIT | Public or pre-IPO operator (BXDC, Equinix, Digital Realty, CyrusOne pattern), 250 MW+ portfolio | $75,000 | Portfolio scope + custom IPMVP M&V framework + LP-grade reporting feeds + dedicated commander handoff + RSMeans-anchored capex line items |
Every tier is month-to-month with a 60-day notice. No equipment vendor relationships, no commission revenue from chillers or CDUs, no white-labeling. The only commercial relationship is with the operator who pays us.
What the structured answer looks like
Below is an excerpt from a real cooling memo (operator anonymized; figures simplified for public posting):
Q: Can we lift Hall 4 from 14 kW/rack to 35 kW/rack without replacing the CRAH plant?
A: No, not at the requested 35 kW. The CRAH plant has 1.18 MW of sensible cooling installed against the 1.41 MW peak that 35 kW × 40 racks implies; air-side delta-T constraints limit you to 22 kW/rack sustained under N+1. Two paths produce different P&Ls.
Path A — Rear-door heat exchangers on the 12 highest-density racks: $1.8M capex, lifts those racks to 50 kW while leaving the remaining 28 at 22 kW. NPV positive at 4.2% WACC across 7 years assuming the AI training revenue assumption holds.
Path B — Full direct-to-chip retrofit (new CDU plant + facility-water loop): $7.4M capex, lifts entire hall to 80 kW/rack ceiling. NPV positive only if Hall 4 utilization stays above 78% through 2030.
Constraint to verify before deciding: the master lease has a 30 kW/rack density cap unless tenant gives written consent (Section 7.3). Density cap was added in the 2021 amendment.
Sources: cre-ts-mep-systems plant capacity analysis 2026-05-14; cre-lease-intel lease abstract; ASHRAE Standard 90.4-2022 §5.2; RSMeans 2026 Class 4 estimate; FIN discount rate pull.
Three things to notice: (1) the answer includes the constraint we found (the density cap), not just the math, (2) every assertion has a source, (3) the agent did not pick a path — it surfaced the path-dependent risk so the operator picks.
Three reasons not to wait
Most operators we talk to delay the cooling-strategy work because it feels like a "next planning cycle" decision. We disagree, for three reasons specific to 2026:
- Refrigerant rules are tightening. The AIM Act phasedown of HFCs (650 GWP cap on new equipment from January 2026) means equipment specified today must be checked against a forward refrigerant transition plan. Equipment ordered against an older spec is a stranded-asset risk.
- BXDC REIT IPO sets the public benchmark. Once Blackstone's vehicle is public, per-building operational data becomes a market benchmark by 2027. Operators on the wrong side of PUE/WUE distribution will see it in their cost-of-capital.
- Water authority queues are 9–18 months out. In Northern Virginia, Phoenix, and the Dallas/Austin corridor, water service applications filed today are returning permits in Q1–Q3 2027. Cooling architecture that depends on water economization needs filing before the architecture is finalized.
Start with the agent
The fastest way to test fit is to ask the agent about one specific facility you are responsible for. No registration, no email gate for the first query.
Ask the AISB agent about your cooling situation →
If you want a structured 30-minute walkthrough of the 9 questions above against a single building, we are running 6 introductory cooling pilots through end of June 2026. Pilot scope and pricing on the enterprise page.
See enterprise terms + book a strategy call →
If you want to see how the embodied-carbon side of the same building is tracked during construction, the companion page is at /embodied-carbon-tracker/.
What people ask us first
Q: Do you sell or recommend cooling equipment?
No. AISB is an intelligence fleet; we have zero equipment vendor relationships and zero commission revenue. Our recommendations are vendor-agnostic by design.
Q: How is this different from a consulting firm doing the same analysis?
Consulting firms produce one memo every 4–8 weeks for $80–250K. AISB produces one memo per question, in minutes, for a fixed monthly subscription, and the institutional memory compounds across every conversation in your portfolio. We expect to be a complement to a small in-house engineering team, not a replacement for an outside specialist on a critical design review.
Q: Can you work alongside our existing OT vendor (Trane, JCI, Honeywell, Schneider)?
Yes. Most of our cooling pilots run in parallel with an OEM contract. AISB consumes the same plant telemetry the OEM gets and adds the cross-disciplinary layer (IPMVP, code, lease, capex, embodied carbon) that the OEM scope does not cover.
Q: What happens if the agent gives a wrong answer?
Every memo lists its sources and its assumptions. If the source is wrong or the assumption is wrong, the memo is wrong, and you should not act on it. We do not over-state confidence; the answer above includes the line "Constraint to verify" specifically because the agent surfaced a lease covenant the operator had not flagged. The fleet is built around what is called "uncertainty as signal" — when the agents disagree, you see that, and you decide.
Q: Are you open-protocol?
Yes. AISB is MCP-native and built on the open BACnet / BIM / IPMVP stack. We are not a closed-loop OT platform and we will not lock your data inside our system. Detail: /agentic-buildings/.
This page is maintained by the AISB Conversion Optimization track. Last source refresh: 2026-05-14. Numeric anchors verified against Blackstone BXDC S-1 (April 2026), Prologis Q1 2026 supplemental, ConstructConnect 2026 starts data, NVIDIA GB200/GB300 NVL72 spec sheets, and AISB internal cre-ts-mep-systems capacity model. For methodology, see /ipmvp-verification/.