CoWoS
Chip-on-Wafer-on-Substrate—TSMC's advanced 2.5D packaging technology used to integrate GPUs and high-bandwidth memory.
As defined in lex_jensen-huang
Where it comes up · 4
- Dwarkesh Patel Podcast - Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute2 claims
- Dwarkesh Patel Podcast - Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat2 claims
- Andrew Homan & Chris Miller - Redefining Semiconductor Progress - [Invest Like the Best, EP.396]2 claims
- @Asianometry & Dylan Patel — How the semiconductor industry actually works1 claim
Most important claims · 7 of 7Search all claims for “CoWoS” →
- As Moore's Law benefits get harder to capture on the front end, value is shifting to back-end advanced packaging, which historically was ignored, and CoWoS packaging was a bottleneck limiting Nvidia GPU output.Andrew Homan & Chris Miller - Redefining Semiconductor Progress - [Invest Like the Best, EP.396]
- As scaling continues, the bottleneck shifts from short-lead-time items (CoWoS, power, data centers) back to long-lead semiconductor supply chains, and ultimately to ASML, which makes the world's most complicated machine: an EUV tool selling for $300-400M, of which only ~70 can be made now, ~80 next year, and barely over 100 by decade's end even under aggressive expansion.Dwarkesh Patel Podcast - Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute
- Going directly after NVIDIA's GPU is a challenging task, so the better opportunities lie in the surrounding bottlenecks of interconnect, memory, power, and packaging, because the data center model of five years ago is breaking.Andrew Homan & Chris Miller - Redefining Semiconductor Progress - [Invest Like the Best, EP.396]
- In rapidly growing AI infrastructure, the 'one bottleneck' narrative is misleading because the moment you unlock one constraint, the next becomes binding - from CoWoS in 2023, to HBM and CoWoS-L, to data centers, transformers, substations, power generation, batteries, UPSs, and water cooling - so the picture is permanently cloudy with shifting limiters rather than a single chokepoint.@Asianometry & Dylan Patel — How the semiconductor industry actually works
- Putting more dies per package (Blackwell's two, Rubin Ultra's four, Tesla's wafer-scale Dojo with 25, Huawei Ascend's progression) is a path to higher in-package bandwidth that works on older nodes too; Huawei leans on packaging scaling because it can't shrink process, and anything done on 7nm packaging can also be done on 3nm.Dwarkesh Patel Podcast - Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute
- No supply-chain bottleneck (CoWoS, logic fabs, HBM, EUV machines) lasts longer than two to three years once a clear demand signal exists, because once you can build one you can build ten and then a million; the only hard, durable constraints are downstream—energy and skilled trades like plumbers and electricians.Dwarkesh Patel Podcast - Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat
- Nvidia's suppliers make large upstream investments specifically for Nvidia rather than competitors because they trust Nvidia has the downstream demand and reach to buy their supply and sell it through; without that demand reach, no one builds a supply chain for an architecture whose business churns are low.Dwarkesh Patel Podcast - Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat