Platform lock-in

Platform dependence can support retention and bargaining power when the cost of rebuilding its surroundings is high. It does not replace product quality: customers may value the core platform and still dislike the exit burden. Gartner’s analyst Michael Warrilow told The Register that companies running the full VMware stack should treat it as a networking vendor first, a storage supplier second, a management-tools provider third and a source of virtualization only after that, because moving networks, storage and management tools is harder than moving the hypervisor. A buyer who thinks of the purchase as one product underestimates what it depends on.

The same web protects the provider from competitors that are better at the core product. A rival must offer a superior platform and also persuade developers, partners and staff to start again, and none of them will move until the others do. That is the reasoning in the Microsoft ruling: the barrier came from how many programs had been written, not from any feature of Windows itself.

It also cuts against the provider when it is used to raise prices. A supplier can extract more from a dependent ecosystem only as long as the ecosystem cannot coordinate an exit, and each increase gives customers, partners and regulators a reason to fund one. Lock-in combines useful fit with dependence. A supplier that uses the exit burden to raise prices can consume customer trust and encourage customers, partners and regulators to fund a way out.

The practical test is to map dependencies as an outsider would. Which applications, partners, certifications and skills would have to change, who owns each, and could any run on two platforms at once? A buyer who cannot answer is exposed to price changes. A provider who cannot answer does not know whether customers are staying because the platform earns it.

The VMware case shows the damage of pricing against the web. AT&T sued, saying Broadcom had not honored its support contract, and by November 2024 the two had reached a settlement in principle on undisclosed terms. Meanwhile CISPE, the trade body for European cloud providers, urged the European Commission in March 2024 to consider action. Broadcom replied that it works with more than 140 European cloud providers and welcomed dialogue.

Exit is still slow, which is what makes the dependence real. Gartner modeled a migration of at least 2,000 virtual machines and 100 servers at 18 to 48 months, and $300 to $3,000 per machine when outside providers are used. Its 2026 Magic Quadrant predicts that 55% of enterprises will start proofs of concept for VMware alternatives by 2029, up from 25% in 2026, while The Register reports that Broadcom is expected to keep most of its customers for at least three more years. The lock-in is loosening, but on a schedule set by the customers’ migration plans, and not by their complaints.

The Microsoft ruling names the mechanism that ends it. The court said that if middleware succeeded, so that applications written for multiple operating systems could run on any of them, the applications barrier would erode. Lock-in weakens when a layer arrives that makes the platform’s complements portable.

Regulators can build that layer by law. The EU Data Act has applied since September 12, 2025, requires cloud and edge providers to remove obstacles to switching and to support interoperability, and bars all switching charges, including data egress fees, from January 12, 2027. Nothing in the evidence shows that it removes the dependence on applications and skills, but it prices exit more predictably.