Nvidia Reshapes the Semiconductor Supply Chain in Its Own Image

Nvidia resides at the cutting edge of chip technology, with its AI graphics processing units (GPUs) pushing the envelope of chip design, manufacturing, and packaging. However, the semiconductor supply chain hasn’t kept pace with Nvidia’s rapid advancements, leading to shortages and soaring prices for its data center AI GPUs.

To help its suppliers stay current, Nvidia is bringing its own technology to different supply chain nodes. At Nvidia’s annual GTC event in mid-March, the company outlined how it is working with partners, including TSMC, Cadence, and Synopsys, to leverage generative AI to develop new GPUs, advancing the company’s initiative to democratize the technology’s benefits.

New Innovations for The Entire Supply Chain

These efforts stand to benefit the electronics supply chain considerably. In his GTC keynote address, Nvidia CEO Jensen Huang noted that his company has already been able to exceed the trajectory of Moore’s Law.

The company’s newest flagship GPU for data center AI acceleration, the Blackwell, illustrates this rapid progress. Compared to its predecessor, the H100, the A100 sports 128 billion more transistors, four times the memory, and a fivefold increase in performance.

By leveraging generative AI, Nvidia is helping its suppliers to enable further advances. For an electronics supply chain stressed by the demand for sophisticated AI accelerators, these advances could be game-changing.

At GTC, Huang discussed a partnership with leading foundry TSMC and EDA vendor Synopsys to use generative AI in computational lithography. GPUs can accelerate inverse lithography by a factor of 40, reducing time, cost, and uncertainty in semiconductor manufacturing.

The technology also promises to accelerate the pace of semiconductor scaling, allowing chip manufacturing to keep pace with innovations in GPU design.

Nvidia also announced that EDA firm Cadence has integrated its digital twin platform with Nvidia’s Omniverse platform to accelerate data center design and simulation workflows by 30 times. Omniverse provides a virtual reality interface that can be used to create and link digital twins.

In another development that has major ramifications for the supply chain, Nvidia announced the Blackwell will employ a chiplet approach, with two separate GPU dice combined in a single package. The Blackwell represents Nvidia’s first chiplet design and follows the lead of AMD and Intel, which have employed this technology for their latest AI accelerators.

The packaging industry is gearing up to accommodate this rising demand, as noted by Commodity IQ.

How the Packaging Industry is Keeping Pace

The increasing sales of advanced packaging are coming at a time when costs are increasing for polyethylene (PE) and polypropylene (PP) used to produce plastic resins for chip containers.PE producers plan to hike PE prices by at least $0.03 per pound in March, while PP suppliers aim to boost contract prices by $0.02 to $0.03 per pound.

For buyers of cutting-edge semiconductors like AI GPUs and microprocessors, Nvidia’s moves to accelerate chip development using generative AI will lead to faster development, improved availability and more affordable prices for high-demand chips. With Nvidia’s H100 now selling in the $30,000 range, such price relief would represent a welcome change.

Meanwhile, the company’s embrace of a chiplet design will continue to place pressure on pricing for advanced packaging as strong demand fuels rising prices.