There’s a scenario that technology industry leaders, defense analysts, and supply chain executives think about more than they discuss publicly: what happens if Taiwan’s semiconductor manufacturing is disrupted? Not “what happens to Taiwan” politically — that’s a separate, complex discussion — but specifically: what happens to the world’s technology supply if TSMC’s fabs go offline?
The answer is deeply uncomfortable, and the infrastructure decisions being made now — CHIPS Act investments, fab diversification, advanced packaging in new locations — are the industry’s response to that discomfort.
The Concentration Problem
TSMC (Taiwan Semiconductor Manufacturing Company) manufactures approximately 90% of the world’s most advanced semiconductors (3nm and 5nm process nodes). There is no close second. Samsung has advanced node capability but significantly lower capacity and yield rates. Intel is working to catch up but is years behind TSMC on process technology.
Every major chip in the products you depend on is likely made by TSMC:
- Apple’s M-series chips (M3, M4): TSMC 3nm
- NVIDIA’s A100/H100 AI accelerators: TSMC 4nm/5nm
- AMD’s Ryzen and EPYC processors: TSMC 5nm/4nm
- Qualcomm’s Snapdragon chips: TSMC 4nm
- Most advanced networking ASICs: TSMC
The concentration isn’t just in final chip manufacturing. The equipment to make leading-edge chips is also highly concentrated: ASML (Netherlands) is the only manufacturer of EUV lithography machines, and those machines require hundreds of components from specialists in Germany, Japan, and the US. The supply chain is a web of irreplaceable nodes.
Why This Matters for Technology Infrastructure
For anyone running significant infrastructure, the dependencies run deeper than you might think:
Servers and compute: Every modern server CPU (Intel Xeon, AMD EPYC) is either TSMC-manufactured or uses TSMC-manufactured components. GPU clusters for AI are almost entirely TSMC.
Networking equipment: The silicon in your switches and routers comes from the same supply chain. Broadcom, Marvell, and Cisco’s custom ASICs are TSMC customers.
Storage: NAND flash and DRAM controllers have their own supply chain, primarily in South Korea (Samsung, SK Hynix) and Japan (Kioxia), which is somewhat diversified from Taiwan but not immune.
Time horizon: A modern semiconductor fab has 8-12 month lead times for manufacturing equipment and 3-6 months minimum lead time for chip orders at scale. An acute disruption would mean no new chips for 12-18 months minimum — no restoring capacity in the near term.
TSMC’s Defenses: The “Silicon Shield”
Taiwan’s strategy has been described as a “silicon shield” — the idea that being economically indispensable to every major power (US, China, Europe, Japan) provides security through mutual dependency. If TSMC’s fabs were destroyed, it would harm China’s economy as profoundly as anyone else’s. Chinese companies buy enormous quantities of TSMC chips.
This deterrent logic has merit, but it’s not guaranteed to hold. Blockade scenarios (which would disrupt chip exports without destroying manufacturing capacity) and scenarios where political decisions override economic logic are both concerning possibilities.
Diversification: What’s Actually Happening
The industry is moving — slowly — to diversify:
TSMC’s Own Expansion
TSMC has announced and is building fabs in:
-
Arizona (US): $65 billion investment announced through 2028. First fab producing N4 process (4nm) chips since 2024. Second fab (N3/3nm) targeting 2026-2028. Third fab (N2/2nm) announced. However: manufacturing costs in Arizona are estimated to be 50-100% higher than in Taiwan, and yield ramp has been slower than expected.
-
Japan (Kumamoto): First fab opened 2024 with 22nm/28nm processes. Second fab targeting N6 (6nm) planned. This serves Japanese automotive and electronics companies.
-
Germany (Dresden): 10nm-class specialty fab planned for 2026-2027, focused on automotive and industrial chips.
Intel Foundry Services
Intel is rebuilding its manufacturing credibility as IFS (Intel Foundry Services), targeting leading-edge nodes. Intel 18A (competitive with TSMC N2) is the target process. Results in 2025 have been mixed — Intel has made technical progress but customer adoption has been slower than hoped.
Samsung
Samsung has aggressive foundry goals and capacity in South Korea and Texas, but has struggled with yield and customer wins at leading-edge nodes compared to TSMC.
The Honest Timeline for Diversification
The gap between “announced” and “producing at scale” is measured in years, not months. A realistic assessment:
2026-2028: TSMC Arizona N3 and N4 at partial capacity, high cost. Still heavily dependent on Taiwan for most leading-edge volume.
2028-2030: Intel 18A potentially competitive, attracting some customers. TSMC Arizona approaching meaningful volume. But Taiwan remains dominant.
2030+: More genuine diversification of supply, but Taiwan will remain a major hub for the foreseeable future.
The CHIPS Act timeline and investments are real, but they’re solving a 2030+ problem, not a today problem.
What Technology Leaders Should Be Doing
For technology executives and infrastructure leaders, the semiconductor concentration risk should be in your planning even if you can’t solve it:
Inventory strategy: For critical hardware, maintain longer inventory buffers than typical. The pandemic supply chain crunch demonstrated that “just in time” doesn’t work when supply chains fail. For servers and networking gear, 12-18 months of buffer inventory is worth the carrying cost for critical systems.
Multi-vendor architecture: Where technically feasible, avoid hardware architectures that lock you into a single chip vendor. This is constrained (NVIDIA is dominant in AI training) but worth planning for where possible.
Cloud as hedge: Cloud providers have the scale to maintain inventory and manage supply chain shocks that most organizations can’t. A hybrid approach where you maintain on-premises critical capacity and have cloud runways is more resilient than either extreme.
Scenario planning: Run a tabletop exercise: “Taiwan supply chain is disrupted for 18 months. What breaks first? What can we not replace?” The exercise is clarifying even if the scenario is unlikely.
Software efficiency: Every optimization that reduces your compute requirements reduces your exposure. If you can do the same work on 20% fewer servers, you have more runway in a hardware shortage.
The Energy Dimension
Advanced semiconductor fabs consume enormous amounts of electricity and ultra-pure water. TSMC’s Hsinchu fabs are major electricity consumers in Taiwan. Climate-related events (droughts affecting cooling water) have already affected Taiwan’s chip manufacturing in recent years.
As AI demand drives explosive growth in chip production, the energy and water requirements are growing. Climate risk adds another dimension to an already complex supply chain risk picture.
Conclusion
The Taiwan semiconductor concentration is the infrastructure risk that underlies every other technology risk. It’s not a reason to panic — the probability of acute disruption in any given year is debated, and the economic incentives against disruption are real. But the concentration is extreme, the diversification timeline is long, and the consequences of disruption would be severe.
For technology professionals: understand your hardware dependencies, build reasonable inventory buffers, and factor supply chain risk into major infrastructure decisions. The CHIPS Act investments and TSMC’s global expansion are the right long-term responses, but they’re not fast enough to protect against near-term risk.
This is a problem the industry is aware of and working on. It’s not solved.
