Samsung Files Documents to Build New, $17 Billion Fab in the United States
The earlier rumors that Samsung might build a factory in the United States have proven true. The company is investigating potential sites in New York, Texas, and Arizona, with plans to build a $17 billion facility.
Anandtech reports that the foundry would be online by Q4 2023. That’s a quick ramp, if true. For comparison, Intel’s Fab 42 in Arizona was under construction from 2011 – early 2014, before Intel paused the project. Development restarted in February 2017, and the facility was declared operational in October 2020. It’s common for a new foundry to ramp up production gradually, so Samsung’s declaration that it wants to have the fab operational by late 2023 doesn’t mean the facility would be producing at anything like maximum capacity.
It’s a little unclear if the facility is guaranteed to be in Austin or not. AT’s article opens by saying Samsung has filed documents in three states seeking to build a fab, but then refers to the Austin site repeatedly as a done or nearly-done deal.
Regardless of where it is, the new facility would produce 1800 long-term jobs in semiconductor manufacturing. Samsung hasn’t named the process node it would deploy, but we can make some guesses about it. During its Q3 quarterly earnings report, Samsung announced it had begun its first shipments of mobile 5nm chips. It also noted it had increased HPC (high performance computing) chip production. Samsung’s biggest known customers in that space are IBM and Nvidia. IBM’s Power10 won’t ramp until later this year, and Nvidia has been stuffing every GPU it can manufacture into the OEM and retail channels.
The $17B investment is separate from Samsung’s S2 facility in Austin, which will continue to manufacture 14/11 nm hardware for the foreseeable future. All client foundries, including TSMC and Samsung, maintain a long tail of manufacturing capability on older nodes to support customers who have no need to move to newer process technologies, or only adopt “new” nodes slowly, when costs are low enough and improvements high enough to justify the effort. Not all types of semiconductors benefit from node shrinks, and even those that do benefit may only gain 10-20 percent improved performance or power consumption.
Samsung has previously claimed it wanted to bring 3nm to market by 2022, so it’s likely that the company will be well into deployment by then. This fab, wherever it winds up being built, is likely to be a leading-edge facility. It’s even possible Samsung could intend to use it to deploy a <3nm node. It isn’t unknown for a new foundry to debut a mature node while a new process is still ramping, then to serve as a launch vehicle for said node 4-6 months later. Samsung has been quite aggressive with its overall foundry roadmap as it attempts to catch TSMC.
Samsung has demanded a 100 percent tax abatement from Travis County, worth an estimated $805.5M. It has also demanded a 50 percent tax abatement from the city of Austin and a $252.5M abatement from the nearby Manor school district. Samsung reported a net profit of $32.5B in 2020. No company in the United States has previously been granted a 100 percent tax abatement for two decades and Samsung has not articulated an argument for why it ought to be the first. That doesn’t mean state and federal officials won’t take the deal, however. The United States currently lacks a leading-edge semiconductor manufacturer, a deficiency state and federal governments are both seemingly keen to remedy. TSMC is already planning to build a facility in Arizona, though it looks as if that factory will be smaller than Samsung’s, and not a leading-edge facility.