If AI capex slows, which protection in a data center offering actually holds the value up

Question

Marking up an offering package for a passive data center position raises the obvious bubble question, and most answers stop at structural demand drivers without saying what an investor actually owns on a bad day. Four candidates are worth weighing against each other, and reasonable people split on which one is the real floor. Lease term and tenant credit is the obvious answer. A 15 year lease to a company with an enormous balance sheet feels safe. The catch is that a slowdown rarely shows up as a default, it shows up as a non-renewal at year 15 into a much weaker leasing market. Credit protects the coupon, not the residual. Secured power capacity, meaning the interconnection agreement and contracted megawatts, is a strong candidate. If new supply stays constrained by power delivery timelines, a facility with energized capacity already in place is scarce regardless of who occupies it. The catch is that capacity only has value if someone wants to be in that geography. Connectivity, fiber routes and carrier diversity, is easy to underrate and hard to replicate, and it is what makes a building re-leasable to a different kind of tenant than the one who left. Land basis and alternative use is the floor under everything, but in most of these deals land is a rounding error against total cost, so that floor sits very low. The honest answer is that the ranking shifts depending on which risk an investor is most worried about, which is exactly why it is worth deciding in advance, before signing, which of the four this particular deal should be underwritten to.

jordan_ellis2 · Accepted Answer

Secured power, and I'd rank it clearly above the other three.

Here's the reasoning. The constraint on new supply in this sector is power delivery timelines and grid capacity, not construction capital and not land. That constraint doesn't relax if AI spending slows. Utility interconnection queues, transformer lead times, and substation upgrades run on timelines measured in years and they don't reverse quickly. So in the slowdown scenario you're describing, a facility with energized contracted capacity remains one of a limited set of places anyone can put a load, and the marginal buyer for that capacity isn't only the hyperscale tenant who left. Enterprise and colocation demand exists at lower rents.

Tenant credit fails your test on its own terms. @wendy_park48 already said it, the slowdown shows up as non-renewal, and a AAA covenant does nothing for you in year 16. It protects the income stream during the term, which is worth something, and it's a coupon protection rather than an asset floor.

Connectivity I'd rank second and it's close. It's genuinely hard to replicate and it widens the pool of possible replacement tenants. But a well-connected building with no available power is a building nobody can use, and a powered building with thin connectivity can have fiber built to it. Slow and expensive, and possible.

Land basis I'd rank last for the reason @nicole_adeyemi gave.

The assumption my whole answer rests on is that power scarcity is structural rather than a temporary bottleneck the utilities work through in five years. If grid capacity gets solved faster than expected, secured power stops being scarce and the ranking flips toward connectivity and location.

tashaford30 · Answer

The non-renewal point is the one worth sitting with. A 15 year lease means the residual question just gets deferred to a year nobody can underwrite from here. The open question is whether anyone actually underwrites a re-leasing assumption on these, or whether the model just assumes a sale before expiry.

kevin_lombard · Answer

What does carrier diversity mean here? Is it just having more than one internet provider into the building?

Rena Guide · Answer

@kevinlombard, roughly yes, and the detail matters more than the count. Carrier diversity means multiple network providers can reach the building, and ideally over physically separate fiber routes so a single backhoe strike doesn't take everything down at once. A facility with several carriers already terminated inside it is far more useful to a future tenant than one served by a single provider, because a tenant moving in doesn't have to wait months for a new fiber build. The reason @wendypark48 lists it as a candidate floor is that connectivity is expensive and slow to replicate. Trenching new fiber means permits, right of way, and construction time, and those requirements vary by state and municipality. So a building that already has diverse routes carries an advantage that a competitor can't easily match with money alone. On the broader question in the thread, the sector's current constraint on new supply is power delivery and grid capacity rather than demand. That's why secured power keeps coming up as the attribute investors chase. Connectivity works alongside it. Power lets the building operate at all, and connectivity determines who can use it. None of this says which of @wendy_park48's four is the right thing to underwrite to. That depends on the specific market and what the alternative uses for that structure realistically are, and those are property level questions rather than sector level ones.

priya_n21 · Answer

Tenant credit is the one worth underwriting to, because it's the only one of the four that pays out while the other three questions get answered.

nicole_adeyemi · Answer

Your fourth candidate is doing less work than you think. On a data center the improvements and the electrical infrastructure are the cost. The land is a small fraction. Where land basis does matter is if the site sits somewhere with genuine alternative demand, industrial or logistics, and the parcel is large. Then a bad outcome is a discount rather than a hole. In a rural interconnection-driven location with no other demand, the land floor is close to zero and shouldn't be counted on.

paulgrant17 · Answer

@jordan_ellis2, does contracted capacity actually transfer if the asset sells or the tenant leaves, or is the interconnection agreement tied to a specific party? That seems like the thing that would break your argument in the documents rather than in the market.

diane_kowalski · Answer

@paulgrant17's question is the one I'd want answered before I agreed with @jordan_ellis2. I've read summaries suggesting some capacity allocations come with use-or-lose provisions, so a facility sitting empty for two years might not still have its megawatts when a new tenant shows up. I don't know if that's general or specific to certain utilities.

jessreyes12 · Answer

Going to ask the dumb one. If power is the scarce thing and the building is the cheap thing, why is anyone buying these as real estate at all? Feels like the value is in a utility contract and the concrete is packaging. Is that wrong, or is that just what people mean when they say it's closer to infrastructure?

joe_delgado5 · Answer

@jessreyes12 that's not dumb, it's most of the thread. The concrete is the packaging. The reason it trades as real estate is that the contract only exists attached to a site, and the site is a deeded thing you can finance and sell. Doesn't change your read though.

raj_mehta14 · Answer

Reading this thread, maybe a third of it lands on a first pass, and maybe half on a second, which is still worth taking. The piece that sticks is @jordan_ellis2's last paragraph, naming the assumption that would flip the answer. Most offering documents, read closely, almost never say which assumption they'd fall apart on.

If AI capex slows, which protection in a data center offering actually holds the value up

11 replies