In February 2026, it became known about the Trump administration’s plans to exempt American tech companies, such as Amazon, Google, and Microsoft, from tariffs on the import of TSMC semiconductors.
This move is related to the Taiwanese company’s investment commitments of over $165 billion as part of a new trade agreement between the US and Taiwan.
This exemption is intended to support chip production in the US and at the same time allow large tech companies to seamlessly expand data centers for artificial intelligence without losing access to Taiwanese suppliers.
The geo-economic significance of this decision lies in changing the logic of dependencies, as Washington gradually moves away from Taiwan’s “silicon shield” model, which supplied American companies with semiconductors and chips.
It instead shifts key production links to U.S. territory, using access to the American market as a tool to force onshoring for tech companies and contractors.
The logic of “Too Big to Fail” in the case of Big Tech becomes the starting point for understanding Washington’s entire policy regarding exempting large companies from tariffs, as it concerns stabilizing the infrastructure on which the US economic, military, and technological architecture rests.
Cloud platforms, computing clusters, and data centers are gradually transitioning into the category of systemic assets, the failure of which creates a risk of chain destabilization of financial markets, defense contracts, and industrial chains, so the state begins to act preemptively, even before the sector formally recognizes a crisis.
The artificial intelligence sector is in a phase of capital trap, where investments in infrastructure amount to hundreds of billions of dollars per year, while economic returns remain deferred in time, and any sharp reduction in spending would mean losing technological momentum in competition with China.
In this logic, tariff exemptions cease to be a tool of trade policy and turn into a mechanism for stabilizing the sector, which allows freezing a potential liquidity crisis even before it manifests on corporate balance sheets.
The exemption from duties for Amazon and Microsoft on semiconductors produced by TSMC essentially acts as a hidden subsidy, reducing the cost of a key AI infrastructure component.
The support boosts the companies’ margins and allows them to continue building data centers even as the market grows nervous about the pace of spending and slow monetization.
This measure also acts as a preemptive rescue of the industry, carried out before the market formally recognizes the presence of a systemic problem.
The trade agreement, within which TSMC will invest approximately $165 billion, is significant because it goes far beyond industrial policy and means the gradual transfer of critical production competencies from Taiwan to the US.
As a result, a closed production cycle is formed, combining American chip design through Nvidia with a domestic energy base and localized production.
This reduces U.S. strategic dependence on external nodes and lowers Taiwan’s long-term geopolitical criticality, giving Washington more freedom in its strategic confrontation with Beijing.
Washington is simultaneously removing obstacles to accelerating the pace of developing next generations of chips, and the reason is the speed of the competitor.
The semiconductor development cycle does not allow pauses, as each new generation of processors—from the current 2-nanometer nodes to prospective 1.4-nanometer solutions and below—requires continuous scaling of production capacities and capital investments. These investments directly determine the speed of transition to the next technological cycle.
Stopping or slowing this process would mean handing the initiative to Beijing, which is consistently closing the technological gap.
China is no longer in the state of dependency it was in five years ago. SMIC is already mastering production at the 7-nanometer node without access to the most advanced lithographic equipment from ASML.
Chinese companies are working to develop their own lithography equipment, but the most advanced domestic SMEE system has only reached the 90-nanometer level, leaving a gap of several technological generations compared to Dutch monopolist ASML.
This forces SMIC to rely on complex workaround solutions using older DUV equipment to produce 7-nanometer chips.
Washington’s export restrictions have slowed Beijing but not stopped it: sanctions pressure has turned into an incentive for accelerated localization of the production chain.
Therefore, tariff exemptions for American tech companies and stimulation of TSMC investments on US territory are not a concession but an element of the technological race, in which the winner is the one who first reaches the next production horizon.
Washington is removing internal barriers to accelerate the deployment of computing power and increase the lead over China before Beijing can replicate the current generation of technologies already operating in the open market.
This race is not limited to semiconductors. The same logic of breaking away from China determines the pace of investments in artificial intelligence, space programs, and defense computing systems—and in each of these areas, slowing down means losing.
In the strategic dimension, such steps by Trump as exempting American companies from tariffs are a consistent three-level plan.
The first level involves freezing risks through tariff exemptions and regulatory benefits, which allows avoiding a sharp reduction in investments.
The second level consists of maintaining the sector’s access to cheap money through signals to investors that the state guarantees the systemic stability of the industry.
The third level is buying time needed for the transition of artificial intelligence from generative applications to industrial use, where computing power will begin to create new materials, energy solutions, and engineering systems capable of providing real productivity growth.
In this setup, private capital builds strategic infrastructure, while the state offers tariff protection, regulatory predictability, and financial stability, expecting that in a decade the global economy will pay American hyperscalers a technological rent for access to computing power, much like the world paid an energy rent for oil in the 20th century.
Alphabet’s record $20 billion debt issuance, with bids exceeding $100 billion, shows that investors are still willing to fund technological expansion, but the size of the funds raised highlights the industry’s growing reliance on cheap money to maintain its pace of development.
Thus, the plans to issue century bonds announced on February 9 signal an attempt to lock in long-term financing for infrastructure projects whose lifecycle may prove shorter than expected due to rapid changes in technological standards and uncertainty in the regulatory environment.
Investors are willing to finance technological expansion, but the financing structure itself indicates a change in the nature of the sector—AI infrastructure is beginning to be financed as a long-term public asset.
Plans to issue ultra-long bonds aim to secure cheap financing over a period longer than the pace of technological cycles, making the stability of these corporations crucial for pension funds, insurance companies, and the financial system as a whole.
Under such conditions, Big Tech transitions to the status of quasi-state structures: their platforms become the basic infrastructure of the economy, defense, and industrial design, and a potential crash of the AI market automatically turns into a national security crisis that the state cannot allow.
The problem is exacerbated by the fact that the current model of artificial intelligence development has built a construct in which the volume of investments significantly outpaces organic demand, and infrastructure accumulates debt burden faster than a stable cash flow from its use is formed.
The slowdown in growth of Microsoft’s cloud segment became a key signal for the market, as it was the cloud that was supposed to turn AI investments into a profitable model, and its dynamics showed a gap between the scale of spending and the speed of monetization.
Microsoft’s capital expenditures on data centers in the fourth quarter of 2025 reached $37.5 billion with annual growth of about 66%, which sharply increases risks if the return on investment remains low for several years.
Accordingly, the market is beginning to consider a scenario in which artificial intelligence agents turn part of corporate software into a standardized product, reducing expectations of future revenues and destroying the valuation logic of SaaS companies built on long-term margin expansion.
The planned expenditures of tech giants on AI infrastructure exceed $380 billion in 2025, but surveys of CFOs show no noticeable effect on productivity, decision-making speed, or cost structure of companies.
Despite the fact that 78% of large companies have invested in AI, only a small portion of businesses have been able to convert these costs into staff reductions or tangible revenue growth, which reinforces doubts about the speed of the technology’s economic return.
The comparison of AI to the end of the dot-com cycle is becoming clearer, as companies build computing power based on expectations of future growth, while business processes adopt AI as a support tool without major changes to the production economy.
Accordingly, the drop in Intel shares after a weak forecast became a reminder of the industry’s production limitations, where even high demand for server chips does not guarantee rapid scaling due to technological complexities and low yield percentages.
The Oracle situation demonstrates an extreme scenario of this model, where the bet on mass construction of data centers in partnership with OpenAI led to rapid debt growth exceeding $110 billion and the need to revise investment plans.
OpenAI’s financial difficulties underscore the structural problem of the sector, as model development requires constant increases in computing costs, while revenues from their use remain insufficient for self-financing further scaling.
The phase of general expansion of computing power is exhausting itself—the sector is entering a period of selection, where projects capable of converting computations into specific added value through cost reductions, new materials, and automation of engineering processes will survive.
It is this capability that will determine the winners of the next cycle of technological competition, in which China continues to close the distance. Under these conditions, tariff exemptions, regulatory relaxations, and protection of access to cheap capital work as tools for maintaining pace.
Losing one technological cycle gives Beijing two to three years to consolidate its own standards, supply chains, and export markets—a gap that cannot be compensated in the next cycle.
