A failed accelerator can become a financial problem long before its replacement arrives. The hardware may remain under warranty, yet the customer could still face lengthy service delays, limited diagnostic access or an expensive replacement process. For AI infrastructure buyers, this raises an uncomfortable question about what procurement teams actually measure before approving major hardware investments. Performance benchmarks, energy efficiency and acquisition costs receive considerable attention during commercial evaluations. Repair access may receive less scrutiny in procurement evaluations that prioritise performance and acquisition costs, despite its potential influence on operating expenditure and service continuity. The industry has become skilled at comparing computational capability while leaving practical maintenance requirements buried in technical documentation. That imbalance deserves closer examination as organisations commit larger budgets to increasingly complex AI systems. A repairability score could give buyers another way to distinguish hardware that delivers strong performance from equipment that remains economically manageable throughout its operating life.