NVIDIA discloses it is now using Samsung for chip production

Last month there was a rumor that NVIDIA planned to use Samsung's 14nm FinFET process for mobile Tegra chip production and it seems there was quite a lot of truth to this rumor. NVIDIA's recently published Form 10-K annual report filing with the SEC confirms the company is now using TSMC and Samsung for chip production:

We utilize industry-leading suppliers, such as Taiwan Semiconductor Manufacturing Company Limited and Samsung Electronics Co. Ltd, to produce our semiconductor wafers. We then utilize independent subcontractors, such as Advanced Semiconductor Engineering, Inc., BYD Auto Co. Ltd., Hon Hai Precision Industry Co., Ltd., JSI Logistics Ltd., King Yuan Electronics Co., Ltd. and Siliconware Precision Industries Company Ltd. to perform assembly, testing and packaging of most of our products. We purchase substrates from IbidenCo., Ltd., Nanya Technology Corporation, and Unimicron Technology Corporation.

RBC Capital Markets chip analyst Doug Freedman responded to the news by suggesting NVIDIA could shift wafer supply quicker than other semiconductor designers because they have fewer products and sell little to industrial customers (who may have strict qualifications).

Freedman believes NVIDIA could also become a customer of Intel's custom foundry business, and suggests that besides the diversification advantage of having multiple foundry options, NVIDIA may be pursuing this strategy to safeguard further patent licensing revenue from Samsung and Intel:

Should NVDA fully move their foundry relationship to Samsung and/or INTC, NVDA would be opening up a $1.7bil foundry revenue stream to suppliers from whom they are seeking royalty revenues. NVDA could be set up in a powerful position as they could dangle the $1.7bil foundry “carrot on a stick” to increase the likelihood of collecting future royalty revenue streams from Samsung and/or Intel. We calculate that 80% of COGS from NVDA in 2016 may be foundry-related costs, or $1.7bil/year using FY2016 COGS in our model. We model zero royalties in F2Q18, approximately 2 years from now, per the present terms of the Intel license. Each $100mil/y increase in royalty revenue would add $0.18 or 11% in FCF.

In slightly related news, rumors are circling the web again about a potential buyout of AMD. This time South Korean media claims Samsung is interested in acquiring AMD to better compete with Intel and Qualcomm. The chances of this panning out seem quite low so unless more credible information comes in we're not going to make a dedicated news post about this.