TSMC has acknowledged that the immense demand from the AI sector has severely hindered their supply chain, claiming that the shortage of AI GPUs from NVIDIA and other companies could last up to 1.5 years.
TSMC Faces Huge Hindrances in "CoWoS Packaging" Capabilities, Taiwan Giant Working on a Steady Supply Chain To Meet Demand For NVIDIA & AMD AI Chips
Speaking at Semicon Taiwan, TSMC's chairman Mark Liu admitted that the business has seen a huge boost in orders from NVIDIA and partners, primarily due to the rise of GenAI development, which requires huge computing capabilities. TSMC reiterated that the bottleneck persists in CoWoS packaging, and the company didn't expect such a huge rise in orders going into 2023.
It is not the shortage of AI chips, it is the shortage of our CoWoS capacity. Currently, we cannot fulfill 100% of our customers' needs, but we try to support about 80%. We think this is a temporary phenomenon. After our expansion of [advanced chip packaging capacity], it should be alleviated in one and a half years
-Mark Liu via Nikkei Asia
TSMC witnessed three times the usual orders, specifically for its CoWoS packaging services, since the process constituents a large share in the production of AI GPUs. However, TSMC has expressed that they are working on upscaling facilities to meet the demand from the AI sector, and it hopes to double down the production by 2024. This won't still be enough for now, as disclosed by the Taiwan giant since the shortage could remain for an extended period.
TSMC's CoWoS packaging facilities are occupied by large orders from NVIDIA for its AI GPUs. While the company is working on expanding facilities, this is a long-term move, and NVIDIA, currently in the spotlight, can't compromise on time for anything. To capitalize on this opportunity, companies like Intel and Samsung are reportedly collaborating with NVIDIA to get their share of the "AI frenzy" by supplying necessary components for producing AI GPUs, such as HBM and GPU packaging.
To justify the above statement, we previously reported that Samsung had acquired a large share of HBM memory orders from NVIDIA, hoping to gain responsibility for 2.5D packaging. Moreover, Intel Foundry expects an upturn in the division, with the company working on improving its packaging facilities due to interest from various "businesses," particularly from Team Green.
Recently, both AMD and NVIDIA have stated that their AI-based revenue could grow to hundreds of billions of dollars but that goal may be a bit too far-fetched given all the supply constraints and the fact that setting up new fabs and plants doesn't happen overnight and takes several years to be formalized.
NVIDIA plans on having an extensive supply chain since it is pushing down the throttle when forecasting sales of its H100s in the upcoming years. Team Green plans to sell 1.5 to 2 million H100s by 2024, which wouldn't be possible by depending on a single supplier; hence, we could soon see a landscape shift in favor of new entrants in the industry.
News Source: Nikkei Asia
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
