TSMC
Taiwan Semiconductor Manufacturing Company (TSMC) is the world's largest dedicated integrated circuit foundry, headquartered in Hsinchu, Taiwan. It provides advanced semiconductor wafer fabrication services to clients including leading chip design companies such as Apple and NVIDIA.
Core Facts
Timeline (last 90 days)
If foundries such as TSMC, Samsung, and Intel want to expand AI chip capacity, they must first place orders with ASML.
Taiwan occupies a unique position in the global industrial chain — it has both cutting-edge manufacturing like TSMC and a night market culture full of everyday life.
Maia 300 is equipped with 8 HBM4 memory stacks, uses compute chiplets manufactured on TSMC's N2P process, and is deployed in the form of 96-accelerator racks.
NVIDIA's GPUs are manufactured by TSMC, and advanced process capacity (such as 4N, 4NP) is a scarce resource.
The high-end GPUs, advanced-process chips and their manufacturing equipment that AI training depends on are highly concentrated in a few regions in terms of production capacity; taking NVIDIA GPUs as an example, their advanced processes mainly rely on TSMC's foundry in Taiwan, and lithography equipment is supplied by the Netherlands' ASML.
Maia 100采用台积电5nm工艺制造。
CoWoS (Chip on Wafer on Substrate) is an advanced packaging technology developed by TSMC that achieves higher bandwidth and lower power consumption by integrating multiple chips on a single substrate.
The M4 is manufactured using a second-generation 3nm process (TSMC N3E).
Determining whether there is a bubble in AI requires a comprehensive assessment combining capital expenditure data from multiple cloud vendors, GPU utilization rates, paid conversion rates of AI applications, and the macroeconomic interest rate environment.
OpenAI launched its in-house chip project codenamed Jalapeño in October 2024 and sent the first version of the design to TSMC for tape-out in just 13 months.
27 more timeline events
All Facts (20)
Blackwell (B100/B200 series) was released in 2024, manufactured using TSMC's 4NP process, with a single GPU integrating 208 billion transistors
75%UnverifiedA13 Bionic采用台积电7nm工艺，集成85亿晶体管，包含2个高性能核心、4个能效核心以及8核神经网络引擎
90%Unverified芯片制程进入亚纳米时代（2nm及以下节点），晶体管结构已从FinFET演进到GAA（Gate-All-Around）纳米片架构
90%UnverifiedMicrosoft于2023年11月首次公布自研AI芯片Maia 100，采用台积电5nm工艺，集成超过1050亿晶体管
85%Unverified台积电在亚利桑那州的芯片工厂投资约为400亿美元
85%UnverifiedBetween 2023 and 2024, the delivery lead time for NVIDIA H100s once reached 36–52 weeks; by the second half of 2025, GPU availability improved significantly as TSMC expanded its CoWoS advanced packaging capacity.
60%UnverifiedThe AMD R7-H255 is manufactured using TSMC's 4nm process.
60%UnverifiedA single NVIDIA H100 card is priced as high as $30,000–$40,000, and global annual supply is limited due to constraints on TSMC's advanced process capacity.
60%UnverifiedIf foundries such as TSMC, Samsung, and Intel want to expand AI chip capacity, they must first place orders with ASML.
50%UnverifiedTaiwan occupies a unique position in the global industrial chain — it has both cutting-edge manufacturing like TSMC and a night market culture full of everyday life.
50%UnverifiedMaia 300 is equipped with 8 HBM4 memory stacks, uses compute chiplets manufactured on TSMC's N2P process, and is deployed in the form of 96-accelerator racks.
50%UnverifiedNVIDIA's GPUs are manufactured by TSMC, and advanced process capacity (such as 4N, 4NP) is a scarce resource.
50%UnverifiedThe high-end GPUs, advanced-process chips and their manufacturing equipment that AI training depends on are highly concentrated in a few regions in terms of production capacity; taking NVIDIA GPUs as an example, their advanced processes mainly rely on TSMC's foundry in Taiwan, and lithography equipment is supplied by the Netherlands' ASML.
50%UnverifiedMaia 100采用台积电5nm工艺制造。
50%UnverifiedCoWoS (Chip on Wafer on Substrate) is an advanced packaging technology developed by TSMC that achieves higher bandwidth and lower power consumption by integrating multiple chips on a single substrate.
50%UnverifiedThe M4 is manufactured using a second-generation 3nm process (TSMC N3E).
50%UnverifiedDetermining whether there is a bubble in AI requires a comprehensive assessment combining capital expenditure data from multiple cloud vendors, GPU utilization rates, paid conversion rates of AI applications, and the macroeconomic interest rate environment.
50%UnverifiedResearch interest in TMD-type two-dimensional semiconductors peaked between 2013 and 2015, with major players including TSMC, Samsung, and Intel all having relevant fundamental research programs.
50%UnverifiedGiants including Intel, Broadcom, and TSMC have all committed to volume production of silicon photonics.
50%UnverifiedDemand for high-end GPUs such as the H100 and H200 from top AI labs and tech giants far exceeds the supply capacity of TSMC and NVIDIA.
50%