TSMC's fab expansion is hitting a skilled construction labour shortage — not a capital or technology wall — as hyperscaler capex races toward $1.3 trillion by 2027. From Google's custom Arm chips to YMTC's sanctioned IPO, today's briefing covers the physical constraints reshaping the AI infrastructure race.
Audio is available on Spreaker — see link below.
The constraint slowing AI infrastructure right now isn't silicon. It's the people who build the rooms where silicon is made.
TSMC has historically built five to six fabs per year. The ambition now is to scale toward twenty to thirty.
Against that backdrop, the hyperscalers aren't waiting. Google has confirmed custom Arm-based server chips targeting data centre deployment in twenty twenty-five.
The scale of that capex is worth sitting with. The big five AI hyperscalers are forecast to spend approximately one-point-three trillion dollars on data centre infrastructure in twenty twenty-seven, up from around eight hundred billion this year.
One more signal worth watching. China's YMTC, which is sanctioned under the US one-two-six-zero-H designation, has had its IPO application formally accepted by the Shanghai stock exchange for first-round regulatory review.
The near-term watchpoints are clear. Can Taiwan's construction labour pool scale fast enough to matter?
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.