When chief technology officers choose hardware infrastructure, they rarely think about Taiwan. When supply chain managers assess supply risk, semiconductors often appear as a technical item, not a strategic decision. This framing was wrong before 2020 and has become dangerously outdated in the wake of the pandemic, American export restrictions, and increased tensions in the Taiwan Strait. The chip war is no metaphor — it's a real competition between powers that is actively changing the cost, availability and access to computing hardware for companies around the world, including in Brazil.
The concentration that no one can replicate quickly
The advanced semiconductor supply chain has a characteristic that distinguishes it from virtually every other industry: extreme geographic and corporate concentration at points with no short-term substitute. TSMC, a Taiwanese company, manufactures approximately 90% of the most advanced chips in the world — those with a process node below 5 nanometers. There is no equivalent second option today. Samsung manufactures in smaller volumes and with lower yields in the most advanced processes. Intel is rebuilding its manufacturing capacity after years of delay.
ASML, a Dutch company, is the only manufacturer in the world of EUV (Extreme Ultraviolet) lithography machines — the equipment without which it is not possible to manufacture advanced chips. A single ASML machine costs between 150 and 300 million dollars, takes years to produce, and uses components from more than 800 suppliers in dozens of countries. Replicating this chain is not a matter of investment — it is a matter of decades.
Arm, a British company controlled by the Japanese group SoftBank, licenses the processor architecture that is in virtually all smartphones in the world and a growing fraction of servers. NVIDIA controls the market for GPUs for high-performance computing and artificial intelligence training with margins that reflect the absence of real competition.
American export controls and their cascading effects
In October 2022, the United States government implemented the most comprehensive semiconductor export restrictions to China in history. The rules prohibited not only the sale of advanced American chips to Chinese companies, but also the participation of American citizens in chip development projects in China, and the sale of manufacturing equipment — including ASML machines — to Chinese factories that produce chips above certain performance thresholds.
The restrictions were expanded in 2023 and 2024, progressively closing the loopholes. The immediate effect was to force companies like Huawei to seek chips with a less advanced process or develop domestic alternatives — Huawei launched in 2023 a 7nm chip manufactured by SMIC, the only Chinese company with advanced manufacturing capacity, demonstrating that the restriction delays but does not stop Chinese development.
For companies outside of China and the United States, the side effect of the restrictions is less obvious but just as real. Massive American investment in domestic manufacturing via the CHIPS Act — $52 billion in subsidies to build U.S. factories — is changing where chips will be produced over the next ten years. This affects pricing, lead times and availability for global buyers.
What does this change for those who buy hardware
For companies purchasing servers, workstations or devices at scale, the geopolitics of chips have already manifested themselves in variations in price and availability that do not follow the normal logic of supply and demand. The chip shortages of 2021 and 2022 were the most visible episode, but structural volatility did not disappear — it just became less acute temporarily.
GPUs for artificial intelligence workloads are the most critical case. Export restrictions on NVIDIA A100 and H100 chips — the most powerful available for training language models — created a parallel market with prices reaching triple the listed price. Companies that make AI infrastructure decisions without considering hardware availability and access risk are making incomplete planning.
Risk exposure varies by hardware type. Less complex generic chips — microcontrollers, memory chips, general purpose processors from previous generations — have a more distributed and less politicized supply. State-of-the-art chips for intensive computing have a supply that can be restricted by a political decision by a single government with global effect. The relevant difference for planning is: what part of the hardware needed to operate is in this second category?
Where is Brazil in this dispute
Brazil is not a producer of semiconductors in commercial volume and does not have the construction of advanced manufacturing capacity on its next agenda — the investment required for a competitive chip factory starts at 10 billion dollars for advanced processes, and the return takes decades. The Brazilian position is that of a buyer dependent on global chains, which creates specific vulnerability to supply shocks and restrictions that are not directed at Brazil but affect it through contagion.
Brazilian industrial policy for semiconductors has focused on chip design — Ceitec in Porto Alegre was the most ambitious project and was privatized after operational difficulties — and on attracting R&D centers from global companies. This is a more realistic entry point than advanced manufacturing, but it does not address the supply vulnerability.
For Brazilian companies that depend on cutting-edge hardware, the practical strategy is diversification of suppliers within what is available, long-term contracts with distributors to guarantee allocation in times of scarcity, and attention to the political calendar of the main producing powers — American elections and tensions in the Taiwan Strait have historically preceded the biggest supply shocks.
How to incorporate geopolitical risk into technology decisions
Semiconductor geopolitical risk is not treatable in the same way as conventional financial or operational risk — it is not directly hedged and is not accurately modelable. But it can be incorporated into architectural decisions in a practical way.
The first lever is to reduce dependence on high-end hardware where it is not needed. General-purpose cloud computing workloads already run on hardware managed by providers, shifting availability risk to them. The second lever is building architectural flexibility — designs that can migrate between hardware vendors with reasonable effort are more resilient to availability shocks than those that assume specific hardware. The third, especially relevant for those investing in AI, is to actively monitor the chip chain that powers your workloads — which processor, from which manufacturer, manufactured in which factory, under which regulatory regime. This information transforms abstract risk into a manageable variable.
Also read
- The new generation of chips: GPU, NPU, ASIC and RISC-V for those who decide
- Energy, the new oil of AI: whoever controls the resource controls the era
- The AI Energy Crisis: What Data Center Consumption Means for Infrastructure Decision Makers
- Digital sovereignty: when the place where your data lives becomes a strategic decision
- Inference chips and ASIC: when the specialized beats the generic
- Specialized chips and the end of the generic CPU era