Samsung runs Tesla's AI5 through Taylor 64 days after tape-out

Sixty-four days separate Samsung's July 13 statement that the Tesla design had reached tape-out from the September 15 report that Taylor is running trial AI5 wafers ^[7][4][15]. Count from the equipment move-in ceremony and the interval is 144 days ^[6][16]. Both are short for a first advanced-node fab on a new site. Samsung says it will finish mass-production verification by year-end ^[3].

The schedule change itself comes from semiconductor industry sources cited by Sedaily, who put the pull-forward down to a sharp rise in AI chip orders from big tech buyers ^[2][1]. The capacity reason sits in the same account: the 2 to 4 nanometer lines at Hwaseong and Pyeongtaek are running full ^[8]. An order won in Korea this quarter has nowhere in Korea to be built.

TrendForce's second-quarter numbers put Samsung at 5.9 percent of the global foundry market against SMIC's 5.4 ^[13], which is SMIC at 91.5 percent of Samsung's share ^[17]. The report describes Samsung as the world's second-largest foundry and SMIC as the fast riser behind it on domestic Chinese demand ^[19]. So the share figures on the table describe a contest for second place. The opening at TSMC, whose advanced-node capacity the report says is largely used up, is attributed to analysts as a capacity judgement ^[12].

The $16.5 billion Tesla supply contract was signed last year ^[9]. The report does not say over how many years it runs, so the annual revenue Taylor is being built against cannot be worked out from it ^[22]. What is specified is breadth: AI5 is meant for the Model 3 and Model Y, the Cybercab, Optimus and AI data centers ^[10], as Tesla expands into autonomous vehicles, data centers and humanoid robots ^[21].

In my view this is a utilization decision. Equipment is in, a customer is contracted, Korean advanced-node lines have no room, and the second Taylor plant is supposed to break ground before December on a 1.4-nanometer process for around 2030 ^[11]. Samsung also intends Taylor to be its core US base for processes below 2 nanometers ^[20]. The counter-thesis is that trial wafers signal confidence and nothing else until yield verification closes, and Samsung has put that at year-end ^[3]. If it slips, next year's full-scale supply date slips with it.

The share test runs separately. SMIC above 5.9 percent in the third-quarter table would mean Samsung lost second place in the same quarter it was starting Taylor up ^[13].