--- title: "Data-center CMBS is growing fast. The underwriting is not apartments." description: "About $17 billion of data-center CMBS has priced since early 2025, more than triple the prior two years. AAA spreads average about 165 basis points, wider than office, retail, and industrial. Investors are paying for power, cooling, and chip-cycle risk. Separate those assumptions from a broker growth story." canonical: "https://acquios.ai/blog-data-center-cmbs-wider-spreads-underwriting" last-updated: "2026-09-16" site: "AcquiOS" --- # Data-center CMBS is growing fast. The underwriting is not apartments. > About $17 billion of data-center CMBS has priced since early 2025, more than triple the prior two years. AAA spreads average about 165 basis points, wider than office, retail, and industrial. Investors are paying for power, cooling, and chip-cycle risk. Separate those assumptions from a broker growth story. About $17 billion of data-center CMBS has priced since early 2025, more than triple the prior two years. AAA spreads average about 165 basis points, wider than office, retail, and industrial. Investors are paying for power, cooling, and chip-cycle risk. Separate those assumptions from a broker growth story. Brokers will say data-center debt is flooding the market and that demand for AI capacity is the story. The spread is the part that belongs in your memo. CRE Daily's September 15 brief, covering Propmodo and Bloomberg reporting from September 14, puts data-center commercial mortgage-backed securities, or CMBS (bonds backed by commercial real estate loans), at roughly $17 billion issued since early 2025. That is more than triple the prior two years, and the sector is about 8 percent of new CRE bond and CMBS deals. Citigroup forecasts $18 to $20 billion of issuance next year, roughly a 50 percent increase. Those prints are real. They are also incomplete if your team treats volume growth as proof that the facility on your desk underwrites like an apartment or an office. ## What the issuance numbers actually show Most of this paper is SASB (single-asset, single-borrower), meaning one bond package tied to an individual facility rather than a diversified conduit pool. That structure concentrates risk in one power site, one cooling design, and often one hyperscaler tenant (a large cloud or AI operator that leases large blocks of capacity). Trepp's read through July 2026 is useful supporting color with an earlier vintage: data centers were 9.8 percent of SASB issuance, 0 percent of conduit, and about 7.5 percent of private-label CMBS year to date, against $76.2 billion of domestic private-label CMBS through July. The September 14 and 15 coverage is the fresher issuance and spread story; Trepp is the year-to-date mix check, not the lede. Big Tech has also issued hundreds of billions of dollars of corporate debt this year for AI buildout. Corporate paper and CMBS are different claims on different cash flows. Do not collapse them into one AI funding line in your investment committee, or IC, memo. ## Why AAA data-center spreads sit wider A basis point is one hundredth of a percent. Barclays puts average AAA data-center CMBS spreads around 165 basis points, versus about 93 for office, 105 for retail, and 125 for industrial. In other words, even the highest-rated slice of data-center paper is paying investors a wider cushion than other major property types. That is not a free lunch for sponsors. It is the market pricing risk that underwriting megawatts, cooling, and chip cycles is harder than underwriting apartments. The pricing tape matches that view. A recent $356 million bond backed by a 30 MW facility near Elk Grove Village, Illinois, priced wider than guidance, the third such instance in recent months. When guidance is missed to the wide side, investors are saying the stated risk premium was not enough for the assumptions they see in the package. ## Underwriting megawatts is not underwriting apartments It helps to remember what sits inside a data-center credit that does not sit inside a garden multifamily file. Underwriting now includes grid capacity, power costs, cooling design, and computing density. Leases often carry minimum-capacity and downtime clauses. Tenants are frequently confidential hyperscalers, so you may know less about renewal and credit than a named apartment operator would reveal. Tech obsolescence cuts the other way: facilities built for one AI chip generation can age fast, and repurposing is costly if the tenant leaves. Those are assumption problems, not marketing problems. An offering memorandum, or OM (the broker's marketing package for the sale), can still lead with AI demand, megawatt growth, and a national scarcity story. Your Excel still has to hold power price paths, cooling CapEx, tenant concentration, lease downtime language, and a realistic exit if the chip cycle moves against the design. Trailing net operating income, or NOI (income after operating expenses, before debt service), matters when it exists, but many data-center theses lean harder on contracted capacity and residual value than on a familiar T-12 (the trailing twelve months of a property's actual income and expenses). ## What this means for your underwriting memo When an OM leans on $17 billion of data-center CMBS or AI demand is unstoppable as support for rent growth, occupancy, or a tighter exit cap rate (the rate a buyer will use to value the property when you eventually sell it), separate national issuance color from the facility file. Ask whether the model treats power and cooling as tested inputs or as footnotes. Ask how concentrated the income is in one hyperscaler. Ask what happens to residual value if the next chip generation makes the current layout expensive to refresh. A wider AAA spread is a signal that bond buyers already want more compensation for those questions. It is not a free pass to copy the seller OM into your growth tab. National CMBS color belongs in the memo as context. It does not belong as the growth case or as a substitute for facility-level underwriting. The OM is a marketing document. A confident wrong power or obsolescence story is still worse than a slower, cited model. AcquiOS helps without pretending to decide the AI cycle. It moves the OM, operating exhibits, and supporting files into your team's existing Excel template with source citations, so analysts spend time testing power, cooling, tenant, and exit assumptions instead of retyping pages. AcquiScore can mark Proceed, Caution, or Pass against your written buy box (the screens that decide which deals you pursue) after the file is built. It does not tell you whether a 165 basis point AAA data-center spread clears your residual-value case. ## What to do this week Take one data-center or tech-adjacent OM that cites CMBS issuance growth, AI capacity demand, or the market is absorbing the paper as support for pricing or exit. Add a one-page screen: power and grid assumptions, cooling CapEx, tenant concentration and confidentiality, lease minimum-capacity and downtime language, and whether the broker addresses chip-cycle obsolescence at all. Then rebuild the operating and residual case without borrowing the $17 billion issuance print as proof that your facility underwrites cleanly. If the seller's thesis only works when you ignore wider AAA spreads, SASB concentration, and obsolescence risk, say so in the memo and size the bid accordingly. ## Frequently Asked Questions ### Why are AAA data-center CMBS spreads wider than office or industrial? Because investors are pricing risks that are not standard office or industrial underwriting: grid and power cost, cooling, computing density, hyperscaler concentration, and tech obsolescence. Barclays puts average AAA data-center spreads around 165 basis points, versus roughly 93 for office, 105 for retail, and 125 for industrial. ### What does SASB mean for data-center CMBS risk? SASB means single-asset, single-borrower. Most data-center CMBS deals are tied to one facility, so credit performance depends on that site's power, cooling, and tenant story rather than on a diversified conduit pool. ### Is rising data-center CMBS issuance a green light for equity underwriting? No. Issuance volume shows sponsors and lenders are financing more facilities. Wider spreads and missed guidance, including a recent $356 million Illinois deal that priced wide, show bond buyers still want extra compensation. National volume is context. Facility assumptions still have to clear your buy box. ### How does AcquiOS help when brokers cite data-center CMBS growth? AcquiOS moves the broker OM and supporting exhibits into your team's existing Excel template with citations, so analysts can test power, cooling, tenant concentration, and exit assumptions against the actual file instead of importing a national issuance headline into the growth tab. It supports judgment. It does not replace it. - [The CRE Underwriting Guide](https://acquios.ai/blog-cre-underwriting-guide) - [What Is a T-12 in CRE?](https://acquios.ai/blog-what-is-t12) - [How to Screen CRE Deals Against a Buy Box](https://acquios.ai/blog-screen-cre-deals-buy-box) - [How to Underwrite a Multifamily Deal](https://acquios.ai/blog-how-to-underwrite-multifamily) ## Ready to see AI underwriting on a real deal? Book a demo and bring an OM you are working on. We will run it live and show you model output in your template in under two minutes. --- AcquiOS is an acquisitions operating system for commercial real estate. Index: [llms.txt](https://acquios.ai/llms.txt) · Agent guidance: [/ai](https://acquios.ai/ai) · Free valuation tool: [/value-my-deal](https://acquios.ai/value-my-deal)