The Debate on AI Bottlenecks: 2027 Peak Bottleneck vs. 2028 Capacity Expansion and Price Collapse

The AI computing power arms race is approaching a critical turning point.

With tech giants like Broadcom and Nvidia signing large-scale, long-term contract agreements with their supply chains in late 2025 and early 2026, the semiconductor industry is undergoing the largest round of capital expansion in its history. The actual implementation of this investment will ignite a fundamental debate about the supply and demand dynamics in 2028.

On September 9th, tech analysts Ben Bajarin and Jay Goldberg engaged in a head-to-head debate on the AI chip cycle on the in-depth industry podcast "The Circuit".

Ben Bajarin believes that 2027 will be the most extreme "bottleneck" in terms of supply constraints for the entire industry. Although new production capacity will be released on a large scale in 2028, the continuously expanding demand for AI will keep supply and demand in a slightly tight state, and there will be no cliff-like oversupply.

Jay Goldberg, citing the 50-year cycle of the semiconductor industry, issued a converse warning: the massive production capacity that TSMC, Micron, Samsung, and SK Hynix are simultaneously advancing will trigger a collapse in pricing power and a brutal shakeout due to the reversion of gross margins to the mean once it is released in 2028.

Despite their fundamental disagreement on the 2028 target, the two reached a clear consensus: the era of "blindly buying AI supply chains and seeing explosive growth" is over. As bottlenecks evolve from single links to multi-polarization, investors must use more stringent standards to identify core assets that truly possess cross-cycle pricing resilience.

Systemic constraints: The weakest link effect dominates the upper limit of chip shipments

Understanding this debate hinges on a fundamental logic of the semiconductor supply chain: the actual shipment volume of the entire system is determined by the weakest single bottleneck .

Even if the supply of GPUs or memory is sufficient in a certain stage, if there is a shortage in any stage such as substrate, MLCC, or wafer testing, the total output cannot be increased.

This "weakest link effect" directly explains the current tight supply chain situation. Substrate, power and analog semiconductor, wafer foundry, and memory suppliers have historically adopted a conservative investment strategy, and will not rashly start new greenfield capacity without obtaining clear endorsement from customers' long-term demand.

It wasn't until late 2025 and early 2026, after Broadcom, Nvidia, and major cloud service providers submitted substantial and stringent long-term contract lock-in agreements to the supply chain, that major suppliers fully launched a new round of capital expenditures.

Since the construction and cleanroom certification of a wafer fab, advanced packaging line (such as TSMC CoWoS) or high-end ABF substrate fab naturally take 24 to 36 months, the actual mass production of these investments will not be available until 2028 at the earliest.

This timeline marks the common starting point of the dispute between the two sides.

Ben Bajarin: Supply will peak in 2027, with a moderate recovery in 2028.

Ben Bajarin's core argument rests on the incompressibility of the physical construction cycle.

He believes that between 2026 and 2027, old production capacity will be fully locked up and new production lines will not be put into operation, and the world's advanced process and advanced packaging will be in the most extreme period of supply bottleneck in history.

Major cloud providers will then have no choice but to accept any available computing power quotas – 2027 may become the year with the most severe supply constraints in semiconductor history.

Ben Bajarin does not agree with the pessimistic narrative of a "flood of production capacity" for 2028.

He pointed out that as AI models shift from plain text to high-definition video, real-time audio, 3D generation, and long-range inference, token consumption is growing exponentially, and the expansion of demand is sufficient to quickly absorb the newly released production capacity.

In his prediction, by 2028 the market will at most emerge from "absolute shortage" and return to a healthy state of slightly tight supply and demand (demand/supply ratio of about 105% to 110%), without experiencing a cliff-like oversupply.

He also suggested that the current supply bottlenecks act as a "safety valve" in a sense, delaying the occurrence of over-investment.

He compared this AI infrastructure cycle to the historical booms in railway, canal, 3G, and mobile internet construction, pointing out that these areas all experienced serious over-construction. The key difference in this round of AI computing power is that even if the demand is clear, the construction speed of production facilities and data centers is also physically limited, and supply constraints themselves are restricting over-investment.

He added that in enterprise-level AI deployments, the high cost of tokens and limited access to computing power have objectively slowed down the migration process from experimental to production environments.

Jay Goldberg: The Iron Law of Historical Cycles – Pricing Power May Backfire in 2028

Jay Goldberg's contrarian warning stems from a historical analysis of the semiconductor industry's fifty-year capital cycles. His core judgment is that in every upward cycle, all participants shout "This time is different," but each time it ends in severe overcapacity.

The scale of projects currently under construction is alarming:

TSMC is advancing nearly 20 new wafer fabs and packaging bases globally;Micron has seven large-scale storage plants under construction globally;Samsung and SK Hynix have both been aggressive in their investments in the HBM sector;In addition, there is the production capacity of manufacturers such as Intel, which has been accelerated by government subsidies.

Jay Goldberg also specifically pointed out the historical pattern of "double/triple bookings".

During periods of chip shortages, customers such as Apple, Qualcomm, and cloud providers often submit overlapping or even doubled production capacity requests to multiple suppliers simultaneously to ensure they receive their orders. Once supply begins to increase and delivery times shorten, these inflated "phantom orders" are instantly canceled.

In terms of pricing, Jay Goldberg believes that Nvidia's current gross margin of over 70% and Broadcom's high ASIC premium are essentially scarcity premiums created by supply shortages.

By 2028, when a large number of wafer fabs are fully operational and fixed asset depreciation creates enormous cost pressure, foundries and chip designers will inevitably engage in fierce price wars to maintain equipment utilization rates. At that time, hardware gross margins will face a brutal mean reversion.

Investment consensus: Selective buying replaces the logic of broad-based price increases.

Despite their significant differences in their assessment of the end result in 2028, they agree on one key market judgment: the window for "blindly buying AI supply chains to obtain excess returns" has closed.

Over the past two years, any company whose business is even remotely related to advanced packaging, HBM, GPU, or high-speed optical modules has seen its stock price significantly benefit from the sector premium brought about by a single bottleneck.

However, from now until 2027-2028, supply chain bottlenecks are evolving from single links (such as the early CoWoS packaging) to multiple dimensions.

With MediaTek, Marvell, and other custom computing players accelerating their deployments, and large-scale production capacity being released in succession in 2028, hardware suppliers lacking irreplaceable software ecosystems or core IP moats will face the dual pressures of compressed gross margins and inventory reduction from the end of 2027 to 2028.

The two analysts concluded that investors can no longer view the semiconductor supply chain as a homogeneous whole. They must focus on the core links that truly possess cross-cycle pricing resilience in order to gain a favorable position in the differentiated landscape following the implementation of the largest capacity expansion in history.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.