The explosive growth in demand for AI chips is pushing the global semiconductor supply chain to a critical point. An executive from Broadcom has warned that Taiwan Semiconductor Manufacturing's production capacity has become a key bottleneck in the supply chain, with shortage pressures now spreading from the chips themselves to upstream and downstream components such as laser devices and printed circuit boards.
Natarajan Ramachandran, Product Marketing Director for Broadcom's Physical Layer Products Division, stated to media on Tuesday that TSMC's capacity has "reached its limits." Just a few years ago, he would have described TSMC's capacity as "virtually unlimited." He anticipates that TSMC will continue its capacity expansion efforts through 2027.
The effects of the supply crunch are being felt in the market. To mitigate uncertainty, a growing number of customers are entering into long-term agreements lasting three to four years with suppliers to secure production capacity.
Samsung Electronics confirmed last week that it is also negotiating with major clients to extend contract cycles to three to five years. This trend reflects a widespread anxiety across the entire industry chain regarding long-term supply security.
TSMC's capacity is under severe strain, as the AI construction boom consumes the remaining advanced process availability. As the primary foundry for advanced AI chips globally, TSMC's clientele includes tech giants like NVIDIA, Apple, and Broadcom. According to a Reuters report, TSMC publicly acknowledged tight capacity in January of this year and stated it was working to close the supply-demand gap. The massive wave of AI infrastructure development has largely exhausted the remaining space on its advanced process production lines.
Ramachandran was candid about this situation. He indicated that while TSMC's capacity expansion plans will extend through 2027, the capacity shortfall will become a substantive constraint on the supply chain by 2026, before the expansion is complete. This implies that chip designers and system integrators will face a sustained period of pressure during the current intense phase of the AI computing arms race.
The bottleneck is spreading, with laser devices and PCBs becoming unexpected weak links. Beyond chips, supply chain pressures are extending into adjacent sectors. Ramachandran pointed out that significant supply constraints also exist in the laser device field – despite multiple suppliers being present in the industry, overall capacity remains insufficient.
The situation with printed circuit boards is particularly acute. He described the PCB shortage as an "unexpected" bottleneck, citing PCBs used in optical modules as an example: delivery lead times for these products have surged from approximately six weeks to six months. PCB suppliers in both Taiwan and mainland China are facing capacity saturation pressures.
In response to the supply tightness, players across the industry chain are adjusting their procurement strategies. Ramachandran noted that a significant number of customers are now turning to three-to-four-year long-term agreements with suppliers in exchange for stable capacity commitments.
The recent disclosure from Samsung Electronics corroborates this trend. The memory chip giant stated it is discussing extending contract periods to three-to-five years with key customers. This move aims to provide customers with longer-term supply security while also helping suppliers mitigate the impact of demand fluctuations.
Regarding the industry's outlook, Ramachandran expressed cautious optimism. He stated that the entry of new players and the continued expansion of existing capacity are expected to gradually alleviate supply pressures over time. However, in the short term, the tight supply chain situation is forecast to persist through 2026, continuously testing the delivery capabilities of all parties across multiple critical segments.
Comments