How Do Export Controls Affect the AI Chip Supply Chain? Nvidia, ASML, and TSMC Risk Analysis

Export controls will not eliminate global AI chip demand, but they can restrict the connection between advanced chips, semiconductor equipment, software, technology, and specific customers. For Nvidia, the key risks are China market access, the competitiveness of compliance-oriented chips, and inventory losses. For ASML, risks center on equipment licenses, delivery, and service. For TSMC, the impact is more visible in customer screening, foundry compliance, and order scheduling. You need to separate policy shocks, financial losses, and long-term supply-chain restructuring.

Key Takeaways

• Export controls cover chips, equipment, software, technology, end users, and end uses.
• Nvidia faces the most direct risk to revenue, inventory, gross margin, and product planning.
• ASML’s risk is concentrated in equipment licensing, service restrictions, and China exposure.
• TSMC faces customer identification, foundry due diligence, and advanced-node order shifts.
• The long-term result is more likely supply-chain fragmentation than the end of AI investment.
• Investment analysis should combine policy, financials, orders, inventory, and valuation.

What Do AI Chip Export Controls Actually Restrict?

AI chip export controls do not simply restrict a few GPU models. They form a broader regulatory framework covering product performance, customer identity, end use, manufacturing equipment, software technology, and re-export routes. When assessing policy impact, you first need to distinguish between a full prohibition, a license requirement, case-by-case review, end-user restrictions, and end-use restrictions. The same controlled product can have very different commercial outcomes, ranging from a complete sales block to limited delivery after approval.

Export controls usually cover six categories. The first is AI accelerators that exceed certain thresholds for compute performance, performance density, memory bandwidth, or interconnect capability. The second is servers, rack systems, and computing equipment containing controlled chips. The third is equipment used for advanced chipmaking, including lithography, etch, deposition, inspection, and metrology. The fourth is controlled software, technical data, and certain support activities. The fifth is restricted entities, military end users, or supercomputing end uses. The sixth is re-export or in-country transfer through third-country distributors, cloud providers, or affiliates.

In 2026, the U.S. Department of Commerce’s BIS adjusted the review approach for some advanced computing products. The shift to case-by-case review for H200 and MI325X means certain chips are no longer automatically reviewed under a presumption of denial, but it does not mean free export. Applicants still need to satisfy security conditions, customer screening, capacity protection, and relevant technical verification requirements.

The suspension of the earlier AI Diffusion Rule is also easy to misread. BIS later emphasized that advanced computing item license requirements still apply to entities whose headquarters or ultimate parent companies are located in specified country groups. In other words, the policy moved away from a broader global tiering framework and back toward a targeted review path focused on priority countries, customers, and products. That does not mean restrictions on advanced AI chips have been fully removed.

For companies, the most difficult issue is uncertainty. A chipmaker may have completed product design, wafer starts, HBM procurement, and packaging capacity planning, only to find that a policy change suddenly makes the product license-controlled or non-deliverable. Equipment companies may have signed orders but face delays in delivery or service because of license changes. Even a foundry that does not directly sell GPUs still has to determine whether a customer’s chip falls within a controlled category.

Summary: The core of AI chip export controls is not a simple list of chips that cannot be sold. It is a system that restricts the connection between advanced computing capability, manufacturing capability, and specific customers. To assess whether a policy affects a company, you need to check product parameters, customer headquarters, ultimate parent company, transaction destination, end use, license policy, and re-export rules. A shift from presumption of denial to case-by-case review only improves commercial possibility; it does not mean revenue will immediately recover. Testing, approval, customer screening, import permits, and product competitiveness can still determine whether orders are actually delivered.

What Risks Do Nvidia, ASML, and TSMC Face?

Nvidia, ASML, and TSMC all sit at the center of the AI chip supply chain, but export controls affect them at different points. Nvidia faces the most direct risks to chip sales, inventory, and product planning. ASML faces equipment licensing, delivery, and installed-base service risks. TSMC faces customer identification, foundry compliance, and order interruption risks. You cannot judge the risk only by China revenue exposure; you also need to assess whether products can be resold, whether capacity can be shifted, and whether customers can be replaced.

Nvidia: Product Restrictions, China-Specific Chips, and Inventory Risk

Nvidia faces the most direct risk because high-end GPUs themselves are core targets of export controls. The stronger the chip performance, interconnect capability, and HBM configuration, the more likely the product is to fall within a controlled category. Nvidia can design lower-specification versions for specific markets, but that creates extra R&D costs, shorter product life cycles, weaker pricing power, and inventory risk if policy thresholds change again.

In its SEC filing, Nvidia disclosed that H20 export license requirements led the company to record $4.5 billion in charges related to excess inventory and purchase obligations in the first quarter of fiscal 2026. This example shows that export controls are not just abstract geopolitical risks. They can directly enter the income statement and affect gross margin, inventory, procurement commitments, and supply-chain contracts.

Even if Nvidia receives certain licenses, the risk does not disappear. A compliance-oriented product must satisfy both regulatory requirements and customer performance needs. If the chip is too weak, customers may turn to domestic accelerators, cloud-based alternatives, or in-house ASICs. If the chip is powerful enough, it may trigger new review standards. Nvidia therefore faces a dual challenge: whether it is allowed to sell, and whether what it can sell is still competitive.

ASML: Equipment Licenses, Delivery, and Service Revenue Risk

ASML’s risk mainly lies in equipment licensing and service continuity. EUV has long been restricted for advanced chip manufacturing in China, and some advanced DUV, metrology, and inspection technologies can also fall under license review. The Dutch government announced that from April 2025 it would expand export licensing for advanced semiconductor manufacturing equipment, meaning more equipment and technology would require case-by-case authorization.

ASML’s pressure is not limited to the inability to deliver new equipment. It also includes delayed signed orders, restricted software upgrades, more complex spare-parts support, and uncertainty around on-site maintenance. In its statement on U.S. rule changes, ASML noted that updated U.S. export restrictions could affect certain products and customers, while also emphasizing that long-term semiconductor demand is still driven by global wafer demand.

This means ASML’s risk is more like a shift in regional and product mix, rather than the disappearance of global demand. If the U.S., Europe, South Korea, Japan, and other regions continue building advanced logic, memory, and packaging capacity, ASML may still absorb part of the demand through other customers. But if China equipment demand falls faster than other regions can offset it, near-term orders, backlog, and service revenue may still fluctuate.

TSMC: Customer Screening, Foundry Responsibility, and Order Interruption Risk

TSMC is not the branded seller of high-end GPUs, but it can face more complex foundry compliance responsibilities. U.S. export rules can affect overseas manufacturing through the foreign direct product rule, requiring foundries to assess whether a customer’s chip uses controlled technology, reaches controlled performance thresholds, or serves a restricted end user or end use.

In its [2025 Form 20-F](https://investor.tsmc.com/sites/ir/sec-filings/2025_20F Report.pdf), TSMC disclosed that certain products using 16nm and more advanced process technologies may be subject to license requirements under specific destinations and conditions, which could lead to order delays, delivery restrictions, or shipment prohibitions. For TSMC, the risk does not come from just one customer. It comes from customer mix, order screening, compliance processes, and advanced-node capacity reallocation.

TSMC’s advantage is that global demand for advanced process capacity remains strong. If some restricted customers can no longer place orders, capacity may shift to other GPU, AI ASIC, CPU, networking chip, or high-performance computing customers. But the transition is not cost-free. Customer validation, masks, design rules, packaging routes, and delivery schedules all need adjustment. If restricted orders are highly customized, near-term scheduling impact can be more visible.

Summary: Nvidia, ASML, and TSMC do not face the same type of risk. Nvidia is most exposed to short-term revenue and inventory shocks. ASML’s pressure is more visible through licenses, equipment delivery, and service revenue. TSMC must manage foundry due diligence across customers and regions. To judge the impact of export controls, do not only compare China revenue exposure. Also assess whether restricted products can be resold, whether capacity can be shifted smoothly, whether supply contracts can be adjusted, and whether global AI demand can fill the restricted market gap.

How Do Export Controls Flow Through the AI Chip Supply Chain?

The impact of export controls flows from chip sales into foundry production, HBM, advanced packaging, semiconductor equipment, servers, and cloud data centers. In the short term, it appears as order cancellations, inventory write-downs, and delivery delays. In the medium term, it appears as product redesign, customer migration, and capacity rescheduling. In the long term, it may lead to fragmentation across hardware architectures, software ecosystems, and regional supply chains. This is not a single-point risk; it is a chain reaction.

Short-Term Shock: Inventory, Orders, and Delivery Timing

When restricted chips cannot be delivered normally, the first pressure falls on chip companies and direct customers. Finished-goods inventory may be written down. Purchase commitments for HBM, substrates, advanced packaging, and servers may create losses. Data center projects may be delayed. License approval timelines also reduce revenue predictability, making management guidance harder to stabilize.

The upstream supply chain is affected as well. TSMC may need to adjust wafer schedules. HBM suppliers may reallocate long-term agreements. Packaging suppliers may shift CoWoS or other advanced packaging capacity to other customers. ODM server companies may have to wait for customers to redefine chip configurations. If customers purchase ahead of rule implementation, this can create a short-term order spike followed by a digestion period.

Medium-Term Adjustment: Product Redesign and Supplier Replacement

In the medium term, companies redesign products around policy rules. Chipmakers may introduce versions with lower compute performance, lower interconnect capability, or lower memory bandwidth for specific markets. Cloud providers may increase their use of in-house ASICs, AMD accelerators, or multi-vendor procurement. Customers in restricted regions may accelerate adoption of domestic GPUs, AI ASICs, software frameworks, and cluster networking solutions.

These changes create new costs. Chip companies need to maintain separate product lines for different regions. Software teams need to optimize for different hardware platforms. Customers need to redo model migration and performance tuning. On the surface, it may look like replacing one chip with another. In practice, it involves drivers, compilers, communication libraries, inference frameworks, cluster scheduling, and operations systems.

Long-Term Change: Duplicate Investment and Supply-Chain Fragmentation

Over the long term, export controls may push the supply chain toward fragmentation. The U.S., Europe, Japan, South Korea, and other regions may expand local advanced manufacturing, packaging, and data center capacity. China may increase investment in equipment, EDA, materials, memory, and domestic accelerators. Different regions may maintain different hardware and software versions for the same AI workload, reducing global coordination efficiency and increasing duplicated investment.

This is why export controls can produce a result where demand does not disappear, but the demand path becomes more complex. Restricted GPU orders may be blocked, but wafers, HBM, and packaging capacity may not stay idle forever. If global AI capex remains strong, these resources can be reallocated to U.S. cloud providers, Middle Eastern data centers, European sovereign AI projects, or other AI ASIC customers. The issue is that reallocation takes time, and near-term financial shocks cannot be fully avoided.

Summary: Export controls do not only affect the chip company named in the rule. A GPU delivery restriction can affect wafers, HBM, packaging, servers, and cloud data center projects, but those capacities are not necessarily permanently idle. If global AI demand remains strong, TSMC, HBM, and advanced packaging capacity can gradually shift to other customers. If a product is highly customized, redistribution will be slower. The most important long-term change is that companies begin maintaining separate products, customers, and supply chains for different regions. Costs rise, efficiency falls, but supply-chain resilience may improve.

Will Export Controls Reduce Global AI Demand or Redirect It?

Export controls are more likely to change the geography, products, and supplier structure of AI chip demand than to immediately reduce total global AI investment. In restricted regions, advanced compute supply declines, procurement costs rise, and deployment cycles lengthen. In other regions, cloud providers, governments, and enterprises may gain access to more advanced chip supply. The final impact depends on whether restricted orders can be absorbed by other customers and whether domestic alternatives become good enough.

Demand changes can appear in four forms.

Demand Is Delayed

• Customers wait for U.S. export licenses or local import approvals.
• Data center projects delay server procurement.
• Chip, networking, HBM, and power plans need to be redesigned.
• Software teams must re-optimize for different hardware platforms.

Demand Is Redirected

• Restricted GPU capacity shifts to the U.S., Middle East, Europe, and other cloud markets.
• TSMC advanced capacity shifts to other GPU, ASIC, CPU, or networking chip customers.
• ASML equipment demand moves toward new fabs and advanced-node projects in other regions.
• HBM and advanced packaging capacity shifts toward more compliant customers.

Demand Is Substituted

• Chinese customers accelerate adoption of domestic AI accelerators.
• Cloud providers raise the share of in-house ASICs.
• Enterprises use model compression, quantization, and inference optimization to reduce dependence on high-end chips.
• Some applications accept lower performance hardware that is easier to obtain.

Demand Is Duplicated

• Different regions build similar fabs, packaging plants, and data centers.
• Enterprises procure separate server systems for different regulatory regions.
• Supply-chain security requirements increase buffer inventory.
• Compliance, audit, and customer screening become new fixed costs.

For Nvidia, export controls may reduce part of China-related revenue, but global demand for Blackwell, Rubin, and rack-scale systems may still absorb some capacity. For ASML, long-term demand is driven by global wafer volume, process complexity, and advanced-node migration, so equipment orders may shift geographically. For TSMC, the most likely effect is customer mix change rather than all restricted orders turning into idle capacity.

But a redirected demand path does not mean every company transitions smoothly. If a product is customized for a specific customer, rack system, or packaging route, resale can be slower. If a customer’s software stack is deeply tied to one hardware platform, migration costs can also be high. The margin impact of export controls often comes from this time lag during reconfiguration.

Summary: Export controls mainly reprice and reallocate global AI demand. Restricted customers must pay higher compute costs or accept lower performance, while other regions may receive more advanced chip supply. If global AI capex keeps growing, total supply-chain revenue may still expand, but market share will shift between companies and regions. You need to distinguish between “global demand is falling” and “one company is losing a specific market.” These two scenarios have very different implications for Nvidia, ASML, TSMC, HBM suppliers, and equipment companies.

How Can Investors Tell Whether Export Controls Are a Short-Term Shock or Structural Risk?

To judge whether export controls are a short-term shock or a structural risk, track policy enforcement, company financial data, product competitiveness, and supply-chain migration together. A one-time inventory charge from a new rule may be temporary. But if a company cannot provide compliant and competitive products over time, customers build alternative ecosystems, and supply-chain contracts move permanently, the risk shifts from a temporary expense to lasting market-share loss.

You can monitor four sets of indicators.

Policy Indicators

• Whether BIS adjusts performance thresholds.
• Whether license policy is presumption of denial or case-by-case review.
• Whether the Netherlands, Japan, and other equipment-rule regimes tighten in parallel.
• Whether Entity List and end-user rules expand.
• Whether China introduces import restrictions or countermeasures.

Nvidia Indicators

• China-related revenue and order changes.
• Inventory write-downs and purchase commitment charges.
• Speed of compliant product development and launch.
• Data center gross margin changes.
• CUDA developer stickiness in restricted markets.

ASML Indicators

• China system sales share.
• DUV, EUV, and installed-base service revenue changes.
• Order delays caused by licenses.
• Whether global backlog can offset restricted demand.
• Construction pace of fabs in other regions.

TSMC Indicators

• Advanced-node capacity utilization.
• High-performance computing revenue growth.
• Customer concentration and new customer ramp.
• Order delays caused by compliance review.
• Overseas fab costs and capacity ramp progress.

Market pricing often moves before revenue changes. After a policy headline, share prices can react immediately, while the actual revenue, inventory, and gross-margin impact may appear later in financial statements. Conversely, a license approval does not mean customers will definitely buy, because they still compare performance, price, delivery time, local import approvals, and alternatives. Investment decisions should not rely only on headlines. They should also assess whether the impact has already been reflected in valuation.

If you follow AI chip supply-chain stocks such as NVDA, ASML, TSM, AMD, AVGO, and MU, trading costs also matter alongside policy changes. U.S. stock trading costs may include commissions, platform fees, external institution fees, transaction activity fees, and other charges. You can use U.S. stock market information to track related names and review Biya U.S. stock trading fees, where commissions, platform fees, external institution fees, and other charges are subject to the fee center and order page.

Summary: The sign of structural risk is not one inventory charge. It is a situation where products remain unable to enter the target market, customers build replacement ecosystems, and supply-chain contracts shift for the long term. Nvidia must show whether it can launch competitive and approvable products. ASML must show whether global orders can fill the gap left by restricted equipment demand. TSMC must show whether advanced-node orders can move smoothly to other customers. Investors should put licenses, orders, gross margin, inventory, and valuation in one framework, rather than interpreting policy easing as immediate revenue recovery or a single restriction as the collapse of global AI demand.

If you want to keep tracking how export controls affect the AI chip supply chain, place policy documents, company financials, inventory charges, license progress, order changes, and valuation in one watchlist. NVDA, ASML, TSM, AMD, AVGO, and MU represent different parts of the chain, including chips, equipment, foundry, custom ASICs, and memory, so policy impact will differ across them. Biya supports multi-asset trading across U.S. stocks, Hong Kong stocks, and digital assets, making it suitable for watching AI chips, semiconductor equipment, and memory supply-chain names in one asset pool. Availability depends on your location, identity verification results, platform rules, and applicable laws and regulations. Public market information and fee structures are for reference only and do not constitute investment advice. Before trading, understand order types, fee details, price volatility, and your own risk tolerance. You can also use the Biya App to continue following related companies.

FAQ

Can Nvidia H200 Be Exported Directly to China in 2026?

No, H200 cannot be exported to China directly and without restrictions. Even if H200 and similar advanced AI chips are reviewed case by case, they still require customer screening, capacity protection, performance and security verification, and license approval. Actual delivery may also depend on China import approvals, customer demand, and commercial competitiveness.

Does Dutch Export Control Ban All ASML Sales to China?

No. Dutch export controls do not impose a blanket ban on all ASML sales to China. They apply export licensing to certain advanced lithography, metrology, and inspection technologies. Approval outcomes can differ by equipment model, customer, destination, and end use. The impact should be judged based on Dutch rules, license progress, and ASML disclosures.

Will Export Controls Reduce HBM Demand?

Export controls do not necessarily reduce global HBM demand. Restricted GPU orders may be delayed, but global demand from GPUs, AI ASICs, and cloud data centers can still reallocate HBM capacity. The key variables are how customized the restricted product is, the strength of other customer orders, HBM long-term agreements, and advanced packaging conversion speed.

Can Compliance-Oriented Lower-Spec Chips Eliminate Nvidia’s Risk?

No. Compliance-oriented lower-spec chips cannot fully eliminate Nvidia’s policy risk. Regulatory thresholds may cover compute performance, performance density, memory bandwidth, and interconnect bandwidth, while the product must still be commercially competitive. Repeated product adjustments can increase R&D costs, inventory risk, and life-cycle management complexity.

How Should Investors Track AI Chip Export Controls?

Investors should prioritize BIS, the Dutch government, company earnings reports, SEC filings, and formal regulatory announcements. Media reports can provide early signals, but product scope, license rules, and financial impact should be verified through official rules and company disclosures. Orders, inventory, gross margin, customer mix, and valuation should be reviewed together.