The thing is, with the amount of reliability/failure data they'll get from this, learning what fails and what doesn't and why, they'll probably get better reliability in the long term than if they'd gone for low-volume-high-cost reliability from the outset.
