Semiconductor Supply Chains Under Pressure: Why “Capacity” and “Licenses” Are Now Competing Like Nodes

A new kind of scarcity: when capacity is plentiful but licensable output is not

A modern chip shortage does not always look like empty shelves. It can look like full fabs that still cannot ship, full tool inventories that still cannot legally move, and production plans that still stall on “due diligence” and licensing timelines rather than on physics. That shift is increasingly visible in 2025 to 2026 policy moves from export-control authorities and in parallel, high-cost capacity expansions by foundries and memory makers.

In the United States, the government has also explicitly framed the problem as a national-security issue tied to insufficient domestic capacity for both semiconductors and semiconductor manufacturing equipment relative to demand. In January 2026, the White House described an investigation under Section 232 that concluded U.S. capacity “is insufficient to meet domestic demand,” and then issued a presidential action adjusting import treatment for semiconductors and semiconductor-manufacturing equipment and derivative products. (whitehouse.gov)

This is not a purely American story. The European Chips Act has been explicitly built to address reliance on external suppliers for both manufacturing capacity and next-generation technology, and the Commission has reported that by 2025 its Chips Act measures had already catalyzed large investment levels for manufacturing capacity. (digital-strategy.ec.europa.eu) But even where money is flowing, licensing and controlled-tool pathways can reshape who gets throughput and when.

At the same time, the market is wrestling with a second pressure point: memory and advanced packaging constraints. Storage and compute are tightly coupled in AI systems, and when NAND and packaging tightness hits, it reverberates upstream into orders for wafers, packaging slots, and logistics. TrendForce has publicly described NAND pricing and the supply-demand response via production behavior, while other reporting has tied ongoing AI demand to the persistence of “sold out” NAND supply expectations through multi-year windows. (trendforce.com; tomshardware.com)

From fab capacity to “licensable throughput”: how export controls reshape sequencing

The semiconductor supply chain has always been a sequencing problem—materials, deposition, lithography, metrology, test, then logistics—but export controls add a second, legally deterministic sequence that can be orthogonal to factory throughput. The key operational shift is that “being able to make” no longer guarantees “being able to ship the specific configuration, into the specific end-market, via the approved parties, within the needed date window.”

In practice, export controls work like a parallel constraint system with its own lead times and failure modes:

• Classification and re-classification lead times: A shipment’s eligibility can change when there are updates to control parameters, end-use assertions, or tool/part characterization requirements. Even if a fab’s yield is stable, the legal classification process can bottleneck output release.
• End-use / end-user diligence as a scheduling gate: BIS’s due-diligence emphasis on advanced computing integrated circuits effectively turns “compliance readiness” into an operational precondition—so order timing becomes dependent on document completeness, customer verification, and the responsiveness of downstream counterparties.
• Use of approved intermediaries (including OSAT) as a throughput filter: When compliance structures rely on “approved” outsourced assembly and test services, output routing becomes part of the production plan. A fab can complete wafers, but if the OSAT path is not cleared for the specific controlled characteristics, the downstream acceptance clock can reset.

In January 2025, the U.S. Bureau of Industry and Security (BIS) announced strengthened restrictions on advanced computing semiconductors aimed at enhancing foundry due diligence and preventing diversion. BIS described “approved” outsourced semiconductor assembly and test services (OSAT) companies and a compliance structure around verified final chip characteristics. (bis.gov)

Those due diligence changes matter because they alter business operating models. Without stable pathways, customer order timing becomes risk management rather than production planning. A useful way to think about the new scarcity is that it converts uncertainty in compliance into uncertainty in delivery—specifically into effective lead time variability: even if wafers are started and processed on schedule, downstream shipping and acceptance can slip because licensing decisions and documentation cycles are not synchronized with factory takt time. The compliance shift also intersects with “where you can assemble” as much as “what node you can fabricate,” pushing companies toward architectures and routing choices that can pass compliance checks in the relevant jurisdictions and packaging channels.

The proof of this new operating model is visible in the administrative detail of how the U.S. has treated China tool access. Multiple reports in late 2025 and early 2026 indicated that Washington replaced “validated end-user” status with annual export licenses for specific Korean memory manufacturers’ tool shipments into China. For example, reporting tied to Reuters described annual licenses for 2026 covering chipmaking tool shipments for Samsung Electronics and SK hynix, while characterizing the end of the earlier validated-end-user regime at the year boundary. (tomshardware.com)

Quantitatively, the effect is not just theoretical. When access becomes “license-counted events rather than continuous operational flows,” the supply chain’s effective capacity becomes “throughput that survives compliance calendars.” In operational terms, this tends to show up as lumpy deployment of tools and packaging steps rather than smooth monthly output: even if wafer starts are planned, downstream shipping and acceptance can be delayed at the moment of controlled export clearance, turning what looks like allocation volatility to customers into a predictable consequence of legal gatekeeping. It also reshapes customer behavior—buyers start specifying delivery windows that implicitly assume a worst-case compliance cycle and build in buffers for acceptance testing and logistics, not just fabrication.

Data point: U.S. policy’s declared domestic capacity gap

In its January 2026 action under Section 232, the White House stated that U.S. capacity for semiconductors and semiconductor manufacturing equipment is insufficient to meet domestic demand. (whitehouse.gov)

China’s industrial chessboard, the counter-move in licensing

Export controls are often described as “cutting off technology,” but operationally they are closer to a switchboard: they can reroute, delay, compartmentalize by configuration, and re-temporize the pipeline—without shutting down every line overnight. BIS’s 2025 actions emphasized diversion prevention for advanced computing integrated circuits and strengthened due diligence structures, including the roles of approved OSAT companies and broader compliance frameworks. (bis.gov)

At the same time, reporting on 2026 licensing decisions suggests that access is not simply halted; it is recalibrated into an annual rhythm that firms can budget for—yet still can’t assume. For companies with manufacturing footprints in China, what changes is the predictability of tool supply and the scheduling risk that follows it. Annual licensing turns uncertainty into a budgeting line item, but it also creates a recurring planning cliff: year-boundary decisions can compress or expand the effective window for tool shipments, installations, qualification runs, and ramp to meaningful wafer output.

This is also where strategic stockpiling begins to look rational—but not in the simplistic “buy more” sense. The stockpile is increasingly targeted at the compliance-sensitive portions of the chain: spares for qualified tools, components with stable classification, packaging inputs whose routing is already cleared, and documented intermediate steps that reduce the need for repeated re-submissions. If legal pathways can change, and if tool access can hinge on annual approval cycles, then inventory decisions become hedges against administrative timing—not merely against physical disruption. That is a major difference from earlier supply shocks, which were often about transport time, factory downtime, or geopolitical disruption. The new shock has both physical and administrative dimensions: you can have wafers, but not always the “right” shipment route, and not always at the moment the customer planned to take delivery and begin downstream assembly.

In other words, “China’s industrial chessboard” is partly about manufacturing capacity—and partly about synchronizing production to the permission calendar. The counter-move is therefore not only to secure equipment, but to reduce the probability that a line is operational yet legally unable to ship into the intended channel when it matters.

Data point: scale of European Chips Act investment catalysis

The European Commission reported that under the Chips Act it had already catalyzed more than EUR 80 billion in investments in chip manufacturing capacity (as referenced in its update on milestones). (digital-strategy.ec.europa.eu)

Fab and packaging capacity is a competition with physics, cost, and time

The most visible industrial race remains “node advancement” and “geographic capacity expansion.” But in supply-chain terms, the competition is not only who can etch smaller transistors. It is who can deliver system-level output at scale under packaging constraints, equipment availability, and compliance barriers.

TSMC’s 2nm ramp and capacity planning is a case in point. Reporting based on TSMC’s own disclosures described 2nm-class chips beginning volume production in the fourth quarter of 2025, with further details on ramp plans and process variants. (tomshardware.com)

The editorial question is what happens when advanced wafers arrive faster than advanced packaging can absorb them. That is why advanced packaging capacity, such as CoWoS-like ecosystems and OSAT buildouts, has become structurally important to downstream customers. When packaging is the limiter, wafer capacity expansion alone does not smooth delivery times.

Data point: explicit “3-times short” framing of advanced-node capacity vs AI demand

One report cited TSMC leadership describing advanced process capacity as “not enough,” characterizing it as about “3-times short” relative to AI demand. (tomshardware.com)

Real-world case example 1: CoWoS and advanced packaging expansion needs more time than node transitions

The U.S.-linked advanced packaging buildout story highlight how packaging lags wafers. One report described Amkor breaking ground on an Arizona advanced packaging and test campus, positioning it as a response to a critical gap in the U.S. chip manufacturing chain, with a production roadmap extending to later node classes. (tomshardware.com)
Timeline and outcome: Arizona packaging campus announced through a ground-breaking in the mid-2025 reporting cycle, with production expectations described as extending to 2028 for later node classes, reflecting how packaging capacity is being treated as a long-lead national bottleneck rather than a “follow-on” investment. (tomshardware.com)

Memory is the quiet shock amplifier: NAND supply behavior and its pricing signals

AI workloads make memory a pacing item, not a supporting cast. NAND flash affects storage performance and upgrade cycles across data centers, edge devices, and consumer hardware, which can translate into rapid order changes. When memory makers adjust supply, it can quickly propagate to system OEM allocations.

TrendForce has described NAND pricing increases driven by spillover demand and changes in supply behavior, including specific quarterly price signal expectations. (trendforce.com) Other reporting has quoted Kioxia leadership describing that “cheap 1TB SSD” dynamics have shifted and that NAND supply was sold out for the year and likely through 2027. (tomshardware.com)

Data point: NAND pricing expectations

TrendForce reported expectations for NAND flash prices to rise by 5–10% in the fourth quarter of 2025, linking this to supply-demand dynamics and broader demand pressures. (trendforce.com)

Real-world case example 2: “sold out” expectations through 2027 reshaped customer planning

Kioxia’s messaging about NAND supply being sold out for the year and likely through 2027 became a planning constraint for OEMs and channel partners. The reported outcome is not just price speculation; it is schedule risk. (tomshardware.com)
Timeline and outcome: the reporting describes a “sold out” state in late 2025 with expectations extending to 2027, effectively turning inventory and procurement strategy into multi-year commitments. (tomshardware.com)

TSMC, Intel, Samsung: competition now includes compliance-ready routes

When commentators discuss “TSMC vs Intel vs Samsung,” the temptation is to reduce it to performance and yield. But in 2025 and 2026, competition increasingly includes which manufacturing plans can keep moving under changing export-control regimes.

TSMC’s geographic expansion is part of the capacity race, and reporting on U.S. buildouts emphasizes acceleration. Axios reported in January 2026 that TSMC was speeding up its Phoenix chip expansion. While such reports can be interpreted as political and industrial resilience narratives, their supply chain implication is straightforward: faster equipment installation can mean earlier volume throughput, which can help customers hedge against disruptions elsewhere. (axios.com)

Intel’s foundry story is also framed as capacity-led. Intel has published a “foundry fact sheet” describing planned investment and facility locations including Ohio, Ireland, and others, along with the scale of commitments. (intel.de) This matters because advanced packaging and equipment lead times are long, and customers increasingly treat manufacturing geography as a resilience feature.

Samsung, meanwhile, competes on node ramp timing and regional footprint strategies. While some sources are less authoritative than primary company disclosures, major reporting has tracked investment and production planning shifts. One article, for instance, described Samsung cutting foundry investment in 2024 to about KRW 5 trillion (about $3.5 billion) versus KRW 10 trillion the prior year, citing a report about allocation changes. (tomshardware.com)

Data point: Intel’s published scale of capacity investment

Intel’s foundry fact sheet states it is prepared to invest up to $200B in capacity within the decade (as published in the referenced document). (intel.de)

Strategic stockpiling is changing shape: from “buy more” to “buy correctly”

Stockpiling has always had a blunt form: reserve inventory against disruptions. What is changing is the sophistication of what “correctly” means. Export controls mean that “correctly” includes compliance feasibility, shipment routes, and packaging compatibility with the licensable supply chain.

This is why export licensing, due diligence requirements, and approved packaging/test ecosystems have become central to supply chain continuity. BIS’s 2025 due diligence framing around advanced computing semiconductors explicitly links supply continuity to compliance structures involving approved OSAT companies. (bis.gov)

In parallel, Europe’s policy response shows that resilience is being treated as an industrial ecosystem problem, not just a storage problem. The European Commission’s Chips Act updates emphasize catalysts for investment in manufacturing capacity and state-aid decisions. (digital-strategy.ec.europa.eu) That investment helps reduce reliance, but it still does not instantly replace the “licensable throughput” constraint.

What companies should do now: inventory strategy, contract design, and compliance-ready supply mapping

The most practical implication for procurement and operations is that supply-chain plans must treat compliance as an operational variable, like lead time and yield. That means redesigning contract incentives around licensable delivery windows, not just delivery dates. It also means building a multi-sourcing strategy that recognizes that packaging and test bottlenecks can be as constraining as wafers.

A good rule of thumb is to map your supply chain into three tiers: (1) wafer starts and process capacity, (2) packaging and test capacity, and (3) controlled-tool and controlled-output logistics. If you only diversify across the first tier, you may still be exposed. The Amkor Arizona packaging expansion described earlier reflects this industry learning: packaging has become a national bottleneck worthy of long-lead investment. (tomshardware.com)

Data point: Annual licenses replacing validated end-user pathways

Reuters-linked reporting indicated annual licensing for 2026 for tool shipments to China by Samsung and SK hynix, as the validated end-user status ended at the year boundary. This is an example of how planning must shift to approval cycles rather than assuming continuity. (tomshardware.com)

Forecast: by late 2027, “chip availability” will increasingly mean “approved route availability”

The next 18 to 24 months will likely intensify the same pattern: advanced node ramps will continue, but customers will keep experiencing constraints through packaging, memory tightness, and export-control calendars. TSMC’s reported 2nm volume production beginning in 4Q 2025 supports the notion that manufacturing capability is moving forward. (tomshardware.com) But demand has been characterized as outpacing capacity by multiple factors, reinforcing that customers will not stop rationing.

However, by late 2027, “chip availability” will increasingly be less about whether wafer starts exist and more about whether the downstream chain can clear the permissioning steps that convert chips into saleable deliveries. Expect a measurable shift in how buyers score suppliers: availability will be weighted toward suppliers who can demonstrate stable compliance routing (approved intermediary coverage, predictable documentation readiness, and historically smooth approval outcomes) rather than suppliers who merely promise production volume.

This forecast is grounded in two observed trends. First, policy instruments that emphasize due diligence structures and approved pathways (BIS 2025) effectively turn compliance readiness into a gating factor for shipment release. Second, administrative reshaping from open-ended access to time-bound annual licenses (2026 reporting on Samsung and SK hynix tool shipments) introduces a recurring rhythm of uncertainty that—over multiple cycles—encourages procurement practices built around approval probability and buffer time. In that world, the “availability metric” becomes composite: production capacity, packaging/test slot commitments, and legal route feasibility all contribute to whether a customer can confidently schedule build, acceptance testing, and revenue recognition.

By late 2027, companies that ignore route feasibility will increasingly experience the same pattern in different guises: production proceeds, then delivery windows collapse when paperwork, approvals, or approved-path constraints lag behind the factory calendar.

Policy recommendation and practical roadmap

Policy cannot undo long equipment lead times, but it can reduce administrative friction that turns uncertainty into cost. The clearest recommendation is for governments coordinating export-control compliance and industrial policy to standardize “time-bound, pre-authorized” compliance pathways for qualified manufacturing and packaging partners, with transparent renewal windows that align with customer quarterly planning cycles.

In the U.S., that means BIS and the Commerce Department should continue refining due diligence and approval systems in ways that reduce year-to-year unpredictability, building on the structure described in the January 2025 BIS rules. (bis.gov)

For practitioners, the immediate roadmap from now through 2027 is simple: (1) include packaging and test capacity in resilience modeling, (2) treat export compliance timelines as constraints in forecasting, and (3) negotiate contracts that reward early notification of approval-related delays rather than penalizing production changes that are driven by licensing. When “availability” stops being a single number and becomes a multi-dimensional variable, the companies that plan best will look less surprised by allocation volatility.