Taiwan’s Alternatives to Chinese Suppliers Still Depend on Chinese Inputs

Taiwan’s supply-chain exposure is not confined to the advanced chips that power AI systems. It also sits in the manufacturing steps and materials that turn chips into usable systems—and in the battery, software and component networks needed to build drones. Across these sectors, the strategic problem is not simply whether Taiwan can replace Chinese suppliers. It is whether alternative capacity can be built, qualified and sustained without disrupting production before substitutes are ready.

The three research areas presented by the Democracy, Society and Emerging Technology institute (DSET) expose different parts of that problem. Jeremy Chang focused on advanced packaging and optics, arguing that semiconductor competition is extending from chip fabrication into how chips are integrated and connected. Samara Duerr described Taiwan’s effort to grow a drone industry through domestic procurement and international cooperation, while emphasizing limits in cost, scale and components. Tsalying Lu traced the mineral dependencies beneath both industries, including materials that can reach Taiwan through suppliers in Japan, Europe or the United States while remaining dependent on Chinese feedstocks.

The policy choices vary by sector. Chang argued that export controls designed around the chip itself may not reach the industrial capabilities that matter next. Duerr stressed that drone exports do not by themselves establish a durable industry: demand must persist beyond wartime surges, and component supply and production capacity must develop alongside finished platforms. Lu described a less visible vulnerability: Chinese controls can delay or raise the cost of materials even when exports have not stopped.

That distinction runs through the discussion. Moving production out of China does not necessarily remove Chinese exposure. A supplier’s location may not reveal who controls it, where its inputs originate or whether alternatives are affordable at the required scale. The speakers’ proposals therefore combine diversification with procurement commitments, supply-chain visibility and cooperation among partners.

Chang’s account begins with a shift in where semiconductor competition takes place. Public debate, he said, has concentrated on the front end: foundries, advanced manufacturing nodes and equipment such as ASML’s. But as applications demand more computing performance, advanced packaging has become another frontier. Packaging brings semiconductor dies together into a finished system; newer approaches can also link packages to one another. Chang called it the “next frontline” because it can improve system-level performance even when progress in making individual chips is constrained.

He distinguished three routes to advanced packaging. The most demanding is foundry-led, with very fine interconnects produced through wafer-level processes. Chang’s presentation named TSMC’s CoWoS and Intel’s EMIB as examples. A second, foundry-OSAT model combines foundry wafer steps with organic substrates and less precise wiring. The third, OSAT-led route uses larger-panel fan-out and coarser wiring. It is the most accessible starting point for Chinese firms, though the presentation characterized its performance ceiling as limited.

OSAT means outsourced semiconductor assembly and test. Chang clarified that OSAT is distinct from a foundry: a foundry makes the wafer; packaging and testing follow, often at a separate facility. Assembly and test were historically treated as lower-value work, and Taiwanese firms helped establish much of the industrial base in China through investments in packaging-related manufacturing. That history left Chinese suppliers with substantial capacity in a stage of production that is now becoming more strategically important.

Chang’s claim was not that Chinese companies already match the most advanced packaging capabilities. Rather, they can build from an existing OSAT base and try to move toward more demanding approaches. He described the OSAT-led route as a way to catch up and compete with the other models over time—not as an immediate substitute for frontier packaging. The distinction matters because routine assembly capabilities can become strategically consequential when packaging itself is part of the performance gains sought by AI systems.

The scale evidence in DSET’s presentation also placed Chinese firms among the largest global OSAT providers. It identified JCET as the world’s third-largest OSAT company, after ASE and Amkor, and described Tongfu and Hua Tian as other large Chinese firms. The presentation’s figures and reporting periods differ across slides, so the ranking is clearer than any single sales comparison. Chang’s broader point was that China’s foundation is not hypothetical: it includes firms with established scale and a history of supplying outsourced services.

The next policy question, in his view, is how to address an industry where China has strength in lower-level manufacturing while the most advanced capabilities remain elsewhere. Chang’s packaging recommendations went beyond controls on advanced chips. DSET proposed specification-based export controls on AI-grade packaging materials—including ABF film, T-glass and packaging photoresist—while exempting consumer-grade materials. It also proposed tighter controls on equipment and processes such as laser drilling, hybrid bonding and AI-grade test equipment, coordinated with Japan, the Netherlands, Germany and South Korea.

Chang also argued for changing the unit that policy measures. Rather than relying only on chip-level thresholds, DSET recommended considering compute density after packaging, bringing all generations of high-bandwidth memory and FC-BGA substrates into scope, and reviewing key suppliers tied to Huawei’s Ascend systems for possible Entity List action. The proposal reflects a concern that a chip-level measure can miss performance created by combining and connecting components. The presentation also called for downstream tracking, end-user certification and checks for unusual orders routed through transshipment hubs, alongside intelligence-sharing among the United States, Japan, South Korea and Taiwan.

The case for such measures is not straightforward. Chang said export controls had helped preserve time and maintain an advantage at the manufacturing frontier. But he argued that controls alone are poorly suited to competition in packaging and other lower-level capabilities where China already has industrial strength. If US and allied firms still rely on Chinese suppliers for key inputs, restrictions imposed too abruptly could disrupt their own production. Chang framed the task as reducing dependency while avoiding measures that inflict harm before alternative capacity is available.

Chang’s second example was silicon photonics and co-packaged optics (CPO). In current systems, pluggable transceivers connect to switches through a chain of optical components. CPO places optical engines closer to the switch chip. Chang described the technology as early, while saying that NVIDIA was adopting CPO solutions supplied by Taiwanese firms. As systems change, the boundary between suppliers can move: companies that once supplied complete optical modules may become less central, while providers of fiber, micro-optics and optical engines remain important.

The transition does not create a clean break from Chinese suppliers. Chang named InnoLight and Eoptolink as major Chinese optical-module manufacturers and identified TFC as a supplier of fiber-related solutions for NVIDIA’s CPO systems. He said that, under “China Plus One” strategies, Chinese suppliers had moved some manufacturing to Southeast Asia, particularly Thailand. He described TFC’s subsidiary Fabrinet as supplying parts for CPO from Thailand. For policymakers, that raises a question distinct from where an item is assembled: should a Chinese-controlled supplier operating outside China still be treated as a source of risk?

DSET’s optics recommendations were designed around that problem. The presentation called for mapping exposure below the finished product, including components, subassemblies, equipment, ownership and control, and sub-tier suppliers. Risk-tiered bills of materials and sub-tier disclosures were proposed as ways to improve visibility. It also recommended pairing restrictions with qualified substitutes: policymakers should account for replacement capacity and qualification timelines before restrictions take effect, phase measures where appropriate, and use procurement commitments to help alternatives enter the market.

The remaining proposals were to make supply-chain policy architecture-aware and coordinate around a shared exposure map. Chang’s presentation urged policymakers to follow capabilities rather than broad product categories, prioritizing harder-to-substitute functions such as integration, laser sources, fiber attachment, packaging and testing. It argued that inbound restrictions, export controls, industrial policy, procurement and qualification support should be based on a common assessment of exposure.

Chang connected the same concern to offshore computing. He described data centers in Malaysia, including facilities in Johor, that he said serve Chinese customers through remote access. He said Taiwanese AI-server assemblers work with NVIDIA and TSMC to bring chips to those facilities, while current US rules do not restrict remote access to computing in the same way they restrict direct shipment of chips to China. Chang called this a policy loophole. He characterized offshore compute as a temporary way to obtain capacity, not the best long-term strategy for China. He also said DSET had identified different business models involving joint ventures with Singaporean capital and Malaysian real-estate investors; he did not claim that every facility or arrangement served the same customers.

He cautioned against treating packaging or distributed compute as a route to parity. Without access to the most advanced lithography, he argued, Chinese chipmakers face performance and energy-efficiency limitations. Packaging and interconnection can help assemble systems around those constraints, but do not eliminate them. His broader warning was that controls on the front end do not settle a contest increasingly shaped by integration, suppliers and system deployment.

Duerr described Taiwan’s drone policy as an effort to build an industry that can meet defense needs and supply partners seeking alternatives to Chinese components and platforms. The Lai administration’s three objectives, as she presented them, are industrial development, greater defense self-reliance and integration into democratic supply chains. A six-year unmanned-vehicle program set out four approaches: expand domestic and international demand, advance technology and cooperation, build industrial clusters, and improve regulation. The program’s stated investment was NT$44.2 billion over 2025–2030, with a target of NT$40 billion in industry production value by 2030.

Duerr reported rapid growth in exports, while distinguishing those shipments from production capacity and government targets. The presentation listed 3,472 exported units in 2024 and 122,726 in 2025. Duerr said first-quarter 2026 exports had already surpassed the full-year 2025 total; the slide put that first-quarter figure at 139,019 units. She then reported 212,209 units exported from January through July 2026, exceeding a stated annual production goal of 180,000 units by 2028. These are reported export figures, not a measure of Taiwan’s annual production capacity or of drones procured by Taiwan’s government.

The demand behind the export surge is closely tied to Ukraine. Duerr said Taiwanese firms’ exports to Europe had gone chiefly to the Czech Republic and Poland, with products transferred onward to Ukraine. She described the market as wartime-driven. The kinds of systems Taiwan is pursuing also reflect lessons from Ukraine’s mass-attrition war, including demand for large numbers of relatively inexpensive FPV drones.

But a surge in exports is not the same as a stable market. Duerr said Taiwanese drones cost two to three times more than Chinese products, in part because Taiwanese manufacturers have not yet reached comparable economies of scale. Greater sales could lower costs and support more production, but companies need confidence in continuing demand before investing in equipment, workers and capacity. Duerr argued that long-term domestic procurement contracts are necessary to provide that signal.

Budget decisions have made demand less predictable. Duerr said an earlier defense budget proposal was cut in May, removing domestic production and defense-manufacturing provisions while retaining Foreign Military Sales. A separate defense budget passed in August restored some provisions, including co-production efforts between Taiwanese and US firms. But she cautioned that the approved framework was phased and subject to annual review. In her assessment, delays and conditional funding meant that there would be no mass procurement during 2026 under that framework. The budget was an important first step, she said, but not yet a dependable basis for sustained production.

The ambitions and the current scale are far apart. The presentation described a long-term goal of 1.2 million units of annual production capability by 2030, equivalent to 100,000 units per month, with half intended for export. Duerr agreed with a participant who contrasted Taiwan’s goals with Ukraine’s needs, which were discussed in the millions of drones. Her point was not that Taiwan must reproduce Ukraine’s wartime production, but that present export totals and targets should not be mistaken for capacity on that scale.

Taiwan’s contribution may also be larger in components than finished-unit figures suggest. Duerr said Ukrainian firms had sought Taiwanese airframes, flight controllers, battery cells and motors, with more recent demand extending to cameras, gimbals, sensors, fiber optics and jet engines. These sales are difficult to measure because many components are dual-use and are not separately classified as drone parts in trade data. A survey of one Ukrainian technology cluster found that seven of 61 UAV companies sourced Taiwanese components; Duerr emphasized that this was not a comprehensive survey of the Ukrainian industry. She said a later delegation of 20 Ukrainian manufacturers had visited Taiwan, and that component sales appeared to be increasing.

A component-first pathway may fit Taiwan’s manufacturing strengths better than relying only on finished platforms. Duerr said some successful drone companies entered from other industries, shifting existing manufacturing and revenue streams into the market. Taiwan has capabilities in precision manufacturing, she argued, but makers face bottlenecks in production scale and cost, US export controls on some military-grade sensors, and dependence on Chinese battery materials and rare-earth inputs for motors.

There are also technology gaps. Duerr identified flight-control, positioning and communications chips; flight and ground-control software; and AI-enabled capabilities such as command-and-control and autonomy as areas where Taiwan is still building capacity. Government subsidies and research programs are intended to help, but their results take time. She said partnerships with US companies are developing around autonomy and AI software, while business-to-business ties with Ukrainian firms have grown since Chinese export restrictions in 2023 pushed some buyers to seek alternative sources.

Certification is another part of market access. Duerr reported that Thunder Tiger became the first Taiwanese company on the US Blue UAS Cleared List, and that Taiwan’s Industrial Technology Research Institute became a designated Green UAS cybersecurity assessor. Taiwan’s National Chung-Shan Institute of Science and Technology and Aerospace Industrial Development Corporation had also achieved CMMC Level 2 certification. These steps can help Taiwanese firms qualify for US markets, but Duerr noted that only one Taiwanese company was then on the Blue UAS list.

Her recommendations for US-Taiwan cooperation included a working group to accelerate certification, a roadmap for interoperability in areas such as spectrum, identification-friend-or-foe systems and airspace coordination, and a testing and data-sharing platform for Taiwan Strait conditions. She also called for formal participation in exercises and measures to ensure small drones are included in contingency stockpiles. The recommendations address a gap between having products and being able to integrate, test and procure them in a partner’s systems.

International partnerships remain uneven. Duerr said cooperation with Central and Eastern European countries, particularly Lithuania and Czechia, appeared stronger than ties with Western Europe. She cited interest in working with Taiwan and a European Parliament call for cooperation on drone technology, but said institutional plans had not consistently caught up with those signals. The EU market is fragmented in its requirements. Duerr argued for more government procurement, co-development, component sourcing and shared testing; she described Lithuania as the only European country with an overall non-red policy.

Ukraine offers another route, though it is constrained by the war and by Ukraine’s own supply-chain dependencies. Duerr advocated a “volume-first” component pipeline, a combat feedback loop to turn field experience into product improvements, and state-backed co-production of long-range and interceptor drones. DSET’s recommendations also called for streamlining import procedures for Ukrainian firms partnered with Taiwanese companies and expanding defense-technology education exchanges. The proposed co-production examples included the long-range AQ-400 Scythe and the Sting interceptor drone.

Participants raised whether Taiwan could gain scale by partnering with Ukrainian firms and manufacturing in Ukraine or neighboring European countries. Duerr noted the obstacle: Ukrainian producers need their output for the war effort. She did not present overseas co-production as an assured shortcut. She also said Ukraine had been cautious about formal government-to-government ties with Taiwan, in part because Ukraine is not fully independent of Chinese supply chains.

Duerr described trilateral arrangements as another possible path—through countries that can work officially with Ukraine, through Japan, or through US companies and subsidiaries. She cited a Ukrainian company’s arrangement with a US company as an example of a possible route for cooperation. The constraint remains financial as well as diplomatic: Taiwan’s suppliers need partners willing to pay higher prices while production scales. Duerr said the United States has non-red requirements in some programs, while European policy is less consistent. Taiwan’s ability to become a premier alternative supplier depends, in her account, on allies helping create a market large enough to sustain that capacity.

For Lu, critical minerals are the underlying vulnerability in both drone batteries and semiconductor-AI hardware. Taiwan may manufacture cells, chips, servers and other components, but production depends on materials and intermediate products for which China retains a strong position. That dependence can persist even when Taiwanese firms import from a supplier elsewhere.

The battery example shows the difference between an alternative supplier and an independent supply chain. Lu said Taiwanese battery exports to Ukraine had risen sharply after 2024, moving from outside the top 20 suppliers to third place in the export figures she discussed. One manufacturer she interviewed reported exporting five to seven million battery cells to Ukraine per month. Lu said these were likely nickel-manganese-cobalt (NMC) cells used in drones.

NMC batteries are useful for drones because of their energy density: they can support longer flight times and heavier payloads. Lu described NMC 811 as a high-nickel chemistry. The slide associated that chemistry with a composition of roughly 75% nickel, 12.5% manganese and 12.5% cobalt by weight for the precursor material. Separately, Lu said cathode and anode materials together account for about 70% of a finished cell’s cost. Those figures refer to different measures—material composition and cost composition—and should not be read as the same calculation.

China’s position in upstream materials creates exposure even when battery cells are made in Taiwan. Lu pointed to export controls on graphite, used in anode materials, and later controls affecting high-energy lithium-ion batteries and other inputs. She emphasized that the restrictions she described had not yet amounted to a complete ban on exports to Taiwan. Instead, they added time to export reviews: Taiwanese firms were waiting around 45 days for some graphite shipments, while adapting by seeking approvals earlier and diversifying suppliers, especially toward Japan. Lu characterized the delays as targeted pressure and an underlying risk, not proof that supply had already been cut off.

That distinction shaped the exchange. Asked whether continued Taiwanese battery exports meant the restrictions had failed, Lu said that the measures were delaying shipments rather than blocking supply outright. She argued that companies and government should still prepare alternatives and consider stockpiling. Substitutes may exist but cost substantially more: Lu said non-Chinese sources for some battery materials could be three to four times as expensive. The premium creates a procurement problem, particularly when the market for drone-specific NMC production remains small.

Taiwan’s domestic battery industry illustrates that demand problem. Lu said most Taiwanese battery manufacturers produce lithium iron phosphate cells rather than the NMC cells used in the drones she discussed, because the market is too small to support scaled NMC production. She argued that government procurement signals should support not only finished drones but also battery cells and components. A larger, more predictable market could give manufacturers a reason to invest in capabilities that are not currently economical.

The semiconductor supply chain presents a parallel exposure. Lu identified materials used in AI servers, power semiconductors and optical components, including silicon carbide, gallium nitride, gallium arsenide and indium phosphide. She argued that importing a finished or intermediate product from a non-Chinese supplier does not necessarily remove Chinese exposure if that supplier depends on Chinese feedstock. Her presentation traced several materials through raw inputs, refining and compound production to uses in servers and semiconductor equipment.

Lu said China’s export-control reviews had taken longer for Taiwan and Japan than for European countries, describing the pattern as targeted. She also noted that China processes many battery precursors and active materials domestically, making alternatives difficult to source. For semiconductors, she said restrictions on minerals such as gallium and germanium were followed by price increases. The leverage lies not only in control of raw minerals but also in refining and intermediate stages.

A slide on Taiwan’s imports illustrated how indirect exposure can persist. Some materials were imported mainly from Germany, Japan, the United States or South Korea rather than directly from China, while the feedstock or suppliers behind them remained connected to Chinese production. Lu’s broader point was that the immediate trading partner does not settle the question of supply-chain risk. She described China’s influence as reaching Taiwan through upstream inputs used by suppliers in other countries. This complicates diversification: a change in the country shown on an import invoice may not change the origin of the material or the availability of its substitutes.

The relevant supply chain is not organized around Taiwan alone. In response to a question about global sourcing, Lu described Taiwan and South Korea as important producers of battery cells and modules. She said Japan was a major source for upstream materials such as precursors and graphite-related inputs. She also pointed to US investment with Polish partners in precursor production as an effort to add capacity elsewhere. In her account, these countries can occupy different roles; building a resilient chain requires enough demand to make new capacity viable.

Recycling can help, but Lu cautioned against treating it as an immediate replacement for primary supply. Taiwan has recycling capabilities for some materials, including gallium and tungsten, but she described them as small-scale or unable to produce the final forms required by semiconductor manufacturing. The specific obstacles vary by mineral:

• Silicon: Lu’s slides described an established wafer-reclaim loop, but no route from scrap back to electronic-grade polysilicon. Wafer reclaim therefore does not close the loop for the material used to make new chips.
• Tungsten: A Taiwanese company was described as recycling carbide scrap into sodium tungstate and tungsten, but the output stopped at intermediate chemicals, much of which was exported for conversion abroad. A domestic carbide-tool loop was planned, not yet established.
• Rare-earth magnets: The slides described magnet recycling at a significant but incomplete stage: materials could be dismantled and processed into rare-earth-oxide-grade feedstocks, while refining into high-purity metals remained limited and costly.
• Gallium and germanium: Lu described recovery from gallium- and arsenic-containing waste streams, but noted the small volume of available scrap and the cost of handling arsenic. The presentation did not provide published recovery volumes.
• Lithium: Lu’s slide described no established large-scale recycling loop, citing limited end-of-life battery feedstock and the absence of a domestic refining pathway.

The limits are therefore both about scale and processing. Recycling may establish a secondary source, but Lu said it could not quickly close the gap in materials or forms that industry needs. She urged Taiwan and the United States to avoid reproducing dependence in emerging areas such as gallium nitride and silicon photonics, which she said were approaching larger-scale production. DSET’s proposed responses included supporting cross-border recycling of semiconductor waste and scrap wafers, and using policy to avoid another dependency trap in newer technologies.

Lu also drew a distinction between having a technical recycling capability and having a supply source capable of serving industrial demand. For example, she said tungsten recycling could produce high-purity powder, but further conversion into materials needed for semiconductor processing remained a challenge. Recycling is part of the response, not a substitute for diversifying primary supply and building processing capacity.

The speakers’ accounts point to different kinds of vulnerability, but not a single remedy. In advanced packaging, Chang’s concern is that a strategic capability is moving into a part of the manufacturing chain where Chinese firms already have industrial depth. His proposed response is to make controls follow the architecture and capabilities that matter, while pairing restrictions with qualified alternatives and a clear view of supplier ownership and sub-tier exposure.

In drones, Duerr’s central constraint is the gap between a fast export surge and a dependable production base. Demand is driven partly by wartime orders, while Taiwan’s higher costs and uncertain domestic procurement make long-term investment difficult. Certification, testing and international partnerships can open markets, but they do not substitute for recurring orders or resolve component dependencies. The proposed “volume-first” approach treats components as a route to build capacity and learn from field use, even as Taiwan pursues finished platforms and co-production.

For minerals, Lu described a different kind of leverage: delays and price premiums can weaken production without a formal cutoff. Diversifying suppliers is not enough if upstream feedstock or processing still runs through China; nor can recycling immediately replace those flows. Here, preparation means developing alternative processing and secondary supply while accepting that non-Chinese inputs may be more expensive.

The cooperative task is consequently broader than shifting production from one country to another. Lu’s battery example distributes roles among Taiwan and South Korea in cell and module manufacturing, Japan in some upstream materials, and new investment with partners such as Poland in precursor production. Chang described a similar division of labor in robotics: Taiwan has strengths in chips, while Japan and Europe are strong in components such as actuators, sensors and motors; in his view, the United States would need to lead coordination around complete systems. Duerr’s account adds that emerging drone suppliers need allies to buy from them before they can reach lower costs.

The unresolved tension is between building capacity and paying for it. Taiwanese drone and battery suppliers may offer alternatives, but Duerr and Lu both described meaningful cost disadvantages. Restrictions can redirect demand, but Chang warned that measures imposed before qualified substitutes exist can also hurt firms that still depend on the restricted inputs. Procurement, qualification support and coordination can help create an alternative market, but the discussion did not treat those tools as quick fixes.

Resilience, in these accounts, requires knowing where capability, ownership and feedstock sit; supporting suppliers before they reach efficient scale; and tailoring policy to the point of vulnerability. A non-Chinese production location may reduce one exposure while leaving another intact. The practical question is not only where a finished product is assembled, but whether its essential components and materials can still be supplied, qualified and expanded when conditions change.