Are AI occupiers about to rewrite the UK institutional office lease?

AI is now the single biggest source of new demand in the London office market, and it is still only in its infancy. Yet it arrives wanting everything the institutional lease was built to resist.

The demand story

CBRE has recently released its second-quarter data on the Central London office market (CBRE, "Central London Office Take-Up Rises in Q2 as AI Occupiers Drive Demand", 27 July 2026), and it contains a number worth pausing on. Take-up by AI firms in the first half of 2026 is already more than four times the same period last year, and more than double the whole of 2025.

This is a legal story, not just an agency one. CBRE projects that AI-led take-up could reach 4m sq ft by 2033, equivalent to around 43% of the Central London pipeline currently under construction and available to let. A market that has spent several years worrying about a structural shortage of best-in-class space is now looking at a single occupier category capable of absorbing nearly half of what's coming. That is not a cyclical blip to be waited out. It is a structural shift in who the tenant of the next decade actually is. Institutional leasing has not yet caught up with what that tenant wants.

Why the AI occupier is different

The AI occupier does not behave like the tenant the institutional lease was designed for. CBRE's sample of 100 London AI start-ups found 75 sitting in flexible space and only nine in conventional offices. Flex is the default, not the exception. As they scale, they graduate into traditional space, and numerous AI developers have already made that journey from a serviced desk to a substantial conventional letting. But they bring the expectations of flexible space with them: fitted, turnkey premises, optionality to grow or shrink, and short decision cycles.

Set against that is the underwriting reality. Most of these businesses have little or no trading history. Some are pre-revenue. Many are backed by venture capital rather than balance-sheet strength. The occupier's demand for flexibility and room to grow sits opposite the landlord's need for covenant certainty over a 15-20 year hold. The institutional lease has traditionally resolved that tension by assuming the second and pricing out the first. That assumption no longer holds, and that is where the drafting has to change.

How the institutional lease has to flex

1. Underwriting the unproven covenant. There is often no trading history to underwrite, only potential. Covenant due diligence is extending beyond the tenant entity to the venture capital funds behind it: who they are, how they've supported portfolio companies before, and whether that support is contractual or merely reputational. The solution is a security package built for a thin covenant rather than a strong one. Rent deposits get larger and longer, with mechanics to release or reduce them as the covenant matures. Group or parent guarantees are harder to rely on where the "parent" is itself overseas or equally young. And, where a conventional guarantee isn't available, expect alternative security, such as rent paid in advance, letters of credit, escrow or cash-collateralised arrangements.
2. Term, flexibility and its price. Expect shorter terms and staged commitments in place of the traditional 10-15 year institutional term, tenant breaks (increasingly rolling breaks) weighed against the "institutional acceptability" of the resulting income, and expansion or contraction mechanics, from pre-emption rights to grow-into and give-back options, built for an occupier that genuinely doesn't know its own footprint in three years' time. None of this is free. The price of flexibility should be drafted, not assumed. Incentive packages should carry clawback on early break or default, drafted with the break clause rather than as a separate negotiation, and turnover, stepped or index-linked rents should let the landlord share in the occupier's growth in exchange for taking the upfront risk. The exit consequence follows through the whole chain. Short, broken or turnover-linked income is priced differently on a sale, feeding straight into the yield an investor will pay. 
3. Dealing with the space. Wider alienation rights, meaning assignment and underletting flexibility that accommodates funding rounds, restructurings and group reorganisations, sit uneasily against a landlord's instinct to control who occupies its covenant. Sharing with group companies and occupation by non-tenant entities, routine in fast-scaling tech, needs addressing head-on rather than left to a narrowly drawn sharing clause.
4. Power, M&E and the operational lease. Where an AI occupier runs compute on site, rather than relying on third-party or cloud infrastructure, the technical requirements start to look closer to a data centre than a conventional office: guaranteed power allocation drafted almost like an energy contract, cooling and resilience infrastructure, and clear rules on who pays, consents and reinstates. Service charge and sustainability provisions need to flex for amenity-rich, energy-intensive occupation. Alterations and reinstatement clauses have to anticipate heavy technical fit-out, and AI occupiers will push hard for dilapidations caps given the scale of what they're installing. Fit-out and capital contributions increasingly need to fund genuinely turnkey, Category A+ space. Landlord capital is being committed against a covenant that is, by definition, unproven.
5. The practical letting overlay. Security and access arrangements often need to go well beyond a normal office letting. There is a live planning question too: does a standard Class E office use class comfortably cover ancillary server rooms and heavy compute installations, or does this need separate thought? Safety risk and insurance need calibrating to a genuinely different pattern of building use, rather than carried over from a conventional office lease.

The operator/JV answer

An important question is whether this flex-to-conventional pattern will push UK institutional capital toward a structure it already knows how to build, in managed and operator-led platforms. Landlord-operator joint ventures, management agreements and hybrid lease-plus-service arrangements could let institutional capital capture growth-stage AI demand without writing a full institutional lease against an unproven covenant. This structure has real value: it lets the landlord underwrite the building and the operating platform, rather than the tenant alone, sharing in the upside of growth without carrying the full downside of an unproven business.

The winners in the UK office market will be the landlords and advisers who redesign the lease, the underwriting and the exit assumptions around this occupier, not those who wait for AI tenants to look like traditional ones.