The UK Court of Appeal has crushed Microsoft’s hopes of overturning a ruling that legitimises the resale of pre-owned software licences. On July 7, judges unanimously dismissed the American giant’s appeal against ValueLicensing (VL), a company built entirely on buying and reselling second-hand on-premise licences. The verdict not only cements the secondary software market but also spotlights a dynamic that directly affects anyone currently evaluating whether to move AI workloads – or keep them – inside corporate boundaries.

The case stems from VL’s 2021 claim that Microsoft deliberately stifled the used-licence market by inserting restrictive clauses: better subscription deals for customers who agreed not to resell old licences. VL argued this dried up the supply of pre-owned on-premise licences, artificially inflating costs for those choosing not to migrate everything to the cloud. The damages sought total £270 million.

Microsoft tried to reframe the dispute in October 2025, arguing that Office – with its icons, clipart and help files – was a creative work protected by the Copyright Directive rather than the software licensing rules underpinning the landmark UsedSoft ruling. It also claimed that bulk licences (e.g., a 1,000-seat purchase) could not be broken up and resold piecemeal. Both arguments were dismantled: the Court of Appeal called the first “odd” and noted it would mean adding a couple of clipart files could circumvent free tradeability, and dismissed the second as groundless.

The judgment’s relevance extends far beyond the immediate claimant. The presiding judge, Justin Turner KC, also oversees Alexander Wolfson’s collective action against Microsoft – a case that, after this ruling, could expose the company to billions in liability. Wolfson was allowed to intervene in the appeal and supported VL’s position.

For those tracking on-premise deployment logic, the ruling restores substance to an often-overlooked economic pillar: the right to resell what you bought. When a company purchases perpetual licences for AI environments – be they LLM serving platforms, orchestration suites or software accelerators – the residual value of those licences can be monetised on the secondary market. In a Total Cost of Ownership analysis, this introduces a variable that has been compressed by restrictive vendor practices. Knowing that a contract won’t permanently lock the asset, and that part of the investment can be recouped, lowers the entry threshold for self-hosted solutions compared to pure cloud subscriptions.

The other second-order effect is pressure on large vendors to redesign the boundary between service and product. If resale threatens recurring revenue, the response may accelerate the bundling of features into cloud-only subscriptions – a trend already visible with language models offered exclusively via APIs. Yet for organisations with sovereignty constraints, air-gap requirements or fine-grained data control, this judgment revalidates software ownership as a choice that is not just technical but asset-based. The cost of running models locally, on owned hardware, is also measured by what that investment might be worth to someone else tomorrow.