Developers can pay with various schemes to receive chips from their contract manufacturers. In some cases, the customer pays for an entire silicon wafer, regardless of the number of defective chips. In other cases, the customer pays only for suitable chip. The first option is more profitable to the contractor, the second to the customer.
Assocaite reports that NVIDIA has moved to pay for the full cost of silicon wafers, which are produced at its request by TSMC. As NVIDIA recognizes , in the case of sufficiently high level of suitable products , it may significantly increase its production costs and reduce the deliveries volume of finished products. Perhaps the transition to such payment scheme explains the relatively high cost of GeForce GTX 680, and their limited availability.
Obviously, if such relationships model between NVIDIA and TSMC will become the norm, the financial results of the first companies may deteriorate and the situation with the pace of migration to the new processes will strongly depend on the product level.