TL;DR
- AI tenants are the cycle's marginal demand engine — concentrated in SF, NYC, London, Toronto, Tokyo.
- Underwriting profile is unique: large headcount ramps, dense seat plans, heavy power and HVAC requirements.
- Frontier labs increasingly negotiate full-floor expansion stacks rather than single floors.
- ESG and amenity performance now table-stakes — but power capacity is the binding constraint.
- Landlords with surplus power and chilled-water capacity are commanding measurable rent premiums.
Demand picture
AI labs and platform AI vendors became the largest single source of Class A absorption in 2024–25 in San Francisco, with material spillover into New York (Hudson Yards, Midtown South), London (Shoreditch tech belt, Kings Cross), Toronto (King West, Liberty Village), and Tokyo (Roppongi, Shibuya). The defining feature of these requirements is an aggressive 24-month headcount ramp: tenants taking 50,000 sqft today underwrite 150,000 sqft in 18 months.
Underwriting profile
AI tenants typically run dense seat plans (60–80 sqft per seat versus the Class A average of 120–150), heavy in-floor power requirements (15–25 W per sqft versus a typical 6–8 W), and elevated HVAC loads from on-floor compute. Many require Tier III equivalent power and HVAC redundancy on a portion of the floor for in-floor inference and dev infrastructure.
Power as the binding constraint
The gating Class A factor for AI tenants is no longer rent or amenity — it is power capacity. Landlords with surplus primary-power feeds, expandable risers, and chilled-water capacity are commanding measurable rent premiums. The fastest-growing Class A capex line item across our coverage cities is power infrastructure upgrade.
Lease structure
Frontier AI labs increasingly demand full-floor expansion stacks (hard expansion rights on three to five contiguous floors above the initial premises) rather than the standard single-floor expansion option. Landlords accommodate by carving out blocks of building NRA at signing, often at premium pricing.
Frequently asked questions
- Are AI tenants creditworthy?
- Frontier AI labs typically come with significant fundraising covenant rather than operating cash-flow covenant. Landlords increasingly require letter-of-credit security packages 18–24 months of base rent rather than the standard 6–12.
- Do AI tenants need data-centre infrastructure on the floor?
- Most do not — production training runs in dedicated data centres. But the on-floor inference, dev, and dense compute racks require power and cooling beyond standard Class A specifications.