The Global Compute Cartel
Export Controls and the Nationalization of Artificial Intelligence
I found myself reading a draft regulatory discussion circulating in Washington concerning the possibility that exports of advanced AI accelerators might soon require approval not merely for shipments to geopolitical rivals, but potentially for large scale deployments anywhere in the world. The proposal was framed in the technical language typical of export control frameworks. It spoke about licensing thresholds, cluster sizes, and compliance mechanisms. Yet the implication was unmistakable. The United States was considering a regime in which the construction of frontier AI compute infrastructure globally could fall under its regulatory oversight.
While going through the text, my mind wandered to a concept that at first glance has nothing to do with semiconductors or artificial intelligence. Treaty shopping.
Anyone who has spent time around international investment arbitration will immediately recognize the term. Treaty shopping refers to the practice whereby multinational firms structure their corporate ownership through specific jurisdictions in order to access protections under bilateral investment treaties. Over the last three decades the practice has produced some of the most elaborate examples of corporate nationality engineering ever attempted.
Consider the logic that emerged within international arbitration practice. Investment treaties typically define investors by their place of incorporation rather than by the nationality of their ultimate shareholders. If a company can route its investment through an entity incorporated in a jurisdiction that enjoys favorable treaty protections with the host state, that company gains access to arbitration mechanisms and legal protections that would otherwise be unavailable. Lawyers learned quickly that corporate nationality could be optimized.
The history of investment arbitration contains numerous examples. Energy companies operating in emerging markets routinely structure their investments through Dutch or Luxembourg holding companies in order to access treaty protections those jurisdictions maintain with host states. Firms have reincorporated, restructured shareholding chains, and created intermediate holding vehicles solely to secure access to arbitration tribunals such as the International Centre for Settlement of Investment Disputes. Entire corporate architectures have been designed not around operational efficiency but around the passporting of nationality.
Reading through Washington’s emerging AI export proposals inevitably raises a question that any practitioner steeped in the world of cross border structuring will immediately recognize.
Why could something similar not occur in the semiconductor industry?
If the United States intends to regulate the global flow of advanced AI chips, what prevents the firms responsible for designing those chips from relocating their legal domiciles or transferring their intellectual property to entities beyond U.S. jurisdiction? After all, the semiconductor ecosystem is already profoundly international. The most advanced GPUs are designed by companies headquartered in California, fabricated by Taiwan Semiconductor Manufacturing Company in Taiwan, assembled using equipment manufactured in Europe, Japan, and the United States, and ultimately deployed in data centers scattered across North America, the Middle East, and Asia.
To the uninitiated observer, it might appear that the jurisdictional architecture of the industry is already sufficiently globalized to enable corporate nationality arbitrage. If the intellectual property behind advanced GPU architectures were transferred to an entity incorporated in Singapore, Abu Dhabi, or Zurich, one might imagine that the resulting products would fall outside the regulatory reach of Washington.
Yet the deeper one examines the structure of the semiconductor ecosystem, the more one realizes that this analogy to treaty shopping collapses rather quickly. The reason lies in a fundamental distinction between how international investment law and export control regimes define jurisdiction.
Investment treaties regulate capital. Export controls regulate technology.
Capital is inherently abstract. Ownership rights can be transferred through legal instruments without altering the underlying asset. If a mining project in Africa is owned by a Canadian company today and by a Dutch holding company tomorrow, the physical mine does not change. The legal nationality of the investment has changed even though the underlying asset remains identical.
Technology behaves differently. Modern export control frameworks define jurisdiction not through the nationality of the firm but through the lineage of the technology itself. Under U.S. export law, intellectual property developed within the United States retains its regulatory identity regardless of subsequent ownership transfers. If a piece of semiconductor design technology originated in the United States, it remains classified as U.S. origin technology even if it is later sold to a foreign entity or incorporated into a multinational supply chain.
Over the last decade Washington has expanded this principle dramatically through the evolution of the Foreign Direct Product Rule. This rule allows U.S. authorities to regulate foreign manufactured products if those products are derived from American technology or produced using American semiconductor manufacturing equipment. The jurisdiction of export control law therefore propagates through technological dependency chains rather than through corporate nationality.
This subtle distinction has enormous implications. It means that relocating a corporation or transferring intellectual property does not necessarily sever the regulatory relationship with the United States. If the design of an AI accelerator incorporates American intellectual property or relies on American design tools, the resulting chip remains subject to U.S. export controls even if it is fabricated in Taiwan and sold by a company incorporated in Singapore.
The reason this mechanism works is that the United States does not merely occupy a single node within the semiconductor ecosystem. It controls several of the system’s most critical technological chokepoints simultaneously.
Start with chip architecture. The modern GPU architectures that underpin contemporary artificial intelligence workloads emerged primarily from two American firms. Nvidia and Advanced Micro Devices have spent decades refining parallel computing architectures optimized for large scale matrix operations. Their designs form the computational backbone of most frontier AI systems deployed today.
Move one layer deeper and one encounters electronic design automation software. Designing advanced semiconductors without sophisticated EDA tools is effectively impossible. The global market for these tools is dominated by Cadence Design Systems and Synopsys, both headquartered in the United States. These platforms provide the simulation environments, layout engines, verification systems, and timing analysis tools required to transform abstract chip architectures into manufacturable designs.
Descend further into the manufacturing layer and the pattern continues. Semiconductor fabrication equipment used in advanced foundries includes critical systems supplied by companies such as Applied Materials and Lam Research. These machines handle tasks ranging from deposition and etching to wafer inspection and process control. They represent decades of accumulated engineering expertise and contain thousands of specialized subsystems.
Even the most frequently cited non American firm within the ecosystem, the Dutch lithography manufacturer ASML, remains integrated into this technological web. ASML’s extreme ultraviolet lithography machines rely on a network of suppliers spanning multiple countries, including American optical, electronic, and materials technologies. The machines themselves represent the culmination of decades of collaborative research across universities, national laboratories, and private firms.
When a single country controls multiple layers of a technological stack in this way, jurisdiction begins to propagate through the system in ways that are difficult to circumvent. Export control law becomes less about territorial authority and more about technological ancestry.
The implications extend beyond the legal architecture of the semiconductor industry. They illuminate the deeper strategic logic behind Washington’s emerging approach to AI hardware.
The current debate surrounding AI chips is often framed in narrow terms as an escalation of technological competition between the United States and China. But the regulatory proposals now under discussion suggest a broader ambition. American policymakers appear increasingly convinced that control over advanced computing hardware offers a mechanism to influence the global trajectory of artificial intelligence development itself.
The scale of infrastructure required to train frontier AI models has expanded dramatically over the past decade. Early deep learning systems could be trained on modest clusters containing a few dozen GPUs. Contemporary frontier models require thousands. The next generation may require hundreds of thousands.
Such clusters are not trivial installations. They require enormous electrical power capacity, sophisticated cooling infrastructure, high bandwidth networking fabrics, and billions of dollars in capital expenditure. Constructing one begins to resemble the development of strategic industrial infrastructure rather than an ordinary corporate investment.
Once policymakers begin to perceive frontier AI clusters as strategic assets, the regulatory impulse becomes easier to understand. If the United States controls the supply chain components required to build the most advanced AI accelerators, it can potentially shape where and how these clusters emerge around the world.
What is being contemplated is therefore not simply an export control regime but a system of compute governance. The objective is to convert structural dominance in semiconductor technology into enduring leverage over the geography of artificial intelligence development.
Whether such a system proves sustainable is an entirely different question.
Technological ecosystems possess their own gravitational dynamics. For decades the global semiconductor industry evolved as a distributed network of research institutions, private firms, and cross border supply chains. Silicon Valley, Taiwanese foundries, Dutch lithography laboratories, and Japanese materials suppliers all contributed to a system whose complexity now defies simple national boundaries.
Attempts to consolidate control over such systems often produce unintended consequences. When chokepoints become too restrictive, incentives emerge to develop alternative pathways. This process is rarely immediate. Building a parallel semiconductor ecosystem capable of competing with the existing stack would require enormous capital investment and sustained technological progress over many years.
Yet history offers numerous examples of technological fragmentation emerging under geopolitical pressure. During the Cold War the Soviet Union developed an entirely separate computing ecosystem, complete with indigenous processor architectures and software environments. That system ultimately lagged behind its Western counterpart, but it nonetheless demonstrated that parallel technological universes can emerge when geopolitical imperatives demand them.
Today we may be witnessing the early stages of a similar dynamic in the realm of artificial intelligence hardware. China has already embarked on an ambitious effort to develop domestic semiconductor capabilities. European policymakers are exploring sovereign compute initiatives designed to reduce reliance on foreign technology stacks. Middle Eastern sovereign wealth funds are investing heavily in large scale AI infrastructure projects that could position the Gulf as a neutral hub for global compute.
None of these initiatives currently threaten the dominance of the existing semiconductor ecosystem centered around the United States. The technical barriers to replicating that ecosystem remain formidable. But the direction of travel is increasingly clear.
The draft export control proposals that prompted my reflections may therefore represent something more than a regulatory adjustment. They may represent the moment when Washington began explicitly attempting to convert technological dominance into geopolitical architecture.
The irony is that the United States achieved its position within the semiconductor industry through decades of open scientific collaboration, venture capital experimentation, and globalized supply chains. The very networks that created American leadership are now being reframed as instruments of strategic control.
Whether that transformation proves durable will depend on forces that extend well beyond export control law.
It will depend on whether the gravitational pull of the existing semiconductor ecosystem remains strong enough to keep the world orbiting within it.
Or whether the pressures of geopolitical competition will eventually lead to the emergence of parallel technological civilizations operating beyond its reach.
As always, thank you for reading.
As we continue exploring the complexities of global markets together, I’m genuinely humbled by the growth of this community—now 54,000+ strong. It’s an honour to engage with such an insightful and globally diverse audience, with readers spanning 107 countries, from the United States to Switzerland.
Your feedback and engagement have been instrumental in shaping the topics I explore. If you’ve found value in these perspectives, I’d love for you to share this newsletter with your networks. Together, we’re fostering deeper discussions and critical thinking about where the markets are heading. Please write to me at kam@amanahcapital.uk