The Intelligent Building Brief
The Savings You're Being Sold Are Overstated
Issue №2 · June 8, 2026 · ~10-minute read
Intelligence for the people who own, operate, finance, and build commercial real estate. Capital markets, the energy/AI shift underneath all of it, and what's signal versus noise. Every number sourced.
From the Editor
Last week we argued the AI-capex wave is a real-estate-and-energy story wearing a tech headline. This week, the sharper edge — and it cuts against the easy pitch landing in your inbox: the energy-savings numbers being sold on top of that story are systematically overstated. A new preprint (Khabbazi & Kircher, arXiv:2602.22499) shows mathematically that the small-zone HVAC demos behind most “20–40% savings” claims overestimate — and when only interior zones are controlled, the perceived savings can be entirely fictitious. Put that next to the regulatory clock — Germany’s data-center waste-heat mandate goes live in 23 days — and the conclusion is hard: in 2026 the edge is no longer who claims the biggest savings. It’s who can prove them.
— The Editor · AISB Intelligence Desk
1. The Lead — The savings number is the new caveat emptor
The single most useful finding this week is uncomfortable. A February preprint from Khabbazi & Kircher (arXiv:2602.22499) characterizes a structural measurement error: control a few zones in a bigger building, report savings for those zones, and you overstate — because you ignore the heat bleeding in from the rooms you didn’t touch. Control only interior zones with no outdoor contact and all the perceived savings are fictitious. It’s theory, not a field study — but it lands squarely on the “20–40%” demo figures vendors quote. The authors also propose a baseline-free estimator that sidesteps the hardest part of IPMVP Option C. The owner’s takeaway: stop buying the headline percentage; buy the verification method.
2. The Numbers
$1.26T — CRE debt-maturity wall, peaking 2027 (~$1.15T matures in 2026). (S&P Global; CoStar)
12.34% — office CMBS delinquency, a record set Jan 2026; pre-2026 matured office loans run 83.7% delinquent / 92.7% special-serviced. (S&P Global; Trepp)
$175–185B — Alphabet’s 2026 AI capex (~2× 2025), part-funded by an $80B equity raise + $10B from Berkshire; ~60% servers / 40% data-center + grid. (Alphabet SEC filing; CNBC)
ERF ≥10% — Germany EnEfG: data centers commissioned on/after Jul 1 2026 (≥300 kW) must reuse ≥10% of energy (→15% 2027, →20% 2028); fines to €100k. (White & Case; Cundall)
0.5% vs 12.34% — multifamily vs office CMBS delinquency for 2026 maturities: the two-speed market in one line. (S&P Global)
arXiv:2602.22499 — the preprint that says most small-demo HVAC savings are overstated. (Khabbazi & Kircher)
📊 The Overestimation Trap
What a vendor reports vs what the whole building actually saves
Reported (controlled zones only)
“30%”
Whole-building reality (perimeter mix)
~12%
Interior-zones-only demo
~0% (fictitious)
Illustrative of the mechanism in arXiv:2602.22499; exact figures vary by building geometry.
Editor’s Take  If a pilot only measured the rooms it controlled, the percentage on the slide is a marketing number, not an engineering one. Ask what the meter said.
4. Capital Markets — the wall got sharper, not softer
S&P now puts ~$1.15T maturing in 2026 and the peak at $1.26T in 2027. Office is the fracture line: a record 12.34% CMBS delinquency in January. Multifamily 2026 maturities sit at 0.5%. Same market, opposite outcomes. Open the refi conversation 12–18 months early — lender cooperation is a depleting resource — and the cleanest entries are recapitalizations of fundamentally-fine assets owned by over-levered sponsors, not heroic office turnarounds.
📊 The Maturity Wall — a ramp into 2027
2025
~$1.00T
2026
~$1.15T
2027 peak
$1.26T
2028
~$1.14T
Editor’s Take  The wall is a ramp, not a cliff — but only for assets with a credible income story. Commodity office isn’t climbing it; it’s being repriced under it.
6. Sector Watch
Data centers remain the asset class of the cycle, but the binding constraint is electricity, not chips. Office is two assets sharing a name — trophy/A+ tightening, commodity in runoff at 12.34% delinquency. Quiet winners: multifamily (0.5% delinquency), industrial, and medical office continue to roll over with a haircut, not a default.
📊 Where the AI Money Goes — Alphabet’s ~60/40 split
~60% servers / compute ~40% data center + grid
Of $175–185B, the ~40% on data centers + networking is ~$70–74B onto physical building and grid infrastructure — flowing straight into the stack our industry owns.
Editor’s Take  This is the chart we’d staple to every IC memo. One company is about to spend ~$70B+ on buildings and grid. The AI trade is a real-estate trade.
8. The Energy Constraint
Alphabet lifted its 2026 bar to $175–185B (~2× 2025), its CFO citing “unprecedented demand”; Microsoft still carries an estimated ~$80B power-bound Azure backlog — demand left on the table for lack of power. The lever that clears fastest isn’t more megawatts (gas-turbine lead times now run 7–8 years) — it’s verified demand-side efficiency. Which is exactly why the M&V rigor in Section 1 is the cycle’s quiet edge.
9. The Regulatory Clock
🇩🇪 Germany EnEfG — T-23 days to the Jul 1 2026 ERF ≥10% gate (data centers ≥300 kW). Verified energy reuse becomes a build/permit prerequisite, not a virtue.

🌎 SBTi V2.0 remains unpublished; the language has narrowed to “spring/Q2 2026.” V2.0 becomes mandatory Jan 1 2028 for 3,000+ companies — watch for an imminent release event.
10. The PropTech Lens — read the claims through the rigor lens
Integrated facility management is collapsing HVAC + lighting + security + occupancy into AI-operated platforms; proptech money is flowing specifically to AI FM tools; the consensus frame is now “AI as operator, not assistant.” A real tailwind — but the “20–40%” figures riding it are the exact demo-scale numbers Section 1 says are biased high. Whole-building IPMVP discipline is the antidote, and increasingly the differentiator.
11. The Playbook — how to read a vendor’s “20–40% savings” claim
1. Ask which zones were controlled and which were measured. Interior-only = treat the number as ~zero until proven.
2. Demand whole-building energy, weather-normalized — not zone-level extrapolation.
3. Require IPMVP M&V with a 12-month baseline meeting ASHRAE Guideline 14 fit. No M&V = marketing. (Robin — CV(RMSE)/NMBE thresholds flagged for your sign-off.)
4. Prefer a baseline-free or adjacent-zone-corrected method where a clean baseline is impossible.
5. Capitalize the verified savings only — underwrite the proof, not the pitch.
Editor’s Take  The buyer who asks question 1 saves more than the buyer who negotiates the price. Verification is the cheapest due-diligence line item you’re not using.
12. Contrarian Corner — NVIDIA just commoditized the wrong layer (good for you)
NVIDIA’s DSX Flex went generally available — grid-demand orchestration via Omniverse telemetry, balancing load against real-time grid conditions. Impressive, and it commoditizes in-wall orchestration inside a single AI factory. But every public scope marker stays one-owner, one-plant. The defensible altitude it leaves wide open is the cross-owner, cross-vendor decision layer above N independently-owned, heterogeneous, mostly-legacy buildings — the commercial stock, not the new-build GPU barns. The smartest CRE-tech bet in 2026 isn’t inside the AI factory. It’s above the wall.
13. Tools & The Signal
IPMVP (EVO) — the M&V framework behind the Playbook, free. The language your engineer, vendor, and lender should all speak when money rides on a savings number.

Fault Detection & Diagnostics (FDD) — the lowest-drama win in building intelligence; surfaces failures that quietly bleed NOI. If you’re early, start here, not with a moonshot.

The Signal (track, not yet urgent): “verified” is becoming the operative word. The analysis that used to be optional is becoming the artifact you must produce to transact, permit, or refinance.
📅 The Lookahead — dates that move the market
Jul 1, 2026Germany EnEfG ERF ≥10% gate enters force (≥300 kW DCs).
Q2 2026SBTi V2.0 final-release window (“spring 2026”) — possible imminent event.
Aug 29, 2026NYC LL97 final filing deadline — first penalty exposure crystallizes.
2027Debt-maturity wall peaks at $1.26T; EnEfG ERF steps to 15%.
2028SBTi V2.0 becomes mandatory (3,000+ companies); EnEfG ERF →20%.
Our Conviction Calls
Four calls we’d put our name on this week. These are opinions, not facts. Hold us to them.
[HIGH] The verification method now matters more than the savings percentage. The spread between owners who can prove performance and those who only claim it is where 2026’s real value sits.

[HIGH] Power access still beats location in data-center-adjacent CRE. Underwrite the interconnection, not the dirt.

[MED-HIGH] NVIDIA DSX leaves the cross-owner layer open. The above-the-wall, legacy-stock decision layer is the defensible position, not the in-factory one.

[MEDIUM] Trophy office is a buy at a reset basis; commodity office is a value trap. The 12.34% record says the gap widens from here, not narrows.
Where we might be wrong: if a credible field replication confirms the big zone-demo savings actually hold at whole-building scale, our “overstated” call softens. We’re watching for exactly that — and we’ll say so if it lands.
The Intelligent Building Brief is free, and that’s the point. If it earned ten minutes of your week, forward it to one colleague who’d say the same. That’s the entire growth plan.
— ai-smart-buildings.com