The Great Debate Over the AI Bottleneck: 2027 Peak Constraints vs. 2028 Capacity Unlocking
TREE NEWS reports: As the AI compute arms race accelerates, a critical inflection point is approaching. With tech giants like Broadcom and Nvidia signing massive long-term supply lock-in contracts between late 2025 and early 2026, the semiconductor industry is undergoing its largest capital expansion in history. The true impact of this investment will trigger a fundamental debate over supply-demand dynamics by 2028.
On September 9, tech analysts Ben Bajarin and Jay Goldberg faced off over the AI chip cycle on the industry deep-dive podcast The Circuit. Bajarin argues that 2027 will mark the industry-wide “bottleneck point” of the most extreme supply constraints, and that while a massive wave of new capacity will come online in 2028, it will continue to be absorbed by exponentially growing AI demand, resulting not in a cliff-like oversupply but in a persistently tight supply-demand balance. Goldberg, invoking 50 years of semiconductor cycle patterns, warns that when the massive capacity being built simultaneously by TSMC, Micron, Samsung, and SK Hynix all comes online in 2028, it will trigger a collapse in pricing power and a relentless mean reversion in margins.
Despite their fundamental disagreement over the 2028 endgame, both analysts agree on one clear point: the era of “buying the AI supply chain indiscriminately and reaping explosive profits” is over. As bottlenecks evolve from a single segment into multipolar fragmentation, investors must apply stricter criteria to identify core assets with cross-cycle pricing resilience.
Systemic Constraints: The Shortest-Board Effect Dominates the Ceiling on Chip Shipments
The premise for understanding this debate lies in the fundamental logic of the semiconductor supply chain. The actual shipment volume of the entire system is determined by the single weakest bottleneck. Even if GPU or memory supply is sufficient, if any one of substrates, MLCCs, or wafer testing falls short, total production cannot be increased. This “shortest-board effect” directly explains the current supply chain tensions. Suppliers of substrates, power and analog semiconductors, foundries, and memory have historically adopted conservative investment strategies, refusing to commit to new capacity until they receive clear long-term demand commitments from customers. Only between late 2025 and early 2026, when Broadcom, Nvidia, and major cloud service providers submitted massive and stringent long-term lock-in contracts, did major suppliers seriously begin a new capex cycle. Because the construction of wafer fabs, advanced packaging lines (such as TSMC’s CoWoS), and high-end ABF substrate plants naturally requires 24 to 36 months for construction and cleanroom certification, the actual volume shipments from this investment will not become effective supply until 2028 at the earliest. This timeline is the common starting point for both sides of the debate.
Ben Bajarin: 2027 Supply Peak, 2028 a Gradual Reversion
Ben Bajarin’s core argument rests on the incompressibility of physical construction cycles. He believes that from 2026 to 2027, legacy capacity will be fully locked while new lines are not yet operational, placing global leading-edge process and packaging under the most extreme supply bottleneck in history. Major cloud providers will have no choice but to accept whatever compute allocation is available, and 2027 may become the year of the most severe supply constraints in semiconductor history. Regarding 2028, Bajarin opposes the pessimistic “capacity flood” narrative. He points out that as AI models shift from pure text to high-definition video, real-time voice, 3D generation, and long-range reasoning, token consumption will grow exponentially, and demand-side expansion will be fast enough to quickly absorb newly unlocked capacity. In his forecast, the market will at most exit “absolute shortage” in 2028, returning to a benign state of somewhat tight supply-demand (demand/supply at roughly 105-110%), rather than a cliff-like oversupply. He also suggests that the current supply bottleneck acts as a “safety valve,” delaying overinvestment. He compares this AI infrastructure cycle to the historic railway, canal, 3G, and mobile internet construction booms, noting that all of these past domains experienced severe overbuilding. The key difference in this AI compute cycle is that even with clear demand, the pace of building production facilities and data centers is equally bound by physical constraints, and the supply constraint itself is restraining overinvestment. He further adds that in enterprise AI deployment, high token costs and limited compute access are objectively slowing the transition from experimental environments to production environments.
Jay Goldberg: The Iron Law of Historical Cycles, 2028 Pricing Power May Backfire
Jay Goldberg’s contrarian warning derives from a historical synthesis of 50 years of semiconductor capital cycles. His core judgment is this: in every upcycle, all participants shout “this time is different,” but each time it ends in severe overcapacity. The scale of projects currently under construction is astonishing:
- TSMC is advancing roughly 20 new wafer fabs and packaging sites worldwide.
- Micron is building seven large memory plants around the world.
- Samsung and SK Hynix are equally aggressive with HBM investment.
- Additionally, there is government-subsidized capacity from Intel and others.
Goldberg also highlights the historical pattern of “double/triple ordering.” During chip shortages, customers such as Apple, Qualcomm, and cloud providers often submit duplicate, sometimes double, capacity intentions to multiple suppliers to secure delivery. Once supply begins to unlock and lead times shorten, these inflated “ghost orders” are immediately canceled. On pricing, Goldberg argues that Nvidia’s current ultra-high gross margins of over 70% and Broadcom’s high ASIC premiums are essentially scarcity premiums generated by supply shortages. By 2028, when the massive fab equipment is all operational and fixed-asset depreciation creates enormous cost pressure, foundries and chip designers will inevitably fall into fierce price wars to maintain utilization, and hardware gross margins will face relentless mean reversion.
Investment Consensus: From Indiscriminate Buying to an Era of Selectivity
Despite significant disagreement over the 2028 endgame, both analysts agree on a key market judgment: the window for “buying the AI supply chain indiscriminately for excess returns” has closed. Over the past two years, companies with businesses tied to advanced packaging, HBM, GPUs, and high-speed optical modules saw their share prices greatly benefit from sector-wide premiums driven by a single bottleneck. But from now through 2027-2028, supply chain bottlenecks are evolving from a single segment (such as early CoWoS packaging) to multiple dimensions. As custom compute players like MediaTek and Marvell accelerate their layouts, and massive capacity is sequentially unlocked in 2028, hardware suppliers without irreplaceable software ecosystems or core IP moats will face the dual pressures of margin compression and inventory adjustment from late 2027 into 2028. The shared conclusion of both analysts is this: investors can no longer view the semiconductor supply chain as a homogeneous whole. To secure an advantageous position in the diverging landscape after the largest capacity expansion in history is realized, one must focus on core segments with genuine cross-cycle pricing resilience.
