Rumours online suggest that NVIDIA turn to Intel Foundry Services to up its production capacities of its H100 chips. As to why this may be the case, it is possible that the GPU brand’s long-time fab supplier, TSMC, may have trouble keeping up with the brand’s growing business appetite.
Allegedly, TSMC is struggling to meet NVIDIA’s request to increase its production capacities for its H100 GPUs by an additional 5,000 wafer units per month. That would total to 300,000 chips for the H100 GPU alone, assuming that the contract is solely for said enterprise-grade AI accelerators.
The issue here is that, unlike TSMC, Intel Foundry Services doesn’t offer the same chip-on-wafer-on-substrate (CoWoS) packaging method; its version for chip packaging technology is known as Foveros. Because of this difference, NVIDIA may have to test and validate the blue chipmaker’s packaging process, and that in turn means that future GPUs assembled with the new chips could sport different characteristics than the ones made by the Taiwan-based fab.
Even if the rumour turned out to be real, analysts predict that NVIDIA will continue to rely on TSMC as its main supplier of chips for the foreseeable future. After all, the fab already has current and upcoming facilities in Arizona, as part of the deal afforded by the US CHIPS deal signed back in 2022.
