SAP has sidestepped an antitrust fine that could have run into the billions. The European Commission said it would accept a set of commitments from the German software group, closing an investigation into how SAP handles maintenance and support for its on-premise enterprise software. The potential penalty was enormous, but the real story is not the avoided fine – it’s the crack opening in the wall of vendor lock-in.

For decades, organizations running SAP on their own servers – banks, manufacturers, multinationals – could turn only to SAP for updates, patches, and troubleshooting. That recurring revenue model gave the vendor predictable income while squeezing customer choice, with annual maintenance fees often exceeding 20% of license value. The EU investigation zeroed in on this asymmetry, suspecting SAP was effectively blocking third-party providers from offering equivalent support.

With the commitments now accepted, the market opens up. Independent providers will be allowed to compete on maintenance and support, and SAP has pledged to remove technical and contractual barriers. For CIOs, the total cost of ownership equation shifts – a timely development as inflation and cloud migration push TCO calculations into the spotlight. For critical ERP workloads, staying on-premise has always meant dealing with a single provider. That stranglehold can now be loosened, without giving up direct control over data and infrastructure.

The SAP case lands as many companies are reassessing digital sovereignty for emerging workloads, including LLMs. The ability to choose who maintains vital software is a key piece for anyone architecting local deployment strategies. Buying GPUs and orchestrators is only part of the puzzle: you need a support ecosystem that doesn’t recreate the same lock-in at a different scale. The European decision could influence other enterprise software vendors – from Oracle to Microsoft – at a time when antitrust authorities are looking not just at prices, but at exit costs.

There is a third-order implication, less immediate but structural. A more contestable support market encourages more open monitoring and diagnostic tooling, potentially integrated with inference pipelines and automation. If the incumbent loses exclusivity, innovation in maintenance services may accelerate, including through predictive AI techniques – a virtuous loop for the whole on-premise stack. This isn’t science fiction: it’s what’s already happening in hardware infrastructure, where the multi-vendor approach has driven down costs and improved resilience.

For the engineering crowd that reads AI-RADAR, this story isn’t about GPUs or quantization. It’s about a lesson enterprise IT has long understood but the Large Language Model world is still learning: deployment freedom doesn’t hinge only on hardware or the model itself, but also on who holds the keys to keep it running.