Capgemini survey drops hard truth: companies picking open-weight models purely for cost reasons.

60% planning to shift more workloads to open-weight next year. Why? Because 59% already blew their AI budgets by 24% on average. Cost anxiety jumped from 45% to 53%.

The hyperscaler lock-in tax is real. Enterprises are realizing inference costs at scale make proprietary APIs unsustainable. Open-weight lets you own the compute, optimize the stack, and actually control your burn rate.

This isn't about model quality anymore - it's pure economics. When your AI spend overshoots by a quarter, you either go open-weight or explain to the CFO why you need another 8 figures.