As the global AI arms race intensifies, vulnerabilities in the semiconductor supply chain have returned to the spotlight. Chip design giant Broadcom (AVGO.US) has publicly warned that it is currently facing supply chain constraints, with capacity limitations at its key manufacturing partner Taiwan Semiconductor Manufacturing (TSM.US) emerging as a critical bottleneck hindering the entire industry's growth. Meanwhile, energy and raw material supply pressures stemming from Middle East conflicts are adding new uncertainties to the global semiconductor sector.
From "Unlimited Capacity" to "Capacity Limits": Broadcom Points to TSMC Bottleneck Natarajan Ramachandran, Product Marketing Director for Broadcom's Physical Layer Products Division, stated on Tuesday, "We are observing that TSMC's capacity has reached its upper limit." He noted that just a few years ago, he would have described TSMC's capacity as "unlimited." Ramachandran added, "While TSMC plans to continue expanding capacity through 2027, by 2026, its capacity has already become a bottleneck, effectively constricting the entire supply chain to some extent." As of the time of reporting, TSMC had not immediately responded to requests for comment.
As the primary foundry for high-end AI chips globally, TSMC acknowledged tight capacity in January of this year, citing that the AI infrastructure construction boom was heavily utilizing its advanced process production lines. The world's largest chipmaker—whose major clients also include NVIDIA (NVDA.US) and Apple (AAPL.US)—indicated at the time that it was working to narrow the supply-demand gap.
Ramachandran further pointed out that supply chain tightness is not confined to chips but has spread to multiple segments of the technology industry chain. "Although there are many suppliers in the industry currently, there is a clear supply gap in the laser device sector," he said, noting that printed circuit boards (PCBs) have also become an "unexpected" bottleneck. Ramachandran indicated that PCB manufacturers in both Taiwan, China and mainland China are facing insufficient capacity, leading to extended product delivery cycles, though he did not disclose specific supplier names.
To secure production stability for the coming years, many customers are being compelled to turn to long-term agreements. According to Ramachandran, numerous clients are now signing capacity assurance contracts with suppliers spanning three to four years. This trend is corroborated by memory chip giant Samsung Electronics (SSNLF.US), which stated last week that it is collaborating with core customers to extend cooperation agreement terms to three to five years. This move reflects both customer demand for long-term supply chain security and suppliers' desire to hedge against market demand volatility.
TSMC's 2nm Capacity Crisis: NVIDIA's Next-Gen AI Chip Platform May Require Design Adjustments TSMC's capacity bottleneck is also evident in its most advanced 2nm process. Reports indicate that TSMC's 2nm process capacity (including the subsequent A16 node) is severely oversubscribed, to the extent that even its largest customer, NVIDIA, may be forced to redesign its next-generation AI chip platform, "Feynman." Originally slated for launch in 2028 as the successor to the current Vera Rubin architecture, any design adjustments due to capacity constraints could impact its performance targets, launch timeline, and cost structure.
Simultaneously, tech giants like Meta (META.US) have joined the competition for 2nm process capacity, further exacerbating the supply-demand imbalance. As a result, order lead times for TSMC's advanced processes now extend beyond 2028. The capacity crunch has also strengthened TSMC's pricing power. Reports suggest that TSMC's advanced processes will see price increases for four consecutive years. This means chip design companies are not only grappling with supply shortages but also facing rising manufacturing costs, which will ultimately be passed on to AI server manufacturers, cloud service providers, and end-user enterprises, further driving up the cost of deploying AI infrastructure.
The Geopolitical "Specter": Middle East Conflict Threatens Energy and Material Supply Lines While the physical limits of manufacturing are being tested, geopolitical clouds are introducing additional variables into the supply chain. As conflicts in the Middle East persist, the global semiconductor industry faces the risk of disruption to key material supplies, with Taiwan, China's high reliance on external energy making it particularly vulnerable. The chip manufacturing industry in Taiwan, China contributes approximately one-fifth of its economic output, but the entire ecosystem is highly dependent on imports. For instance, about one-third of the world's helium is processed in Qatar, while key materials like sulfur are derived from oil and natural gas refining.
More critically, Taiwan, China's energy supply is heavily reliant on the Middle East—97% of its energy needs are met through imports, with 37% of its liquefied natural gas (LNG) sourced from the Middle East. TSMC alone consumes about 10% of Taiwan, China's total electricity generation, and its advanced processes demand extremely high power stability, where millisecond-level fluctuations can render wafers useless.
An analyst team at Goldman Sachs warned that commercial transit through the Strait of Hormuz remains severely disrupted, with Qatar having declared force majeure. For Taiwan, China, the key risks extend beyond oil prices to include the actual availability, pricing, and delivery timelines for natural gas. Although Taiwan, China's economic authorities have stated that LNG supplies for March and April are secured, and noted that local companies can source helium from multiple origins like the US and Australia, the potential for soaring power costs and material shortages in a prolonged conflict remains a Sword of Damocles hanging over TSMC.
Shawn Kim, Head of Asia Technology Research at Morgan Stanley, pointed out that while disruptions in the Strait of Hormuz may not immediately halt chip production, they will impact power costs, material supply, and the economics of building AI infrastructure. He added that companies constructing high-energy-consumption facilities, such as large data centers, may face rising operational costs alongside sluggish revenue.