The UK 40% First-Year Allowance changes AI capex maths in 2026

By Dr Aalok Shukla, CEO · Published 1 June 2026 · Updated 1 June 2026 · 9 min read

From 1 January 2026, plant and machinery gets a 40% First-Year Allowance

HMRC (the UK's tax authority) confirmed the new First-Year Allowance in its 26 November 2025 policy paper, with legislation following in Finance Bill 2025-26. The 40% FYA applies to qualifying main-rate plant and machinery expenditure incurred from 1 January 2026. The Office for Budget Responsibility has costed the measure at £1.035 billion in Exchequer impact for 2026-27, rising to £1.505 billion in 2027-28, which gives a sense of how broadly HMRC expects the relief to bite.

At the same time, the main-rate writing-down allowance drops from 18% to 14% on a reducing balance basis, effective 1 April 2026 for Corporation Tax and 6 April 2026 for Income Tax. So the tax picture for capital assets shifts in two directions at once: a much larger up-front deduction in year one, then a slower tail.

This matters for any UK CFO planning AI infrastructure spend. The 40% FYA is permanent, unlike the temporary super-deduction it loosely echoes, and it extends to unincorporated businesses and assets bought for leasing, neither of which can use full expensing today. For finance directors weighing a self-hosted AI rollout, the post-1 January 2026 tax position is genuinely different from the position they modelled at Q4 2025 board approval. AIOS Command sits on the other side of the line entirely, since it is delivered as a subscription. More on that in the next section.

AI software-as-a-service subscriptions are revenue expenses, not capital

The cleanest CFO answer first. SaaS subscriptions, including AI agent platforms billed monthly, are revenue expenditure. ACCA's February 2026 guidance and HMRC's capital allowances overview both treat off-the-shelf software licences and annual subscriptions as deductible in full against trading profits in the year incurred. No capital allowances claim, no pool, no writing-down tail.

That means AIOS Command (from £250/mo) enters the P&L as an operating expense and is 100% deductible against the corporate tax charge in the year it is paid. There is no 40% FYA event, because there is no qualifying capital asset to claim it against. The same logic applies to seat-based AI copilots, API token spend, and managed AI agent retainers.

The practical implication: the new 40% FYA does not make AI software cheaper. It makes capitalised AI infrastructure cheaper, but not for everyone, and not for the cases where most UK mid-market operators actually spend their AI budget. Most of those operators (per a 2025 Department for Science, Innovation and Technology survey, with 23% of UK businesses now using some form of AI) are buying SaaS, not buying GPUs. AVA (the platform's enquiry-and-inbox operator), DEX (the deal-flow analyst), LEXI (the compliance-and-policy reader), KIA (the customer-success orchestrator), and KORA (the back-office automator) are all delivered as part of an AIOS Command subscription, not a hardware install.

The £1 million AIA ceiling decides which assets need the 40% FYA

Before the new FYA bites, the existing Annual Investment Allowance (AIA) of £1 million still gives 100% relief on qualifying plant and machinery in year one. The AIA was permanently set at £1 million from 1 January 2019 and remains there for 2026. It is available to companies, sole traders, and partnerships, but a £1 million cap is per business per year.

So the running order for a UK mid-market firm buying new AI hardware in 2026 is:

• Annual Investment Allowance first. 100% relief on the first £1 million of qualifying spend.
• Full expensing next (companies only, new and unused). 100% relief, no cap.
• 40% First-Year Allowance on main-rate expenditure that does not qualify for full expensing, such as assets bought for leasing or by unincorporated businesses.
• 14% writing-down allowance on the residual balance, per year, reducing balance.

For a typical UK mid-market firm spending under £1 million a year on plant and machinery (the vast majority, per HMRC's own impact assessment of 650,000 affected businesses), the AIA covers everything and the 40% FYA never enters the calculation. The FYA matters when a finance director is signing off a multi-million-pound capex programme, when the business is unincorporated, or when assets are being bought for UK leasing.

AI hardware and on-premise deployments are where the 40% bites

The narrow population that genuinely gains from the 40% FYA, in an AI context, is the firm capitalising significant on-premise AI infrastructure: GPU servers, networking and storage to support self-hosted models, and the specialist computing kit that some UK enterprises run for regulated workloads. HMRC's plant-and-machinery definition explicitly covers computers, laptops, servers, networking equipment, and most IT hardware as main-rate expenditure.

Three CFO scenarios where the new FYA actually changes the answer:

• Capex above the AIA ceiling. A UK mid-market group spending £2 million on AI-related plant in 2026 takes £1 million via AIA and (if a company) £1 million via full expensing, or, if not eligible for full expensing, £400,000 via the 40% FYA plus £600,000 entering the main pool at 14%.
• Leasing or unincorporated. A partnership or sole trader that does not qualify for full expensing now has a 40% upfront write-off where it previously had only WDAs. The new FYA explicitly extends to assets bought for UK leasing (overseas leasing excluded).
• Slower-tail asset planning. For any spend that does enter the main pool, the rate drops from 18% to 14%. That is a slower recovery of the residual 60%. A pure WDA approach gives roughly 5.6 percentage points less relief in years two through five than the previous 18% rate, before the long tail catches up.

For most UK mid-market firms running their AI through SaaS platforms and managed agents, none of this applies. For the smaller cohort with serious on-premise AI ambitions (typical of regulated financial services, healthcare, or specific public-sector workloads), the new FYA is a real planning input. Either way, the maths is now different from the maths CFOs ran at Q4 2025 board.

Connect and operate all your systems in one place.

The CFO question underneath all of this is not really about capital allowances. It is about which AI delivery model gives the cleanest economics and the fastest path to operating leverage. Capitalising on-premise hardware feels familiar and gives a one-off tax break, but it lands you with depreciation, refresh cycles, talent costs, and a 14% writing-down tail. SaaS gives you a P&L line, full deductibility, and the ability to scale agents up or down each month. The 40% FYA narrows the gap on the capex path, but it does not close it for most UK operators.

AIOS Command sits in the SaaS camp by design: a two-layer model with an insight team (read-only analysis across every connected system) and an action team (AI agents that act on what the insight team finds). Case studies across multi-site retail, dental, fitness, and field-services show the pattern: connect first, see the leakage, then deploy named agents. AIOS Workforce handles the agent-led execution layer for ops, sales, customer success, and finance.

For CFOs already running 2026 capital plans, the cleanest question is: which AI spend belongs above the line and which belongs below? AI agent token spend is opex and should be governed accordingly. The continuous close is a P&L outcome, not a capex programme. Self-hosted GPU clusters are capex and now attract the 40% FYA if they spill over the AIA. Treat each line on its own merits, and the new allowance lands in the right place.

CFO action list for AI capex planning in 2026

Five steps to close out before the next board pack:

• Split AI spend by treatment. Tag every line as revenue (SaaS, API tokens, retainers) or capital (hardware, capitalised internally developed software). Capital lines need a capital allowances claim. Revenue lines do not.
• Confirm AIA headroom for 2026. The £1 million ceiling is per business per year. Multi-entity groups should plan AIA allocation across companies before AI hardware orders land.
• Update the post-1 January 2026 capex model. Switch the residual main-pool rate from 18% to 14% in any multi-year capex model. Apply 40% FYA on the first year for spend above AIA where full expensing does not apply.
• Check leasing structures. If your AI hardware is being acquired through a UK lease-back or third-party leasing partner, the new FYA opens a relief route that was previously unavailable. The exclusion is for overseas leasing.
• Reconcile board-level AI ROI with after-tax cash. The headline AI ROI a vendor quotes is pre-tax. The CFO version is post-tax, post-allowances. For SaaS, that gap is small. For capitalised hardware, it now widens by up to 40 percentage points in year one.

If the question on the next board is "should we capitalise our AI stack or stay on SaaS?", the answer for most UK mid-market operators is to stay on SaaS, take the full deduction in year, and pay only for the agent work you actually consume. For the minority running serious self-hosted infrastructure, the 40% FYA is a real input but not a reason in itself to switch. Either way, the maths is moving on 1 January 2026.

Frequently asked questions