May 1, 2026
As AI computing power experiences explosive growth, chip areas continuously challenge physical limits, and AI Agents reconstruct the overall computing architecture, the resonance of these three mega-trends is pushing the PCB and IC substrate industry into an unprecedented "super shortage cycle." The entire supply chain is undergoing a comprehensive upgrade from traditional consumer electronics accessories to core components of AI computing infrastructure. Recent in-depth reports clearly indicate that from 2026 to 2028, the entire industry chain will face persistent capacity crowding and material shortage crises. Upstream high-frequency materials and ABF substrates will be the absolute core of this "volume and price surge" dividend.
The foundational driver of this PCB industry boom comes from the unreserved capital expenditures (CapEx) of North American Cloud Service Providers (CSPs). Statistics show that North American CSP CapEx grew by a staggering 61% year-over-year in 2025, and forecasts for 2026 have been drastically revised upward to 90% (with Amazon's growth rate hitting 77%).
While global server shipments are climbing steadily, the more formidable factor is the exponential growth in “value per board” and “chip area”, which has become the root cause of the across‑the‑board emergency in high‑end PCBs and substrates.
Starting in the first half of 2026, the supply and demand for upstream PCB materials have become extremely tight, with capacities booked up by tech giants and clear upward pricing trends. The following four materials have become super bottlenecks in the supply chain.
As the lifeblood of AI servers and 800G/1.6T switches, HVLP4 copper foil faces severe yield challenges, resulting in extreme shortages of effective capacity. Monthly demand in 2026 is estimated at 2,600‑2,700 tons, but effective capacity is only 1,000‑1,100 tons – a massive 48% shortfall, with the gap projected to remain at 43% into 2027. Several major manufacturers fired the first shot on price hikes in Q4 2025, and multiple price adjustments are expected throughout this cycle.
AI servers utilize high‑end laminates with increased board thickness and extremely hard fiberglass, causing a catastrophic plunge in drill bit life (dropping to just 1/6th of traditional FR4 levels when drilling advanced materials). This has led to a severe shortage of high‑end coated drill bits, with quotes soaring 20% to 50% higher than traditional uncoated bits and gross margins hitting 40% to 50%. This shortage is expected to persist into 2027.
Low Dk (Generation 2) and T‑Glass have become the most out‑of‑stock items on the market. While demand for Gen 1 Low Dk is 15 million meters per month, Gen 2 demand has reached 3.5 million meters, yet yarn manufacturers’ capacity expansions are severely lagging. Specifically, T‑Glass – 70% of which is consumed by ABF substrates – will remain in short supply through 2027. Price hikes across various high‑end glass fibers in 2026 are estimated to hit 20% to 40%.
Designed exclusively for top‑tier architectures like Nvidia Rubin and LPUs. With the volume ramp‑up of LPX racks, Q‑Glass demand entered a frantic stockpiling phase in 2026. The supply‑demand gap is projected to exceed 60% in 2027, making it the most scarce strategic material currently available.
IC substrates are the absolute centrepiece of this cycle, driven fundamentally by the runaway scaling of chip areas.
This directly resulted in global ABF substrate capacity being devoured instantly. While the ABF market was still oversupplied in 2025, it has completely flipped to undersupply in 2026. The shortage gap is estimated to reach 26% in 2027 and widen to 46% by 2028. Western chip giants are frantically reserving capacity, leaving the ASIC ecosystem facing a severe double squeeze of “scrambling for materials and scrambling for capacity.”
The explosion of AI Agents in 2026 is altering the monolithic structure that previously relied solely on GPU compute. Traditional LLM training is GPU‑dominant, but under the AI Agent architecture, CPUs must handle massive amounts of scheduling, coordination, and tool orchestration. This causes CPU latency to account for up to 90% of processing time, creating a new system bottleneck.
To resolve this, Intel and AMD have drastically increased server CPU core counts (Intel Oakstream hitting 144‑288 cores, AMD Venice reaching 256 cores). CPU packaging areas are ballooning simultaneously, forming a “GPU + CPU dual‑engine drive” that is further draining the market’s remaining ABF substrate capacity.
PCB fabricators historically reliant on consumer electronics are temporarily falling behind in this AI wave; AI servers demand ultra‑high reliability High Layer Count (HLC) boards (40‑60 layers).
Meanwhile, to mitigate geopolitical risks, Southeast Asia (particularly Thailand and Vietnam) has become the absolute core for high‑end board capacity expansion, prioritising the demands of major North American AI clients. Furthermore, the explosion of 800G/1.6T optical communication switches (with 800G demand projected at 50 million ports and 1.6T at 20 million ports in 2026) is forcefully driving the capacity consumption of 38‑48 layer high‑end PCBs.
Facing this three‑year super‑shortage cycle, has your next NPI (New Product Introduction) project already secured high‑end laminates and ABF capacity in advance? Do not let the supply chain material crisis be deliberately downplayed, ultimately turning your initiative into a “Watermelon Project” – looking green and on‑track on the outside, but red and full of hidden supply risks and unexploded bombs on the inside.
When “having the budget still doesn’t guarantee materials” becomes the industry norm, how will your team break through using upfront stack‑up optimisation and material contingency planning? Drop a comment below to share your battle strategies, or forward this article to your procurement and hardware R&D teams to initiate a risk assessment immediately!
