Chip Wars: How TSMC and Samsung Will Navigate US-China Tensions

The Semiconductor Chessboard: Why 2026 Is Different

The global semiconductor industry is no longer shaped by pure market competition — it is increasingly a function of geopolitical negotiation. US export controls on advanced chipmaking equipment to China, tightened progressively since October 2022 and expanded again in 2024, have forced the world's two largest memory and foundry giants into uncomfortable strategic pivots. Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung are not merely adjusting supply chains; they are effectively being asked to choose sides in a technological cold war while shareholders demand growth.

TSMC, which manufactures an estimated 60% of the world's most advanced logic chips, finds its geography — 90% of production concentrated in Taiwan — transformed from a cost advantage into a political liability. Samsung, the South Korean conglomerate that once dominated memory markets, is fighting on two fronts simultaneously: navigating Western trade policy while scrambling to regain competitiveness in High Bandwidth Memory (HBM), the critical component powering Nvidia's AI accelerators.

The stakes are measured in hundreds of billions of dollars. Global semiconductor market revenue has grown to roughly $660 billion annually, with AI-related demand now the primary growth driver. But this growth is fragmenting geographically. The CHIPS Act has committed over $50 billion in US subsidies to reshore manufacturing. China is investing an estimated $150 billion in domestic semiconductor self-sufficiency. Caught in the middle are TSMC and Samsung — companies whose technologies are irreplaceable but whose markets are being actively partitioned.

TSMC's Geopolitical Balancing Act

TSMC's response to mounting geopolitical pressure has been the most expensive realignment in corporate history. The company is investing an estimated $65 billion in Arizona alone — not a satellite fab, but a full-scale advanced manufacturing campus intended to produce 3nm and eventually 2nm chips on US soil. Production at the first Arizona fab began in late 2024 with N4 process technology, with the more advanced N2节点 scheduled to come online toward the end of the decade.

This is not philanthropy. TSMC has secured billions in CHIPS Act grants and tax credits, and its Arizona customers — Apple, AMD, Nvidia — have explicitly demanded US-based supply chains for sensitive products. The Kumamoto, Japan fab (a $8.6 billion joint venture with Sony and Denso) represents a parallel strategy: diversify production across allied democracies without fully detaching from Taiwan's unmatched engineering density.

The financial impact is already visible. TSMC reported approximately $90 billion in revenue for 2024, with net margins consistently above 35% — among the highest in manufacturing globally. The Arizona expansion is designed to protect these margins by insulating TSMC's largest customers from supply disruption scenarios, whether triggered by trade escalation or worse.

But TSMC faces a constraint it cannot engineer around: the most advanced process nodes — the 2nm and eventual 1.4nm generations — require Chatham lithography systems that the Dutch government (under US pressure) will not license for export to China. This means TSMC's China-facing business, already constrained, faces a hard ceiling. The company has reportedly stopped accepting advanced-node orders from Chinese hyperscalers for chips below 7nm.

Sources: TSMC investor materials, SIA, industry estimates

Samsung's Two-Front War: Trade Policy and the HBM Crisis

Samsung's semiconductor division presents a more complex picture than TSMC's. While TSMC is a pure-play foundry with a single mission — make the most advanced chips possible — Samsung operates across memory, foundry, and consumer electronics simultaneously. This integration once seemed like a moat. In the AI era, it has become a vulnerability.

The immediate crisis is High Bandwidth Memory. HBM3E, the memory standard used in Nvidia's latest AI accelerators, is not merely faster DRAM — it is a co-designed component that sits physically adjacent to the GPU die. SK Hynix, Samsung's smaller Korean rival, captured first-mover advantage by qualifying HBM3E with Nvidia earlier and more reliably than Samsung. This created a supply-chain triangle: Nvidia designs the GPU, TSMC manufactures it, and SK Hynix supplies the memory. Samsung, despite having comparable technical capability, has struggled to pass Nvidia's qualification thresholds for the latest HBM generations.

The financial consequences are measurable. SK Hynix has reported consecutive record quarters driven by AI memory demand, while Samsung's semiconductor margins have compressed under competitive pressure. Samsung is reportedly investing aggressively to qualify its next-generation HBM products, but catching SK Hynix in a market where design-win cycles last 2–3 years is not a matter of spending alone.

On the trade policy front, Samsung faces constraints similar to TSMC but with additional Korean geopolitical complexity. South Korea's economy is deeply integrated with both the US security alliance and Chinese export markets. Samsung has announced a $17 billion fab in Taylor, Texas, but simultaneously maintains substantial legacy production in China for NAND flash and older process nodes — facilities it cannot easily wind down without destroying shareholder value.

Samsung's challenge is structural: it is fighting SK Hynix for AI memory dominance while navigating US-China restrictions, all while its foundry division struggles to match TSMC's yield rates at 3nm. No amount of capital can solve all three problems simultaneously.

Controversial Take: The CHIPS Act Is Creating Bubbles, Not Security

Here is an argument you will not hear from lobbyists: the CHIPS Act may be manufacturing geopolitical dependence rather than reducing it. TSMC's Arizona fabs are technically impressive, but the most advanced process nodes — the 2nm and 1.4nm generations that will define AI performance through 2028 — are still being developed and initially produced in Taiwan. Moving leading-edge R&D out of Hsinchu is not economically feasible at current scale; the ecosystem of suppliers, PhD talent, and tacit process knowledge is geographically anchored.

This means the US is spending tens of billions of dollars to build fabs that will, by design, lag Taiwan by one to two process generations for the foreseeable future. Arizona will produce chips that are advanced by American standards but mid-tier by TSMC's global standard. The political value is real — supply chain diversification — but calling it "reshoring leading-edge manufacturing" is marketing, not engineering reality.

Meanwhile, the aggressive US export controls are accelerating China's domestic substitution faster than Washington appears to have modelled.SMIC has reportedly produced 7nm-class chips using older DUV lithography through multi-patterning — an inefficient but functional workaround. Chinese equipment makers are years behind ASML, but in a market as large as China's, even imperfect domestic capacity can absorb enough demand to reduce Western leverage over time.

Practical Playbook: What Operators Should Watch

For operators managing technology procurement or semiconductor-dependent product lines, three variables matter above all others in 2026.

First, watch lithography equipment export licences. The Dutch government's decisions on ASML's advanced EUV systems are the single most consequential bottleneck in advanced semiconductor manufacturing. Any further tightening of export rules directly constrains what TSMC and Samsung can build in China — and by extension, what Chinese companies can build domestically.

Second, monitor HBM qualification announcements. If Samsung regains Nvidia qualification for its latest HBM products, the memory market dynamics shift rapidly. If SK Hynix maintains its design-win lock through the next GPU generation, Samsung's semiconductor division faces a multi-year margin compression scenario that could force restructuring.

Third, track CHIPS Act disbursement timing. The gap between announced subsidies and actual cash transfers has been significant. TSMC and Samsung have both signalled that the pace of US fab construction is partly contingent on reimbursement schedules. Delays could push back production timelines by 12–24 months, affecting supply expectations for 2028–2029.

What Happens Next: Fragmentation or Bifurcation

The most likely trajectory through 2028 is not decoupling — total separation is economically impossible given China's share of global electronics manufacturing — but bifurcation. One semiconductor ecosystem will serve the US, Europe, Japan, and their aligned markets, built around TSMC Arizona, Samsung Texas, Intel's foundry ambitions, and European efforts through STMicroelectronics and GlobalFoundries. A parallel ecosystem will serve China and non-aligned markets, centred on SMIC, Hua Hong, and increasingly domestic Chinese equipment suppliers.

TSMC and Samsung are the only companies with the technical capability to operate in both ecosystems, and both are being pushed by policy to choose. Samsung's dual Exposure — memory leadership plus foundry ambitions plus Chinese production — makes it the more vulnerable of the two. TSMC's narrower focus and earlier diversification investments give it more strategic flexibility, but its geographic concentration in Taiwan remains a risk no amount of Arizona spending can fully hedge.

The Bottom Line

The semiconductor industry in 2026 is not merely navigating trade tensions — it is being actively reorganised by them. TSMC's $65 billion Arizona bet and Kumamoto expansion are defensive moves designed to protect customer relationships under political pressure. Samsung's simultaneous struggle with HBM qualification, US fab construction, and Chinese legacy operations illustrates the operational cost of geopolitical dual exposure.

For operators and investors, the core insight is that semiconductor supply chains are no longer optimisable purely for cost or performance. Geographic risk, export-licence timelines, and geopolitical alliance structures are now first-class procurement variables. The companies that adapt procurement and inventory strategies to this reality will maintain continuity. Those that treat geopolitics as a temporary distortion will find themselves exposed at exactly the wrong moment.