Key messages[1]
- With Pax Silica, the US focuses its semiconductor diplomacy on the technological frontier while keeping markets open for previous chip generations. Access to global markets supports Nvidia, because it helps finance the jump to the next architecture and thus sustain an 18-month ‘moving gap’ vis-à-vis China.
- The alliance creates an internal hierarchy within the US-led system by separating two types of territories: allies and clients. Allies are part of the stack and build technological, financial and security interdependence. Clients can access chips and services, but through a transactional relationship and under rules set by Washington, which can arbitrate access to the system.
- Access to the stack requires large-scale capital to coordinate mining, refining, manufacturing, logistics and design within a single industrial architecture. No country can cover all these stages without a system of co-investment, risk insurance and financial flows. That umbrella operates under political, monetary and regulatory rules defined by the US.
- The EU remains a guest and not a signatory, while the US prioritises bilateral relations with key European countries, starting with the Netherlands, given its control of the lithography machinery bottleneck through ASML. This fits in with the US National Security Strategy, which sees a fragmented Europe as a more manageable space for US interests. The lack of a cohesive semiconductor industrial policy reduces Europe’s bargaining power.
Analysis
After the Pax Romana, the Pax Britannica and the Pax Americana comes Pax Silica. The Pax Silica alliance has been formed by the US, Japan, the Republic of Korea, Singapore, the Netherlands, Israel, the United Arab Emirates, the UK and Australia. Territories with guest status include Taiwan, the EU, Canada and the OECD. The founding declaration of Pax Silica was signed on 11 December 2025 in Washington DC, at a summit convened by the US Department of State. Its aim is to set a political and material perimeter around the semiconductor sector, which underpins the development of artificial intelligence and, with it, a large share of economic and military competition among major powers.
Over the past decades, the microchip value chain has become highly distributed and specialised. The extraction and refining of raw materials are concentrated in China, Australia, Chile, Peru, the Democratic Republic of the Congo, South Africa, Canada, Norway and Brazil. Chip design is dominated by US companies such as Nvidia and AMD and relies on electronic design automation tools mainly controlled by US firms like Synopsys and European ones such as Siemens. The manufacturing of advanced wafers is concentrated in Taiwan and the Republic of Korea, with logic chips mainly produced by TSMC and memory chips by Samsung and SK Hynix. Japan leads in the production of ultra-pure materials and critical machinery, while the Netherlands holds a near-monopoly in lithography through ASML. Packaging and testing are carried out mainly in China, Malaysia, Vietnam, the Philippines, Singapore and Mexico.
Across this chain there are more than 50 points where a single region controls over 65% of the global market. With such a distributed architecture, the supply chain is highly exposed to political and geographic risks. The concentration of advanced production in Taiwan in particular, and the complexity of logistics in general, have become a source of vulnerability in scenarios of conflict or disruption, such as the pandemic or the recurring tensions in the strait separating the island from China. Control over the semiconductor stack is therefore, beyond a way to manage supply-chain risk, a sine qua non condition for leading the global race for artificial intelligence.
In recent years, the US had already been deploying a multi-instrument strategy to re-order the global semiconductor chain, both with the countries now visible in the Pax Silica picture and with others that are not. Export controls, cross-investment bans, industrial subsidies, technology alliances, mining agreements and rules to steer capital flows have all been put in place. The CHIP 4 alliance, mining agreements with Australia, Argentina or Japan, export alignment with the Netherlands and Japan, and industrial pacts with Taiwan and the Republic of Korea have gradually outlined an architecture of agreements around an ‘allied’ stack.
Pax Silica is a non-binding declaration of intent. It should be read mainly as a signal to the rest of the world, since it introduces no new semiconductor policies but rather gives an epic, ‘Trump-style’ framing to what was already under way. Emilio García, in his weekly newsletter Chips y Poder, describes it as a solemn opening of a silicon curtain, with the US asserting its role as global regulator of the supply chain on one side of that divide.
What message is the US sending with Pax Silica to the countries included and to those left out?
Message 1: ‘we focus on the frontier’
The US assumes and signals implicitly through Pax Silica that at the current stage of semiconductor development the key issue is to focus on the technological frontier, that is, on the chips that make it possible to run the most advanced AI models. Washington no longer sees it as essential to try to slow down its competitors. Instead, it prioritises its own leadership and the expansion of markets to keep financing frontier development. The alliances it will privilege are therefore those that provide control over strategic bottlenecks, firms that dominate key technologies and territories that concentrate critical capabilities.
The December 2025 decision, when the US authorised the export of the H200 AI chip to approved Chinese customers, can be read as a signal of the ‘moving gap’ doctrine. This gap at the technological frontier of semiconductors has changed over time, but Washington’s objective remains to keep China in a position of permanent lag. At the start of the Biden Administration, the goal was to maintain an advantage of four to five years over China. Today, that margin is closer to between six and 18 months.
The new step in this doctrine is to allow China access to technologies that have already been surpassed by a new generation. Thus, when China finishes mastering a given technology, such as the H200 today, the US is already deploying the next architectural leap, such as Rubin, the successor to the Blackwell family. This mechanism sustains the industrial investment cycle, because revenues from the sale of earlier generation chips finance the development of the next ones and can also be backed by public instruments that follow the US Chips Act.
The most advanced US chips offer higher tensor compute capacity and higher HBM memory capacity than their Chinese equivalents. Tensor cores, which are the units that run the matrix operations on which AI models are trained and executed, determine how many parameters can be processed per second inside the chip. HBM memory, in turn, determines how many parameters can be kept close to computation without having to split the model across multiple accelerators. On both axes, the technological frontier is clearly on the US side.
Until now, the only Nvidia model that could legally be exported to China was the H20, a restricted version of the Hopper family. Authorising the sale of the H200 therefore represents a qualitative leap in Chinese access to AI compute. In practice, China was already gaining limited access to advanced chips through smuggling via the United Arab Emirates, Taiwan and other territories, or through Chinese capital data centres located in Malaysia and Singapore. However, the change of jurisdiction prevented those centres from being used for sensitive data or military applications. For that reason, it is reasonable to expect that the first H200 chips arriving in China will be prioritised for military uses.
The selection of Pax Silica countries has produced a visible shock in states that had aspired to be part of a central semiconductor alliance, in particular India and several ASEAN members. Over the past two years, India had signed technology agreements with the US, Japan and Australia within the Quad, as well as a memorandum of understanding with Micron for an assembly plant and investment commitments in chip manufacturing from Applied Materials. In parallel, countries such as Mexico, Vietnam, Thailand and Malaysia had been strengthening their role in assembly and testing, backed by US, South Korean and Japanese investment. This is not expected to change, at least for now.
However, their exclusion from Pax Silica is explained by the scale and sophistication of their industries. These countries do not yet have frontier capabilities in any critical stage of the chain and therefore do not qualify as strategic actors under this new logic. The bet by the US and its allies is that, in the AI race, semiconductor diplomacy must focus on the technological frontier. US officials compare Pax Silica to the G7 of the industrial era, a coordination platform among those who control the most advanced segments of production. The logic is that future economic power will be decided in those nodes of maximum sophistication.
This opens a window that China could use to consolidate its technological presence in the Global South. For example, as Apple seeks to decouple the production of iPhones and other products from China, it has opened plants in India that end up being subsidiaries of Chinese conglomerates operating through local affiliates, such as Luxshare, YUTO Packaging or Sunny Optical. In parallel, Chinese battery and electronics firms have been expanding their presence in Vietnam and Indonesia.
As a result, it is assumed that non-aligned countries, including China, will end up buying Chinese hardware solutions and adapting to their architectures. This would amount to a form of technical soft power. At the same time, other industrial sectors that rely on semiconductors but do not require cutting edge chips could be left outside US diplomacy, creating a peripheral environment in which China strengthens its economic and technological control over non-aligned countries.
Message 2: ‘markets for all, the frontier for allies’
Unlike most other productive sectors, this one, both in the tangible semiconductor industry and more broadly in compute-intensive sectors, requires very large capabilities in investment, infrastructure, industrial capacity and talent to reach competitive levels. For that reason, it is a sector that tends toward full control by one or two large players. The message here is that countries will not be able to benefit from leadership in the value chains unless they first define themselves politically.
Under the Biden Administration, the US activated a control architecture based on export restrictions, cross-investment bans and asymmetric bilateral agreements. The Netherlands, Taiwan, Japan and the Republic of Korea accepted, and continue to accept, a significant share of these conditions. Companies such as ASML, which had been selling advanced lithography equipment to China, were forced to halt those exports. The same happened with South Korean firms, which restricted the sale of DRAM and NAND memory and the upgrading of their plants in China after direct pressure from Washington. The US Foreign Direct Product Rule prevents them from selling to China goods that were produced with even a minimal amount of US technology.
Since the arrival of Trump, countries such as Israel and the United Arab Emirates have taken clear steps towards decoupling from China. In return, they secure a seat at the table that organises the technological frontier.
Pax Silica is a declaration of intent. What it does is to add a logic of geopolitical loyalty to the AI economy. It does not create or dismantle any instruments or agreements between states or with their firms, which in the end is what matters in semiconductor diplomacy.
Given the US assumption that the semiconductor sector is in a phase where you are either at the frontier or you lose, this second message is aimed at countries that still try to keep an equidistant position between the US and China. It seeks to convey that in semiconductors it is no longer possible to operate successfully without first choosing to align with a hegemonic technological power while decoupling from the other, with the US acting as the arbiter of how far that alignment goes on its side of the silicon divide.
Pax Silica introduces an explicit separation between two types of territories, allies and clients. Allies are those that form part of the stack and with which technological, financial and security interdependence is built. Clients can continue to buy chips, models and services, and in fact it is in everyone’s interest that this market remains as large as possible, but their relationship with the system is transactional and based on business-as-usual conditions. In the event of a disruptive situation, priority will go to allies, not to clients.
It may seem that Trump’s more transactional doctrine, compared with Biden’s approach of controlling the diffusion of technology, is more compatible with middle powers that seek to maintain good relations with both US and Chinese technology. However, Pax Silica shows that this equidistance is possible only as long as it does not put the security of critical technology supply chains at risk, and it remains largely subject to US discretion.
The Biden Administration articulated this logic through the small yard, high fence doctrine, whose goal was to protect a limited set of critical technologies through strict controls, while keeping the rest of the technology trade open. Trump takes this doctrine one step further. He has made the yard even smaller, now focused almost entirely on the AI and semiconductor frontier, and he has made the fence higher. At the same time, he has relaxed the commercial perimeter for second generation chips.
What does this mean for countries that are left outside Pax Silica? In practice, they will still have access to products. They will be able to buy chips, servers and AI models, just as China can today acquire certain Nvidia GPUs under specific licences, but always under conditions set by the US. However, this is a position of disadvantage in terms of economic security, because the alliance aims to control all levels of the supply chain and to block any external actor from gaining control over a bottleneck. It will seek to prevent the emergence of a new European ASML or a Taiwanese TSMC unless it is within a framework of close cooperation with the US and decoupling from China.
While this position could raise fears of renewed restrictions on chips and models for external countries, that risk declines as large technology firms gain greater weight as actors in technology policy. The state capitalism doctrine promoted by Trump works as a double-edged sword for the White House. On the one hand, the interdependencies that the world has built with US firms give Washington wide room to manoeuvre in global geoeconomics. On the other, it forces the state to negotiate with corporate interests that no longer respond docilely to the President and that are unlikely to give up access to certain markets.
The decision to allow the export of Nvidia H200 chips to Chinese providers is a clear example of this tension. It is neither an attack on nor a concession to Xi Jinping. It is a concession to Jensen Huang, Nvidia’s CEO. Nvidia knows that if it does not increase revenue in foreign markets, while large US hyperscalers design their own chips to reduce dependence, one of the main geopolitical assets of the US is weakened, namely Nvidia’s dominance and the CUDA architecture.
Nor is the argument that this opening seeks to make China addicted to the CUDA architecture convincing. China has already shown that it designs its systems with multichip architectures precisely to avoid dependence on a single supplier and that they can migrate across architectures within months.
Message 3: ‘access to the stack requires a US financial and trade umbrella’
No matter how good a country’s geology may be, no one can sustain a full AI supply chain on their own without access to very large amounts of capital. Owning minerals does not give you secure and sovereign access to the stack. The only card left for countries with large mineral resources, but weak industrial bases, is to become part of an ecosystem of co investment, conditionality and shared financial governance. That ecosystem operates under Washington’s political, monetary and regulatory umbrella.
Unlike in other production sectors, the financing architecture is not one directional, with the US acting as the sole funder. Pax Silica brings together countries with strong financial ecosystems of their own, each in its own way. Singapore is included as an Asian node for financial and logistics services, Israel for its venture capital base, the Emirates for the liquidity of their sovereign wealth funds and the UK as the main financial hub of European technology. In 2025 the UK received US$23.6 billion in venture capital, a figure equal to the combined total of France, Germany, the Netherlands and Spain, the four largest recipients within the EU.
This financial push, together with the inclusion of Australia and Canada, is Pax Silica’s bet to counter China’s dominance in the most sensitive links of technological mining. China accounts for 69% of global rare-earth mining, but more importantly for almost 90% of global refining, including a near monopoly over heavy elements such as dysprosium and terbium, which are essential for high-performance magnets. Even when the ore is extracted elsewhere, intermediate processing, for example into NdFeB magnets or specific grades of battery graphite, remains highly concentrated in China.
This dominance reflects decades of industrial policy, massive state investment and process scalability. In 2023, when China activated export controls on gallium and germanium, it showed that this dominance also serves as a lever of geoeconomic pressure.
Pax Silica proposes to respond through a strategy of simultaneous containment and construction, but it can only do so if it builds a functional network linking countries with mineral reserves, firms with extraction and processing capabilities and capital.
This new industrial architecture may involve:
- Financing refining plants in trusted jurisdictions under public private co investment models.
- Establishing buffering mechanisms against geopolitical disruptions, such as shared reserves and supply insurance.
In the short term, however, there is no way to replicate from scratch the scale of Chinese refining or its network of intermediate suppliers. Even if alternatives are built, residual dependencies will persist for at least a decade, in technologies, chemical inputs and know-how.
For that reason, Pax Silica treats capital as a tool of expansion and as a tool of exclusion. Those that do not enter this logic of shared co investment under aligned rules will not have preferential access to new flows. Without access to capital flows, they will be pushed to the margins of the semiconductor industry and, in turn, of the AI race.
Message 4: ‘the US blurs Brussels’ role in the supply chain’
The EU appears as a collaborator in Pax Silica, and only the Netherlands, home to ASML, is a formal member. This second tier role for the EU aligns with the US strategy of dealing with European countries on a bilateral basis rather than strengthening the EU as a unified actor. Although the choice of observer entities can be read in different ways, it is worth recalling that in its National Security Strategy the US government has already stated that a fragmented Europe offers more room for manoeuvre for US interests.
Over the past few years, the EU has tried to build a common semiconductor policy. Its main instrument has been the 2023 European Chips Act, which set the goal of raising Europe’s share of global production from around 10% to 20% by 2030. More than €43 billion have been mobilised through a mix of EU funds and national budgets, combining R and D funding, industrial investment support and a coordination framework among member states. European semiconductor boards, networks of competence centres and mechanisms to relax state-aid rules have been created.
However, implementation has taken place mainly through national policies, which have produced a limited collective success. Germany sought to attract advanced manufacturing by Intel through a subsidy package of more than €10 billion, a project that was ultimately cancelled. France and Italy strengthened STMicroelectronics through a large joint plant in Crolles focused on intermediate technologies. The Netherlands has centred its policy on protecting ASML, Europe’s monopoly in advanced lithography, even by accepting export restrictions aligned with the US. Spain launched its PERTE Chip with the aim of investing €12 billion to build domestic capacity in design, packaging and manufacturing. Austria, Belgium and Ireland have pursued similar strategies in their own niches. The final result has been a total investment of barely €80 billion in the sector, with a public funding multiplier well below what has been achieved in the US or Japan.
This pattern is visible across much of the European economy. Fragmentation and internal competition between countries weaken European competitiveness and make Europe more exposed to a deeper dependence on the US. Each government seeks to attract factories, R and D centers, and subsidies to its own territory.
In a sector such as semiconductors, which depends on the concentration of resources and the control of bottlenecks, this fragmentation has a direct cost. A Europe that acts as 27 separate industrial policies has less capacity to negotiate with actors such as the US, to set its own standards or to sustain an autonomous technological frontier.
Conclusions
Pax Silica should be read, above all, as a political message rather than as a new set of policies. The US is stating that the race for semiconductors and artificial intelligence will be decided at the technological frontier, and that operating in that space requires countries to accept an architecture of alliances, markets and capital flows shaped in Washington. Whether this bet will work for the US and its closest partners remains uncertain. All signs suggest that China will reach positions of supremacy in key segments of the semiconductor value chain within a few years, and the final outcome will depend on where the real needs of AI models move.
If the trajectory of AI remains tied to a moving frontier of ever larger and more expensive models, the semiconductor industry will tend towards bipolarity, or even unipolarity, dominated by those who control that frontier. If, instead, models take on a more commodity-like character, where a level of sophistication sufficient for most uses is reached and the pace of progress slows, other factors will gain weight, such as the ability to manufacture chips quickly and cheaply. In that scenario, China and its ASEAN partners, which have been left outside Pax Silica (except for Singapore), could take a stronger position. For Europe, the lesson should be the following: the sector tends towards concentration because of the scale of investment it requires. The EU will need, on the one hand, to keep building peer-to-peer alliances with hegemonic actors and middle powers and, on the other, to prepare for the next technological phase, centred on more modular and more efficient semiconductors.
[1] The author would like to thank Emilio García García for his comments and revisions.
