Taiwan’s National Security Bureau (NSB) confirmed in an intelligence report to lawmakers — obtained by Reuters in April 2026 — that China is running a systematic campaign to steal semiconductor manufacturing technology and recruit engineers from Taiwan Semiconductor Manufacturing Company (TSMC). The report identifies this as a state-directed effort to overcome U.S.-led export restrictions that have denied China access to the advanced chips powering the current AI race. China’s semiconductor manufacturing gap with TSMC runs roughly five to seven years — and Beijing has concluded the fastest path to closing it is not innovation, but extraction.
What Taiwan’s Intelligence Report Actually Says
The NSB’s report, presented to Taiwan’s legislature, describes a multi-vector campaign combining corporate espionage, talent recruitment, and supply-chain infiltration. Chinese state-linked entities have been systematically identifying TSMC engineers with access to process integration and advanced packaging knowledge, then approaching them through shell companies and third-country intermediaries designed to obscure Beijing’s involvement.
The document warns that significant recruitment activity is being conducted through ostensibly legitimate technology firms in Singapore, Malaysia, and the United Arab Emirates, according to Reuters. Offers range from two to five times current salaries, plus housing packages and research funding — for engineers willing to relocate, or simply share proprietary data. The NSB report does not name individual targets but describes a recruitment pipeline that has been active for at least two years.
The Three Technologies China Is Actually Targeting
Not all of TSMC’s capabilities are equally valuable to Beijing. The NSB report identifies three specific technology clusters as primary targets:
- EUV process integration: TSMC has developed unique process recipes for its 3nm and 2nm nodes that work with ASML’s extreme ultraviolet lithography machines — machines China cannot legally import under Dutch and U.S. export restrictions. China’s Semiconductor Manufacturing International Corporation (SMIC) is currently limited to 7nm-class production without EUV. Semiconductor analyst firm TechInsights estimates TSMC’s yields on advanced nodes run 20–30 percentage points higher than SMIC’s best comparable process.
- Advanced packaging (CoWoS and SoIC): TSMC’s Chip-on-Wafer-on-Substrate packaging is the manufacturing process that makes NVIDIA’s H100 and H200 GPUs possible at volume. This technology cannot be reverse-engineered from a finished chip — it lives entirely in TSMC’s fab processes and the working knowledge of its engineers.
- High-NA EUV readiness: TSMC is preparing for the next generation of ASML’s high-numerical-aperture EUV tools, targeted for volume production in the 2025–2026 window. Getting ahead of China on this transition would extend the capability gap by an estimated three to four additional years.
How China’s Recruitment Pipeline Operates
The approach is methodical, not blunt. Chinese state entities work through layers of cutouts: headhunting firms in Southeast Asia approach TSMC engineers on LinkedIn or through mutual contacts, initially framing opportunities as roles at “global semiconductor startups.” The Chinese connection surfaces only later, according to the NSB report and corroborating reporting from The Financial Times.
Engineers are also targeted after they leave TSMC. Taiwan law bars former TSMC employees from joining Chinese chip firms for two years after departure, but enforcement collapses once an engineer relocates. SMIC, Hua Hong Semiconductor, and multiple state-funded research institutes in Shenzhen and Shanghai have documented histories of hiring former TSMC staff, according to Bloomberg’s review of trade filings. The talent transfer is a slow leak — but a continuous one.
The U.S. Containment Strategy and Its Structural Weakness
The espionage campaign exists precisely because the U.S. export control architecture has been effective — as a hardware embargo.
Since October 2022, the U.S. Bureau of Industry and Security (BIS) has progressively tightened semiconductor export restrictions targeting China. The controls cover chips, fabrication equipment, and the foreign direct product rule — which requires any chip manufactured anywhere using U.S. technology to comply with export restrictions regardless of where it was made. The result: NVIDIA’s H100, H200, and Blackwell-series GPUs are effectively unavailable to Chinese buyers. ASML is blocked from shipping EUV systems to China. Applied Materials, Lam Research, and KLA are restricted from upgrading Chinese fabs beyond defined capability thresholds.
The structural weakness is this: it is a hardware embargo, not a knowledge embargo. If China extracts process knowledge from TSMC engineers, the physical equipment becomes a secondary obstacle. As the race to build AI infrastructure across geopolitically sensitive borders has shown, physical hardware is increasingly the bottleneck in the compute race — and the engineering knowledge of how to use that hardware efficiently may matter as much as the hardware itself.
China’s Domestic Progress — and Its Ceiling
China has committed more than $150 billion to domestic semiconductor manufacturing since 2014 through its National Integrated Circuit Fund — known as “Big Fund.” The results are genuine but structurally capped.
SMIC’s 2022 production of Huawei’s Kirin 9000S processor demonstrated a 7nm-class capability achieved without EUV, using multi-patterning workarounds. But that process runs at low yields, high cost, and has not scaled to the volumes required for AI training clusters. Building an H100-equivalent chip through SMIC today would produce yields estimated at 40–50% below TSMC’s, according to semiconductor research firm SemiAnalysis — making the economics of large-scale AI model training effectively unworkable at competitive cost. The gap at 2nm is wider still.
The espionage campaign is, in part, Beijing’s acknowledgment of that ceiling. Fifteen years of industrial policy have not closed the gap. Recruitment and theft are the accelerant China is reaching for instead.
What a Successful Technology Transfer Would Mean
The consequences of a successful breach are specific, not abstract.
If China acquires TSMC’s EUV process integration recipes and CoWoS packaging know-how, state-funded foundries could realistically compress the manufacturing gap from five to seven years to two to three — within a decade, potentially producing domestically the H100-class accelerators currently under embargo. That is enough compute to supply Chinese AI labs training the next generation of frontier models without dependence on smuggled chips or diminished-capability workarounds. The race to control who builds those frontier models — visible in consolidation moves across the AI industry — makes chip access not just a manufacturing question but a structural one about who shapes AI’s next decade.
The U.S. compute lead over China — measured in accessible high-performance AI accelerators — is estimated at three to five years by the Special Competitive Studies Project. A Taiwan fabrication breach would compress a strategic advantage the U.S. has spent billions constructing. The implications extend well beyond AI: advanced chips underpin hypersonic weapons guidance systems, electronic warfare platforms, autonomous drone navigation, and satellite intelligence processing. This is not a commercial story with geopolitical overtones — it is a geopolitical story with a commercial surface.
The broader pattern — where the most sensitive intellectual property resides not in classified government systems but in private company engineering teams — echoes incidents like Anthropic’s accidental exposure of Claude agent source code, which demonstrated how concentrated and vulnerable advanced AI knowledge becomes when it lives with a small number of engineers rather than inside hardened institutional infrastructure.
Taiwan’s Exposure and the Countermeasures Being Taken
TSMC sits 180 kilometers from China’s coastline and produces approximately 90% of the world’s most advanced AI chips. Its exposure is not incidental — it is structural.
Countermeasures are being deployed. TSMC has implemented enhanced employee monitoring, tightened data access controls, and partnered with Taiwan’s Ministry of Justice Investigation Bureau on insider-threat detection programs. Taiwan’s intelligence services arrested at least three individuals in 2025 on charges connected to semiconductor espionage, per government statements. TSMC’s Arizona fabs — now producing 4nm chips, with 2nm capacity scheduled — represent a geographic hedge, though they account for less than 5% of TSMC’s total advanced-node output.
The intelligence report reaching legislators is itself a deliberate escalation. Taiwan is pressuring democratic allies — the U.S., Japan, the Netherlands — to treat chip espionage with the same urgency applied to military deterrence. The same U.S.-led alliance architecture that restricts hardware exports needs, in Taiwan’s framing, an equivalent counterintelligence commitment. Whether that coordination materializes before a successful technology transfer does is the operative question in the AI chip race right now.
The semiconductor supply chain is not just an economic asset — it is an active intelligence target. China has calculated that stealing chip knowledge is faster, cheaper, and more reliable than building it from scratch, and Taiwan’s intelligence report confirms that campaign is structured, state-directed, and ongoing. For every shipment blocked by an export control, there is a headhunter somewhere between Singapore and Taipei working the other side of the equation.
