In a silicon civilization, software is the basic infrastructure of state power. States, armies, factories, hospitals, banks, and households all run on code — and almost none of them control, see, nor understand the source of the systems that run and manage their daily lives. As that code is increasingly being written and operated by AI, the dependency deepens. Users are discovering that their data is being handed off to a handful of closed frontier models concentrated in a handful of American labs, accessible only on their terms. From a political and strategic standpoint, the question of who controls software is therefore a question of sovereignty.

The power imbalance between the United States and its European allies has been widening for decades, and one of the widest gaps has opened in software. Depending on American security guarantees is one thing; relying on American software to run the state and power the economy is another. European countries are right to lament how access to their own basic services and most important institutions depend on systems they do not fully control. But recognizing this vulnerability does not mean that those dependencies can be reversed by decree. Taken literally, software sovereignty would require every country to build its own cloud provider, operating system, office suite, cybersecurity industry, frontier AI laboratory, and developer ecosystem, closer to the fantasy of autarky than to the reality of sovereignty. It’s also unattainable for virtually every country on earth, the US and China included.

European countries must decide which dependencies are acceptable and which are not. In some areas, like defense, intelligence, or critical infrastructure, domestic ownership or direct control of the stack may be necessary. In others, sovereignty is better secured through competition among providers, open standards, portable infrastructure, contractual protections, and the ability to mix domestic and foreign, closed and open-weight technologies without becoming captive to any one of them.

America’s Software Monopoly

The role of software in sovereignty is not a new concern. The dominance of Microsoft’s operating systems during the PC era already raised questions about Europe’s dependence on American technology, and spurred interest in domestic and open-source alternatives. The most consequential to emerge was Linux, released by Finnish student Linus Torvalds in 1991 and developed thereafter as open-source software, by a global community of enthusiasts. Linux never unseated Windows as the dominant desktop operating system, but it became a major rival in servers and, eventually, the cloud (which most of the internet now runs on). We have seen similar dependence on American software in almost every category: Google in search, Zoom and Teams in workplace software, and the American social networks — X, Facebook, Instagram — that shape media narratives around the world. Now, AI has raised the stakes of American technological dominance once more; the models writing most of the world’s software are themselves made in American labs.

Everyone agrees that America has dominance over the software industry upon which the world runs. This gives the United States a remarkable source of national — even imperial — power. American software companies make up much of the operating layer of modern life: AI models, cloud computing, operating systems, office software, cybersecurity tools, payment systems, developer platforms, enterprise software. It’s a form of power with no traditional equivalent — not military, diplomatic, nor cultural — and most of us carry a symbol of it in our pockets.

When a foreign company uses American services, it is entering an American-led ecosystem, adopting US standards, workflows, interfaces, and even its jurisdiction. European powerplayers like Renault or the Deutsche Bank run on Google Cloud, and their governments are no different. European tax systems, intelligence agencies, public services, and defense operations increasingly depend on software that is supplied, maintained, or influenced by American firms. In March 2026, Britain’s tax authority awarded Amazon Web Services a £472.8 million (roughly $630 million) contract to move off three legacy data centers and move services into the cloud; every other bidder withdrew, so AWS was the only option for this contract. Australia, meanwhile, is investing at least $2 billion (AUD) in a Top Secret Cloud built with AWS for its defense and intelligence agencies. The European Commission relies on Microsoft 365 for its own administration. Indeed, in March 2024, the European Data Protection Supervisor found that the Commission, which likes to fashion the EU as a “regulatory superpower,” had violated the bloc’s own data-protection rules, including those governing international transfers and control over how personal data was processed. The watchdog ordered the Commission to suspend its contractual, technical, and organizational arrangements with Microsoft unless it came into compliance. Even in China, the country that has done the most to pull away from the American software ecosystem, Windows still runs on 79 percent of desktops.

American software power is potent because it is self-reinforcing. Foreign firms use American software because it helps to make their business more efficient and is easy to adopt; governments use it because it has already proven itself; developers build on it because the surrounding ecosystem of tools, libraries, documentation, and users is nearly impossible to reproduce. The ecosystem is the moat. GitHub alone had passed 180 million developers and 630 million repositories in 2025. China has made a serious effort to build a parallel ecosystem, yet its GitHub equivalent, Gitee, reports only 14 million developers and 40 million repositories. Most Europeans use GitHub or other US alternatives like GitLab, but Europe’s most prominent independent alternative, German nonprofit Codeberg, has only recently passed 200,000 accounts and 300,000 repositories.

This kind of power cannot be built by governments, only by a dynamic private sector. But in an era when autonomy matters more and more to individual states because of the shattering of the illusion that global institutions and international alliances could compensate for power disparities between states, how the US wields its software hegemony may determine whether allied governments keep embracing American technology. If American technology comes to be seen as a geopolitical liability because it is perceived as intertwined with the American state, allies will have every incentive to reduce their dependence on it, even if that means turning to less desirable alternatives, including Chinese ones. That outcome would weaken the technological position of both America’s allies and of the United States itself.

What is Software Sovereignty?

The most intuitive answer is that software sovereignty requires every country to build its own software stack — a digital cuius regio, eius religio. The principle, settled in the early modern European wars of religion in the Peace of Augsburg in 1555 and consecrated in Westphalia in 1648 as the foundation of the modern system of nation-states, held that a prince could set the faith of his territory, free from the Pope or the Holy Roman Emperor. But even the sovereigns who chose to set the faith of their state did not whip up religions from thin air; they were built on pre-existing doctrine — Roman, Lutheran, Calvinist — keeping what served them and discarding what did not. The software beneath sovereignty was always borrowed.

Sovereignty, then, is the capacity to decide. Ownership, of course, helps — it buys more of that capacity — but it carries the cost of building and maintaining what you own. Software sovereignty is better understood as the ability of institutions to choose, govern, adapt, and replace the software they depend on, to avoid captivity to any single provider or model, and to know as much as possible about where their data sits, who can access it, how decisions get made, and how systems can be recovered when something breaks.

There are roughly three approaches to pursue the capacity for software sovereignty. The first is a maximalist approach that seeks domestic capacity across almost every layer of the technological stack, from semiconductors and cloud infrastructure to operating systems, software applications, and AI models. This view assumes software is only truly sovereign if it is domestic. It turns a question of control into a question of origin. Under that logic, using foreign software becomes a sign of weakness even when the foreign product is cheaper, better, safer, and more useful.

Only the United States and China have the market size, capital, industrial base, cloud platforms, frontier laboratories, and developer ecosystems to even attempt broad autonomy across the stack. But even they cannot achieve complete self-sufficiency. For smaller countries, attempting to reproduce the entire stack would burn enormous resources to produce weaker products.

The second approach is sovereignty through interdependence. Rather than build everything at home, a country makes itself hard to exclude by embedding itself into strategic supply chains, becoming a customer too important to lose, and turning itself into an attractive host for AI infrastructure like data centers. Part of the play is stoking the already fierce competition among American software providers.

The third is a hybrid version between the other two: selected sovereignty through open source and modularity, in which the different layers of the software and AI stack can be combined, adapted, and replaced without needing to rebuild the entire system. Countries and companies mix closed and open-source models, open standards, and portable infrastructure to build their own adaptations without starting from zero. The goal is enough control to switch providers, tailor systems to local conditions, and avoid being trapped by a single vendor in a single foreign jurisdiction.

The European Model

Europe’s approach to software sovereignty, and its relationship with the United States, rests on three assumptions that don’t hold up to scrutiny. The first is that EU member states share broadly similar interests in industrial and technological policy. The 2026 collapse of the €100 billion FCAS fighter-aircraft program, Europe’s most ambitious joint defense project, sank because of irreconcilable industrial disputes between France’s Dassault and Airbus. The failure for these closest allies to come to a solution is a preview of what we could expect in any attempt to build a pan-European software alternative. The second assumption is that the United States is purely an external actor, with no European participation in the American technological ecosystem. The third is that Europe lacks indigenous talent and software capability. Europe has real weaknesses in financing, scaling, and retaining successful companies — but those are different problems than lacking the capacity to build technology itself.

Europe therefore lacks a serious strategy for negotiating the terms of its dependence. A European company using Microsoft, Google, Oracle, Amazon, or Salesforce is not automatically less sovereign because the vendor is American. What matters is control over its data, bargaining power over the terms, alternatives if something goes wrong, and the ability to keep operating in a crisis. Most firms don’t need a local provider for services like spreadsheets or email; they need good, cost-efficient, secure tools, and the ability to move to other providers.

Europe is neither incapable of producing successful software nor lacking the people required to build it. Consider the German firm SAP, whose software runs finance, supply chains, manufacturing and human resources for companies worldwide; SAP reported €36.8 billion in revenue in 2025. France’s Dassault Systèmes earned more than $7.1 billion from industrial design and simulation software. The Dutch payments company Adyen processed approximately $1.6 trillion in transactions during the same year. Additionally, on the hardware side ASML remains the only supplier capable of producing extreme ultraviolet lithography systems, without which the most advanced semiconductors cannot be manufactured. Others fill out the picture: OVHcloud, Switzerland’s Proton, Mistral AI, and Linux itself, just to name a few.

Europe also has a substantial technology workforce that is contributing significantly to American tech. Across the EU, there were more than 10 million working ICT (information and communications technology) specialists in 2024. Meanwhile, nearly 887,000 European-born workers held STEM jobs in the United States in 2023. European founders are responsible for founding one-tenth of American technology companies. By the estimate of the State of European Tech report, the transatlantic drift of entrepreneurs produces about 800 American startups a year, companies founded at Europe’s expense.

European countries have failed to retain ownership of what their engineers and entrepreneurs create. American software power is American in ownership and institutional form, but the engineers, researchers, founders, industrial customers, and markets on which that power depends are substantially European. Europe is a key participant in the American ecosystem. Building a wholly separate European stack sounds attractive in theory, but it ignores the extent to which the two sides are already integrated. The real question is how Europe can convert its participation into bargaining power, and how to turn that bargaining power into sovereignty.

This is where much of the European debate goes wrong — the reflex that if the software is American, the answer must be European software. Sometimes that is true: defense, intelligence, critical infrastructure, and sensitive public systems may demand tighter national or European control. But as a blanket doctrine, substitution is counterproductive because, short of moving the clock back to the 1980s, there is no space for those alternatives to mature.

Europe’s real leverage lies elsewhere. American technology companies are not indifferent to Europe: they want Europe’s customers, infrastructure, industrial clients, and talent. Much of the AI and cloud infrastructure now going up in Europe is being built by American firms precisely because those firms want to remain embedded in the European market. European countries can use this position to demand stronger guarantees concerning data, continuity of service, legal exposure, local infrastructure, operational control, and access during a crisis.

That bargaining power is a form of sovereignty. A country does not become more sovereign by cutting itself off from the strongest technology ecosystem in the world; it becomes sovereign by forcing that ecosystem to compete for its business and build on its soil. The paradox is that the surest way for European nations to keep their sovereignty in the silicon age may be deeper integration with American tech, on European terms. Washington has already run this play. Through the CHIPS act, which passed in 2022, it used market access subsidies, and political incentives to bring a greater share of TSMC — the Taiwanese foundry the chip industry depends on — into building in Arizona. European countries could apply a similar principle to cloud computing and AI by demanding local infrastructure, local personnel, and operational guarantees as the price of market access.

The Japanese Model

Japan may offer the most instructive model for countries in the Western world. Japan has pursued several experiments that reflect the different approaches to software sovereignty, carving out a space between full ownership of the stack and total foreign dependence. At the maximalist end, SoftBank’s SB Intuitions is building homegrown Japanese-language models and aims, in the medium term, to eventually develop a model exceeding one trillion parameters. Japan is also pouring billions of dollars in support to Noetra, a state-backed consortium whose members include SoftBank, Sony, Honda, and NEC. These are serious efforts, even if the models remain behind their American and Chinese competitors, and even though full-stack sovereignty remains ruinously expensive and tethered to foreign supply chains.

Another approach comes from Sakana AI, a Tokyo-based AI lab founded by a coauthor of the original Transformer paper. Rather than training a new frontier model, Sakana built Fugu, named for the poisonous delicacy pufferfish, a system routing tasks across a pool of existing frontier models and learning which one to call for which job. Sakana sells it to Japanese enterprises as a cleaner, more local alternative to relying wholly on a single foreign lab and, at least in theory, insulated from the export controls that can abruptly cut off access to frontier models. This does not make Japan immune to foreign restrictions — Fugu itself needs access to at least one foreign frontier model — but it reduces dependence on any single provider. It points to a version of sovereignty more grounded for middle powers: not owning the intelligence, but owning the judgment about which intelligence to use, and when.

In general, Japanese firms and consumers use American software where it makes sense. But Japan has preserved strong domestic ecosystems in areas most shaped by language and consumer habits. Companies like ELYZA and Preferred Networks built models tuned for keigo, the honorific register of Japanese business speech, and for the cultural references a general-purpose American model still handles less gracefully than a Japanese one. These sorts of companies win on the things that matter locally, like register, cost, and the guarantee that Japanese data will stay in Japan. This has a long tradition in Japan’s digital ecosystem. NTT DoCoMo’s i-mode pioneered the mobile internet years before smartphones became widespread, even if it never became the global standard. LINE became central to Japanese messaging culture. Not to mention, Japan’s Nintendo and Sony have led the gaming world for decades.

Japan’s government cloud strategy hasn’t shut out American providers. Its Digital Agency has selected AWS, Google Cloud, Microsoft Azure, and Oracle (and the domestic operator Sakura Internet) as Government Cloud providers. Sakura had to pass similar vetting as foreign providers, and indeed failed to meet Japanese government standards in its first attempt in 2023. The existence of Sakura thus matters because it gives Japan a domestic option inside the system. Japan did not pretend that sovereignty required excluding American hyperscalers. It set requirements, opened the door to a domestic provider, and used the best tools available.

That is the sane model of software sovereignty that builds where local knowledge creates an advantage, preserves capacity where security requires it, and uses foreign tools where they are better. That way, dependence does not degrade into captivity. But the Japanese model owes much of its feasibility to the fact that Japanese firms are genuinely better than foreign firms in adapting to Japan’s linguistic and cultural needs. Europe has no equivalent as its languages and markets are more similar to American ones. It can still, however, learn the structure of the Japanese approach, even while lacking the cultural autonomy of Japan.

The Coder and the Prince

The West has a great opportunity to build a shared technological space that strengthens the sovereignty of Europe while reinforcing the strength of American technocapitalism. The Prince does not need to write the code, as long as he retains influence over it. American software, as the institutional infrastructure of the Western-aligned world, will play a key role in this endeavor. It gives the United States influence, and with it a responsibility to provide useful, reliable, and secure software to its allies.

Helping European nations build their technological sovereignty should not be seen as charity, but an act of self-interest which expands the market for American companies, encourages investment in compatible infrastructure, and keeps allied governments, talent, and industries connected to an American-led technological sphere. The alternative is worse for everyone. If American software comes to be associated with extraterritorial control, or if access to essential services could be withdrawn during a political crisis, Europe will respond with protectionism, regulation, and subsidies for domestic alternatives. A frightened ally makes for a bad customer and a worse partner. Europe may not be able to build equal competitors today, but it can create enough friction that would undermine the Western technology ecosystem, and hand openings to America’s chief rival by cultivating Chinese software instead.

That the West can build on the idea of software sovereignty beyond a zero-sum proposition is already a reality. Beyond the established technological giants, a wave of companies is building the sovereign capabilities of Western nations. Canada’s Cohere lets its models to run inside a customer’s private cloud, and has partnered with Germany’s SAP to provide sovereign AI services to European governments and companies. The American chipmaker Cerebras is similarly working with Germany’s Aleph Alpha to develop secure AI systems for European public-sector and defense use. Again, we see here a clear example of an American company assisting in reinforcing the sovereignty of a European nation. Other companies provide particular layers of the sovereign stack. Groq has built AI-inference infrastructure in Helsinki that runs models from multiple developers on European soil. Cloudera lets organizations manage data and AI workloads across their own local infrastructure and sovereign clouds.

Open-weight models are another important layer of this sovereign stack, and perhaps the most important. Their strategic importance was recently emphasized in July 2026 by “Open Weights and American AI Leadership,” a letter fronted by Nvidia and signed within days by more than 200 companies, including Microsoft, Meta, IBM, Palantir, Mistral, and Hugging Face. It argued that American AI leadership should sustain a broad, open ecosystem — the natural successor of the open-source software of the ‘80s and ‘90s that came to run most of the internet. Because open models can be downloaded, adapted, and operated on infrastructure the user controls, governments and companies gain a credible fallback and the ability to tune models to their own cultural and institutional needs. The strongest open models are also narrowing the gap to closed models like DeepSeek V4 and Kimi K3, and smaller models already match proprietary alternatives in specialized tasks at lower cost.

The rise of Chinese open-weight models like DeepSeek and Kimi makes competitive Western alternatives more urgent still. Yet open weights still lag behind the best closed models in general capability, so they are best understood not as substitutes for frontier models, but as part of a mixed architecture that prevents any single provider from becoming a point of dependence.

Hugging Face, an open-source AI platform and model hub, makes these models easier to move between providers and infrastructure, while American companies provide the computing systems required to run them locally. Palantir offers an operational layer that allows institutions to connect their own data and applications to different models, rather than letting one model provider control the whole system.

None of these arrangements builds a wholly European supply chain. Nor does sovereignty require one. The Western technological system can accommodate specialization, competition, national control, and companies from different allied nations all working across the same tech stack. Europe can maximize its sovereignty while benefiting from American technology; the US can keep its technological leadership while accepting that its allies need operational autonomy to be useful allies. That is the bargain the West must accept. It is in the United States’ best interest to ensure that American software remains the answer to Europe’s sovereignty problem, rather than the reason Europe goes looking for another.