What is TSMC?

Analog in-memory compute achieves 150 TOPS/W at 8-bit precision in 16nm, vs ~5 TOPS/W for best digital matrix-multiply in the same node, a 30x efficiency advantage.

“We're basically doing 8bit compute at 150 tops per watt in a 16 nanometer technology. So just as a point of reference the best digital matrix multiplies will give you sort of like five tops per watt in that technology. is 30x better at the level of that core technology.”

TSMC alone consumes 10% of Taiwan's entire electricity grid.

“TSNC alone takes up 10% of the grid.”

Etched. A startup building transformer-specific ASICs targeting inference. Claims 10x concurrency vs. NVIDIA, sub-40-day silicon bring-up, and a novel low-voltage process. Raised $800M and secured early TSMC cooperation before its Series A.

“Our hardware is going to generally be able to get you an order of magnitude more concurrency at a given level of interactivity.”

AI chip startups cannot use early TSMC process access as a moat because Nvidia already knows new processes better than any small entrant can.

“I promise you that Jensen saw that process when it was a twinkle in Taiwan Simmy's eyes and it they know more about it than this little company with 200 people can imagine.”

TSMC resolved the challenge of cross-utilizing N7 and N5 node capacity across fabs by massively scaling its trucking service to move work between facilities.

“The solution TSMC figured out to cross-utilize the N7 and 5 nodes is by massively expanding the trucking service.”

Eiso Kant on TSMC holding more effective power over AI compute than the US government.

“I think the one entity that has more power than the US government here because like basically whoever gets the allocations gets the compute. You can take it down to TSMC.”

eFlash remains resilient in embedded memory despite TSMC and Samsung actively pushing STT-MRAM and ReRAM as its successors.

“Despite both TSMC and Samsung positioning STT-MRAM or ReRAM, eFlash remains quite resilient.”

Brilliance, a Chinese minivan maker, raised $70 million in a 1992 NYSE/NASDAQ IPO, making it the first Chinese company to list on a major US exchange, five years before TSMC.

“They raised 70 million US dollar like 5 years before TSMC.”

TSMC found a way to use older N7 capacity to support newer N5 production, contradicting earlier indications that such cross-node utilization would be very difficult.

“They have figured out a way to use N7 capacity to support N5.”

Gavin Baker on Nvidia's unconstrained GPU demand potential

“If Taiwan Semi did what Jensen wanted, I think Nvidia could sell two trillion dollars of GPUs in 26 in 26 or 27, maybe two.5 trillion, maybe three trillion.”