Why Semiconductors Are the Foundation of the Modern Economy
Semiconductors — integrated circuits that process and store digital information — are the enabling technology for virtually every product and service in the modern digital economy. The smartphone in every pocket, the cloud server processing every web request, the car managing every safety system, the industrial robot executing every manufacturing step, and increasingly the AI system making consequential decisions in every sector all depend on semiconductors. There is no meaningful substitute for semiconductor technology in any of these applications, and the performance of semiconductor technology — measured in transistors per chip, power consumption per operation, and memory bandwidth — directly determines the performance ceiling of everything built on top of it.
The semiconductor supply chain complexity that most people who use semiconductor-powered devices do not appreciate: the modern leading-edge semiconductor takes thousands of steps to manufacture, uses more than 50 distinct chemical elements, requires manufacturing equipment that itself takes years to build and costs hundreds of millions of dollars per unit, and can only be produced at the highest performance levels by a handful of fabs globally. The most advanced logic chips — the ones that power the latest smartphones and AI accelerators — can only be manufactured by Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung at the leading edge, with TSMC producing the majority of the world’s most advanced chips. This geographic and corporate concentration of leading-edge semiconductor manufacturing has become one of the most significant geopolitical concerns in technology policy.
The Design-Manufacturing Split: Fabless vs IDM
The semiconductor industry structure that has enabled the specialisation that drives much of the industry’s innovation: the separation between chip design (creating the logical specification and physical layout of the chip) and chip manufacturing (the physical process of fabricating the chip in a silicon wafer). The integrated device manufacturers (IDMs) like Intel and Samsung perform both functions; the fabless companies design chips without owning manufacturing facilities and outsource manufacturing to contract fabs. The fabless model — used by Qualcomm, AMD, NVIDIA, Apple, and most of the semiconductor companies that have produced the most successful chips of the past decade — allows chip designers to focus entirely on design innovation without the capital requirements of semiconductor manufacturing.
The fabless model’s dependence on contract manufacturers that most clearly reveals the sector’s strategic vulnerability: the leading-edge fabless chip design capabilities are distributed across many companies in many countries, but the manufacturing capability to produce those designs at the leading edge is concentrated in two companies (TSMC and Samsung) in two countries (Taiwan and South Korea), with TSMC producing the large majority of the most advanced chips. The geopolitical scenario in which this manufacturing concentration is disrupted — whether by cross-strait conflict, natural disaster, or other event — has no short-term mitigation because building a new leading-edge semiconductor fab takes five to ten years and tens of billions of dollars even under the most favourable conditions.
The Geopolitics of Semiconductors
The semiconductor export controls that have most reshaped the industry’s competitive dynamics: the US restrictions on semiconductor technology exports to China, expanded significantly in October 2022 and subsequently tightened further, which restrict the export of leading-edge semiconductor manufacturing equipment, advanced chips (particularly AI accelerators), and chip manufacturing software to Chinese entities. These restrictions are intended to prevent China from developing the leading-edge semiconductor manufacturing capability that would enable its AI development and military applications — and they represent a significant departure from the decades-long policy of largely unrestricted technology trade.
The China semiconductor response to export controls that most clearly reveals the strategic nature of the competition: China’s dramatically increased state investment in domestic semiconductor development, including subsidies to domestic chip manufacturers, accelerated development of domestic alternatives to restricted equipment, and sustained efforts to recruit semiconductor talent from other countries. SMIC, China’s largest domestic contract chipmaker, has continued to advance its manufacturing process capability despite equipment restrictions, though it remains several generations behind the leading edge achievable with unrestricted access to ASML’s extreme ultraviolet lithography equipment.
The AI Chip Investment Wave
The semiconductor investment category that has most captured market attention and investor capital since the launch of ChatGPT in late 2022: the AI accelerator chips designed specifically for the matrix multiplication operations that underpin training and inference for large neural networks. NVIDIA’s dominance of this market — its H100 and H200 GPUs command prices above thirty thousand dollars per unit and waiting lists of months even for the largest technology companies — has produced the most rapid revenue growth in the company’s history and has elevated it to the highest market capitalisation of any semiconductor company in history.
The AI chip alternatives to NVIDIA’s GPU dominance that are emerging from multiple directions: AMD’s competing GPU products (which are achieving meaningful adoption among hyperscale customers, particularly for inference workloads where NVIDIA’s architectural advantages are smaller), the custom AI accelerators that hyperscale cloud providers are developing to reduce their dependence on NVIDIA (Google’s TPU, Amazon’s Trainium and Inferentia, Microsoft’s Maia), and the specialised inference chip startups (Groq, Cerebras, Tenstorrent) that are optimising for specific inference workloads. The AI chip market of 2027 will almost certainly be less concentrated around NVIDIA than the 2024 market, as the volume of investment flowing into alternatives scales up.
The Fab Investment Wave and Its Long-Term Implications
The government-sponsored semiconductor manufacturing investment programmes that are reshaping where leading-edge chips will be made in the next decade: the US CHIPS and Science Act (which provides approximately $52 billion in subsidies for domestic semiconductor manufacturing, with significant investments being made by TSMC, Intel, and Samsung in new US fabs), the European Chips Act (which targets 20% global market share for European semiconductor manufacturing by 2030), and equivalent programmes in Japan (which has attracted a new TSMC fab to Kumamoto), India, and South Korea. The collective ambition of these programmes is to diversify the geographic concentration of leading-edge semiconductor manufacturing away from Taiwan.
The timeline reality that most tempers expectations for the fab investment wave: semiconductor fabs take five to ten years from announcement to first production, and producing chips that are competitive with Taiwan’s fabs requires not just the physical facility but the deep expertise, supply chain relationships, and manufacturing process knowledge that TSMC has accumulated over decades. The TSMC Arizona fab that began production in late 2024 initially targets less advanced process nodes than TSMC’s leading-edge Taiwan fabs, reflecting the genuine difficulty of replicating TSMC’s full capability in a new location. Geographic diversification of semiconductor manufacturing is achievable over the long term; the short-term strategic vulnerability created by Taiwan’s concentration of leading-edge manufacturing capability will not be resolved quickly regardless of the scale of current investment.