A combination of a growing interest in dual-use technologies and government contracting, concerns over sovereignty and the an AI valuation bubble could open up the global AI race to a wider range of competitors.
The following chapter is split into four subsections, each focusing on a significant trend in the AI sector observed in 2025 and early 2026. These sections identify specific dynamics that could result in a further securitization and levelling of the AI race. The section then analyses where those dynamics may lead. The trends under discussion include: the dual-use and defence AI boom; the rise of ‘patriotic tech’; the growing push for sovereignty in AI and defence; and concerns over an AI valuation bubble in financial markets.
A lot can change in 12 months. This chapter therefore adopts a relatively short time-horizon and assumes a relatively stable progression in terms of technological advancement. The potential for any AI company to achieve artificial general intelligence (AGI) is outside the paper’s scope for several reasons, including the lack of expert consensus on what achieving AGI would actually entail; the emerging consensus that such a breakthrough remains hypothetical and, at a minimum, several years away from happening; and the profound societal and economic upheaval AGI might bring.
Trend 1: A ‘boom’ in dual-use and defence tech
Investment in dual-use and defence AI has increased sharply in recent years. Venture capital (VC) spending on defence tech reached record levels in 2025, with VC-backed defence startups in the US and Europe, by some estimates, raising a combined $7.7 billion between January and October in that year – more than double 2024’s total. Overall private defence investment in 2025 exceeded $48 billion, driven by large funding rounds for companies in the AI and autonomous drone sectors. This growing interest is largely a result of the promise of high levels of government spending on defence and heightened interest in AI’s battlefield utility, which at time of writing is being put to the test in several ongoing conflicts.
The war on Ukraine has not only spurred a significant increase in global military spending, but has also showed the utility of commercial, off-the-shelf technology innovations on the battlefield.
Though investment in battlefield autonomy and AI for military purposes dates back many decades, Russia’s full-scale invasion of Ukraine in 2022 has been a major factor behind the resurgent interest in dual-use and defence technologies. The war on Ukraine has not only spurred a significant increase in global military spending, but has also showed the utility of commercial, off-the-shelf technology innovations on the modern battlefield. While large platforms and traditional weapons systems will not lose their importance in war, Ukraine has revolutionized the extensive and innovative use of cheap, disposable drones, enabled by dual-use systems like Starlink. For many military planners, the Ukraine war has reinforced the lesson that success in modern warfare will depend on a country’s ability to leverage not just their defence–industrial bases, but also their commercial technology ecosystems and the innovation that flows from them.
However, collaboration between commercial ecosystems and the defence sector has in the post-Cold War era been constrained by several structural challenges. These hurdles include the difficulty for the public sector of competing with commercial investors and markets, opaque and burdensome procurement processes, and public scepticism towards companies working with the military.
To overcome these hurdles, many governments have begun implementing policies aimed at opening up investment, reducing procurement friction, and encouraging commercial technology companies and new defence startups to participate more actively in defence contracts. In some cases, this effort has included broader industrial policy initiatives, or, on a smaller scale, the creation of new specialized agencies designed to make it easier for non-traditional players to work with the military.
These efforts appear to be bearing some fruit. In the post-Cold War era, most defence markets have come to be dominated by a small number of incumbents (or ‘primes’), such as the UK’s BAE Systems, the US company Lockheed Martin and Italy’s Leonardo. But recent years have seen a rapid influx of a far more heterogeneous set of competitors into the sector, in part spurred on by governments’ increased push for dual-use technologies. Although traditional defence companies remain key beneficiaries of the defence spending surge, so-called ‘neo-primes’, such as Germany’s Helsing, the US’s Anduril and Finland’s ICEYE, have also entered the marketplace, keen to capitalize on renewed interest and the promise of sustained high levels of government defence spending.
The large technology companies, including the major US-based ‘hyperscalers’, have been increasingly active participants in the defence industry. Between 2018 and 2022, the Pentagon signed contracts worth an estimated $53 billion with such firms. By 2025, Anthropic, Meta and OpenAI each signed deals with the US Department of Defense to embed their AI solutions into the US military. Several are part of the Pentagon’s Project Maven and its GenAI.mil platform, launched by the Pentagon in December 2025 to provide secure generative AI access to its 3 million staff members. GenAi.mil, among others, embeds Anthropic’s Claude, Google’s Gemini, xAI’s Grok and OpenAI’s ChatGPT models. (Although it should be noted that, at time of writing, Anthropic’s involvement was in question after a highly publicized stand-off between the company and the White House over the use of the former’s products.)
A similar shift can be observed in terms of financing. Traditional commercial funders, which historically had steered clear of defence investments and were frequently institutionally or legally restricted, or even barred, from doing so, are now becoming more active. Across the globe, both governments and, increasingly, VC firms, banks and other investors are pouring money into companies developing military or dual-use AI applications, with investment rising particularly rapidly in Europe, the US and, to a lesser extent, the Middle East and Indo-Pacific. In Europe, at least five new specialist defence-technology venture funds were established in 2025, increasing the continent’s total to 13. The newest of these funds, DTCP’s Liberty Fund, allocated $500 million to European technology companies. At the same time, large financial institutions are beginning to move more decisively into the sector. For example, JPMorgan Chase announced in October 2025 that it would invest $10 billion annually in US national security and resilience solutions.
The US and China continue to dominate the defence technology landscape, mirroring their leadership of the global AI marketplace. These two markets account for the largest share of global defence-technology investment. Defence tech investments in the US are still approximately three times higher than in the combined European NATO countries.
The EU cumulatively spent approximately €381 billion on defence in 2025, nearly double what it spent a decade earlier, with an increasing share of this spending directed towards new technologies, including AI-enabled capabilities.
Yet, while the US and Chinese markets continue to dominate in terms of absolute numbers, the rate of growth in those markets is not necessarily the fastest. Investment in defence AI has been growing rapidly in Europe, driven by rising market demand, higher defence spending by European governments and a growing emphasis on strategic autonomy. The EU cumulatively spent approximately €381 billion on defence in 2025, nearly double what its member states spent a decade earlier, with an increasing share of this spending directed towards new technologies, including AI-enabled capabilities. The EU is now one of the fastest-growing major markets for dual-use AI investment, with VC-backed investment increasing by around 80 per cent between 2024 and 2025 by some estimates, and the highest in terms of numbers of deals (though later-stage, higher-value deals remain elusive). It is important to put these numbers in perspective: the European market starts from a far lower base, and many institutional and financial barriers remain. Nevertheless, where Europe’s technology industry was previously wary of working with the military, several of the continent’s largest so-called ‘unicorns’ – such as Helsing and Portugal’s Tekever – are now primarily active in the military AI sector. Cities like Stockholm, London, Paris and Munich are emerging as key defence-technology hubs.
For the US, the goals of leading on military AI development and integrating the technology into systems to gain a battlefield edge are intimately intertwined with wider ambitions to win the AI race and gain dominance over the technology. Successive US administrations have described such dominance as the geostrategic determinant of the 21st century. Export controls preventing Chinese rivals from accessing the advanced semiconductors required to develop frontier AI, which Washington fears could aid in the PLA’s modernization, are a key manifestation of this, The increasingly intimate relationship between the Pentagon and Silicon Valley is another. For China, winning the AI race similarly means being the most effective at deploying AI solutions in its military – with new domestic tech challengers seen as vital to helping Beijing achieve this objective.
Other, smaller actors also give defence and dual-use AI heightened importance. Governments are increasingly treating the military deployment of AI as a ‘silver bullet’ that may help solve critical challenges across a range of geostrategic, political and economic objectives. AI is believed to be able to help ease personnel shortages and may bring real battlefield advantage. But it could also bring spillover economic benefits, help countries to cultivate a wider range of sovereign AI systems and grow their global influence and market share.
AI’s potential as a force multiplier and, more mundanely, the efficiency gains predicted from its use are the most important reasons why governments and the private sector (keen to cultivate governments as customers) are dedicating substantial resources towards building domestic defence AI industries and solutions. These ambitions are as frequently about addressing persistent structural weaknesses, such as constraints of time, money and manpower, as they are about countries seeking to place themselves at the absolute frontier.
Government spending in cutting-edge technologies on the battlefield, and rearmament more generally, are often presented as an opportunity for economic growth, job creation and generating spillover benefits from AI developed for military ends. The ‘defence dividend’ from high-tech investment (i.e. the idea of using higher levels of military spending to generate wider economic growth) has frequently been referred to by governments in the UK and elsewhere as an important opportunity.
The perceived benefits of dual-use AI go beyond domestic economic growth to foreign policy. The development of a sovereign, competitive and independent dual-use AI and technology industry could simultaneously serve as an instrument of both soft and hard power to expand a country’s influence. The US and China have been able to achieve this. Middle powers with a relatively large defence footprint can do the same at a smaller scale. For example, the UK, France and Russia have well-developed defence export industries and are keen to replicate this strength in defence AI. Turkey and Ukraine, too, have been using their defence equipment exports and expertise in drone development – which increasingly involve degrees of autonomy – as a means of building wider partnerships.
How this trend may lead to multipolarization
Higher defence spending by middle powers can lead to innovation spillover
Just as the initial rise of Silicon Valley relied on large defence contracts, other sectors may similarly find themselves with an opportunity to grow on the back of high levels of defence spending. This opportunity could be especially relevant in the case of a general-purpose technology like AI, which has significant dual-use applications – meaning that innovations made with a primarily military objective in mind may eventually find civilian uses (or vice versa).
South Korea has been notably successful in cultivating dual-use technology and reaping some of the spillover benefits. This in no small part because, unlike its partners in the West, it has maintained a constant state of war readiness given tensions with North Korea. South Korea has been able to cultivate a large domestic market and fostered a symbiotic, government-led relationship, geared towards bringing the country’s large national corporations and its well-developed tech sector into military production. Though initially primarily focused on satisfying its domestic market, South Korea has become an important exporter of military equipment to partners in Asia, Europe and the Middle East.
As access to capital and talent remain among the main obstacles preventing alternative AI ecosystems from competing, the emergence of innovation clusters around specific defence applications may also provide the seed for the growth of a more mature commercial funding ecosystem focused on wider applications. Governments in places like the EU and Japan are reducing some of the investment barriers to market growth – for example, by making it easier for pension funds and other institutional funders to invest in dual-use applications.
Defence spending, in this scenario, functions as industrial policy and is used to promote domestic innovation. As governments and militaries are the primary customers of defence AI applications as well as dual-use applications like logistics systems, cloud services and AI-enabled space-based applications, strategic spending can help strengthen a wider ecosystem. High levels of defence spending may, for example, be used to help overcome some of the challenges around scaling – which has hitherto been among the biggest difficulties faced by European startups and has led to promising new businesses relocating to the US, the Gulf and other markets to access sufficient capital. The growing interest in dual-use AI, and the formation of small clusters, can also concentrate more AI talent in a specific location – the ‘agglomeration effect’ – from which other new actors outside of the defence-tech sector may benefit in terms of transfers of ideas and people.
Battlefield experience can give small and medium-sized countries a comparative advantage
Ukraine has been referred to as an ‘AI war lab’, with both Ukraine and Russia increasingly deploying AI-enabled solutions in decision-making processes or scaling up drone operations that have been a defining aspect of the four-year conflict. For Ukraine, bringing technology startups and other providers of cutting-edge solutions into the military became a wartime necessity. But the country’s burgeoning defence-tech industry, its growing tech talent pool and a permissive regulatory environment that allows for the rapid testing and procurement of new solutions provide a potential model for its post-war economy. Ukraine is seeking to build on its experience with rapid war-time innovation in AI and drones in particular, brokering partnerships with large defence contractors elsewhere in Europe and striking deals with the US and in the Middle East to scale production and export its knowledge.
Ukraine has been referred to as an ‘AI war lab’, with both Ukraine and Russia increasingly deploying AI-enabled solutions in decision-making processes or scaling up drone operations that have been a defining aspect of the four-year conflict.
Another example of a country that has turned battlefield experience into developing its wider technology sphere is Israel, which maintains one of the world’s most tech-mediated and AI-powered military and intelligence apparatuses. It has turned this expertise and battlefield experience – coupled with the fact that many of its tech workers have served in the Israel Defence Forces (IDF) – into a competitive cyber and defence tech industrial complex. By some measures, Israel’s AI industry ranks seventh in the world, despite the country ranking 29th in terms of GDP and 97th in terms of population.
These examples show that growing investment in dual-use AI can potentially be a leveller for small and medium-sized countries that are able to internalize battlefield knowledge and find opportunities to apply it to other domains.
Real-world use can enable less-advanced countries to catch up in the defence AI race
Ukraine’s creative use of these technologies, and its ability to test and rapidly iterate them directly on the battlefield while developing doctrines for their use, will inevitably have implications for AI deployment beyond the frontline. According to several reports, Ukrainian AI and drone systems routinely outperform those from Western companies, many of which are also using the war to analyse the strength of their technologies.
The US and China may still be able to capitalize on their sizable lead to gain battlefield advantage and pull further away from less-advanced peers in the race. But equally, others may have a chance to catch up. As AI systems are deployed more in combat, grand promises will be tested against operational reality. The true state of the defence AI market will become clearer, revealing the actual utility of technologies and moving discussions beyond simple comparisons between the size of national markets or the scale of investment. A case in point is the fact that Ukrainian drone tech experts have been called in to support Gulf Arab countries, as well as the US military, to help them defend against Iranian drone attacks.
Trend 2: The rise of ‘patriotic tech’ and the blurring of boundaries between civil and military
2025 was characterized by a more intimate relationship between AI companies and governments, with many commercial operators becoming increasingly open in their willingness to explore the defence and national security applications of their AI products. Though this is a global dynamic, the trend is especially visible in the US, driven not just by new defence tech companies entering the AI marketplace, but also by the hyperscalers and frontier AI labs pursuing contracts with the Pentagon and allied defence ministries, and presenting themselves as working in support of US government (and, by extension, NATO) objectives.
Just as governments have become increasingly proactive in courting the companies behind cutting-edge AI solutions, so too have commercial companies become more interested in pursuing military contracting and exploring the dual-use and defence applications of their products. This is the result of sustained levels of high government spending and the implied prestige of developing high-tech solutions for the military, but also a heightened, more insecure and combative security environment, in which private sector actors are encouraged to do their ‘patriotic duty’.
The latter is particularly important. In the US, just like everywhere else in the West, the end of the Cold War ushered in an era in which defence spending declined precipitously and the defence industry rapidly consolidated into a smaller number of large companies. During this time, the focal point of innovation shifted away from the Cold War model in which government-supported laboratories and government-aligned companies developed defence solutions which would eventually flow down into the commercial sphere (such as GPS, the semiconductor and the internet, which all originated from government contracts), towards a commercial-first approach. This narrative shift is now almost fully complete, with especially leading-edge private sector companies in Silicon Valley driving frontier innovation, while militaries and defence contractors purportedly struggling to fully harness these solutions.
The inability of the US military to benefit in full from the country’s cutting-edge technology companies and their innovations has become a source of bipartisan fear in Washington that US military capabilities are atrophying, while more agile rivals – above all China – better able to use the full strength of their domestic ecosystems, are rapidly catching up. As a result, from the 2010s onwards, the US government has made repeated appeals to Silicon Valley to bring more of their frontier AI technologies and expertise into the military. The Defense Innovation Unit (DIU), for example, was created under President Barack Obama to help bridge the gap between Silicon Valley and the Pentagon, by making it easier for tech companies to speedily and easily access funding and providing them with support to rapidly test and scale solutions for military use. The intent behind DIU, the remit and funding of which has increased with each administration, is not just about opening up new sources of funding, but also about building trust between government and tech companies, and allaying concerns about complex bureaucracies and the ethics of working with the military.
Especially under the current US administration – which has set the ambition to accelerate America’s AI dominance by becoming an AI-first warfighting force – closer ties between the tech industry and the military appear to have taken on a distinctly ideological dimension, over and above commercial incentives. While, for a long time, Silicon Valley companies had preferred to keep a certain distance from the state, presenting themselves as more neutral, and global, entities, many are now beginning to position themselves as key actors in preserving US hegemony, security and providing the government with support in its competition with China.
Some new tech ‘primes’ present themselves explicitly as exponents of US power. For example, Palantir’s CEO, Alex Karp, has repeatedly said that his company’s mission is to ‘defend the West’ and to ensure the latter continues to ‘dominate’ technologically. Palantir’s 2026 manifesto wrote that ‘Silicon Valley owes a moral debt to the country that made its rise possible’ and that ‘the engineering elite of Silicon Valley has an affirmative obligation to participate in the defense of the nation’. It has also been widely reported that executives at several large tech companies have joined the military to gain more personal battlefield experience and to bring tech expertise directly into the US Army.
This turn is not just reflected in rhetoric, but increasingly also in decisions around investment, partnerships and the activities these companies are involved in. Project Maven, the Pentagon’s programme to help accelerate the military’s adoption of AI launched in 2017, illustrates this evolution. When news reports emerged in 2018 that Alphabet had been supporting the initiative, the backlash was so immediate that employees staged a walkout – which ultimately prompted the company to limit its involvement. As of 2026, several large US tech companies now openly take part.
Under Beijing’s doctrine of ‘civil–military fusion’, private companies are expected to contribute directly to national technological self-reliance, an objective which has gained further momentum amid increased competition from the US and its allies.
Although the US is currently in the process of giving shape to this evolving tech–military relationship, close relations between state and the private sector are long-standing and more explicitly formalized in China. Under Beijing’s doctrine of ‘civil–military fusion’, private companies are expected to contribute directly to the modernization of the People’s Liberation Army (PLA) and to national technological self-reliance, an objective which has gained further momentum amid increased competition from the US and its allies. The Chinese state, including the military, acts as a major patron and customer of new solutions, providing early scale and support and championing the most promising domestic challengers.
Several of China’s largest tech champions, such as Huawei, reportedly have their roots in the PLA. While the maturity of China’s ecosystem means these companies are now less dependent on government as a customer, many now primarily commercial companies remain entangled with China’s military. For example, one of China’s open-source AI champions, DeepSeek, now underpins some of the PLA’s experiments in UAV autonomy and is being diffused and integrated across China’s military, to help it transition towards what it calls ‘intelligentized warfare’.
This growing intimacy between the state and private AI sector is not just the preserve of China and the US. Significant parts of Israel’s tech ecosystem are intimately intertwined with the IDF, while South Korea has long embraced self-reliance in its defence production. The latter’s successful embrace of dual-use technology development in no small part relies on extensive government involvement and deep defence–civilian integration. While Korean companies initially participated in this system for mostly patriotic reasons, the country’s emergence as a leading global defence exporter has created additional commercial incentives.
The EU is home to a rapidly growing and increasingly well-funded sector of AI and other tech startups actively pursuing military contracts, with many of these companies explicitly citing strengthening European strategic autonomy as a motivation. For example, Torsten Reil, the co-founder of Helsing, has said: ‘[Europe] should develop homegrown systems that we control, both in terms of actually controlling the whole technology, but also the ethics side of it… what degree of autonomy are we prepared to accept is something that we need to be able to control.’
How this trend may lead to multipolarization
The phenomenon is spreading
Russia’s full-scale invasion of Ukraine has prompted an interest in defence tech among young European entrepreneurs. Many founders cite their ambition to strengthen Europe’s tech ecosystem as an important motivator. This phenomenon is set to continue: in 2026, for the first time, tech workers leaving the US for Europe outnumber those going the other way.
Europe’s existing champions and new entrants alike are calling for more concerted action to strengthen Europe’s competitiveness and security. These endeavours are now frequently framed as attempts to strengthen European strategic autonomy, rather than that of individual member states. One of the major hinderances to the growth of both Europe’s technology and defence sectors has been fragmentation of efforts. This new wave of entrepreneurs may be better placed to overcome some of these barriers.
The resulting intimacy between defence tech players and governments is unlikely to decline in the near future. Increased entanglement can be used to grow the market for companies’ AI solutions and to strengthen the overall competitiveness and strategic alignment of domestic AI ecosystems. This dynamic may especially benefit challenger markets with more limited resources. One of the current race leaders – China – provides an example of how governments and companies working in close harmony to shape markets can rapidly scale-up domestic tech industries to compete with a hegemonic power. Israel and South Korea also provide models at a smaller scale.
This kind of symbiosis may allow new challenger markets to gain an advantage and become better able to direct the full strength of their domestic industry towards the pursuit of strategic goals. It can, however, create a dynamic in which both the government and private sector become increasingly reliant on each other – where both sides could feel pressured into removing safeguards and rushing potentially harmful deployment.
‘Civil–military fusion’ could raise concerns over coercion
Symbiosis between the AI sector and governments – whether real or merely perceived – can also become a source of global tension and fragmentation, which may further spur on national sovereignty and decoupling efforts.
China’s doctrine of ‘civil–military fusion’ is often cited by policymakers in Washington as the motivation behind the US’s increasingly all-encompassing (although, under the Trump administration, somewhat unfocused) export-control regimes targeting Beijing’s ability to develop frontier AI. US concerns centre around the perception that, owing to this doctrine, virtually all technology developed by China is inherently dual-use in nature because any company can be called on to ‘do its civic duty’. Similar perceptions of a lack of neutrality may start to take hold in relation to other markets, as more governments become more actively involved in cultivating and scaling their own sovereign champions (see Trend 3, below). Customers around the world may grow increasingly concerned that the products they buy and use are contingent on the strategic objectives of the home governments of countries where suppliers are domiciled.
Governments may, in turn, come under rising pressure to reduce the regulatory burden on, or even to bail out, companies that are regarded as pivotal for national prestige and security. The companies involved would in turn become more reliant on government and military funding and contracts, and may find themselves pressured to, for example, take actions to support geopolitical objectives that further augment the power of the state. Private sector actors may find themselves increasingly pressured to become involved in the state’s international affairs. For example, the International Criminal Court (ICC)’s chief prosecutor was reported to have had their access to email and many other cloud-based services removed in July 2025, following the Trump administration’s imposition of sanctions against the ICC over its actions against the Israeli government.
In an era of weaponized interdependence, governments with closer ties to (or even holds over) their private sectors are better able to co-opt the power of these companies. In a farewell address from the White House in January 2025, Joe Biden spoke of his concern about the possible rise of a ‘tech-industrial complex’. This dynamic will leave the private sector trying to achieve a delicate balance. Refusing to comply with their home government’s demands could lead to serious repercussions. Anthropic, for example, at the time of writing, still faces a supply-chain risk designation in the US, which could lead to the US government, as well as companies doing business with it, being barred from working with the company. However, complying and working closely with a government that is increasingly seen to be engaging in coercive diplomacy can also erode global trust in a company’s brand and the reliability of their services. Over time, companies domiciled in countries which are seen to be adhering to the rule of law, may benefit by attracting talent and customers.
Trend 3: The pursuit of sovereign AI
As the geopolitical climate becomes more tense and the geostrategic utility of having independent AI capabilities more evident, a growing cohort of countries has started to reassess the composition of their technology stacks – encompassing everything from the underpinning physical infrastructure (e.g. undersea cables and data centres, computing power, chips and so on), to the protocols, standards and code, as well as the software and solutions that run on top. This reassessment has added urgency to existing calls for governments to develop their own sovereign capabilities and sees countries attempt to disentangle themselves from a perceived over-reliance on external AI companies and solutions, and instead aligning with trusted domestic or allied partners that can contribute to more resilient supply chains and software stacks.
Sovereign AI strategies and ambitions, in which countries try to cultivate either fully or partially independent AI capabilities, should be seen as laying on a spectrum. Some countries or groupings – such as China and, to a lesser extent, the EU – are aiming to develop their own capabilities across all layers of the AI stack. Other middle and smaller powers – including countries ranging from Brazil to Pakistan and Vietnam – are pursuing more limited strategies, focused on a variety of approaches. These strategies include hedging between Chinese, US and others’ AI products, developing their own niches in the AI industry and underpinning supply chains, adopting cheaper, open-source AI solutions, opening up access to digital public AI infrastructure, and ensuring that the most critical capabilities for national security remain under domestic control. Rather than seeking full technological self-sufficiency (which – with the possible exception of China – is in practice not achievable), these strategies typically aim for greater strategic autonomy through reducing external dependencies, cultivating domestic capabilities and, in some cases, creating technological ‘chokepoints’ that can be used as leverage against others.
These efforts are not new, but were initially primarily motivated by economic considerations. They have increasingly taken on a national security dimension, as governments worry about dependencies in supply chains and solutions that may be weaponized against them. Japan and India illustrate this point. Japan has increased its efforts to cultivate a sovereign AI stack by seeking greater control over foundational infrastructure and reducing reliance on external cloud and AI providers. Tokyo subsidizes domestic cloud capacity, supports homegrown AI companies and aims to increase self-sufficiency in AI infrastructure, partly by leveraging the country’s long-standing strengths in semiconductor manufacturing. These initiatives are linked to Japan’s broader economic security agenda and reassessment of its post-Second World War security posture and are complemented by efforts to stimulate domestic defence AI innovation. In 2026, Japan made diverting government R&D spending towards dual-use innovation a priority, and also seeks to promote these activities through its Acquisition, Technology and Logistics Agency (ATLA), as well as cooperation frameworks such as Pillar II of AUKUS and the Quadrilateral Security Dialogue group (or Quad).
India, similarly motivated by concerns about strategic dependence and by recent security tensions with Pakistan (during which its high-tech solutions were perceived to have underperformed), has begun leveraging its growing commercial technology ecosystem to strengthen sovereign dual-use capabilities and homegrown defence AI. In 2025, New Delhi launched a dedicated technology fund and expanded initiatives such as Innovations for Defence Excellence (iDEX), aimed at cultivating domestic high-tech defence companies and reducing reliance on imported equipment from the US, Russia and elsewhere. Sovereign AI was also a main theme of the 2026 AI Summit hosted by India – a forum which the Indian government also used to promote its latest defence innovations. These initiatives exist in parallel with its long-standing, and already successful, efforts to build public digital infrastructure, which have allowed India to build its own independent payment and identity systems.
The EU has long treated the need to grow its presence in the global AI markets and achieve strategic autonomy in both AI and the wider technology sector as a policy priority. Previously, these efforts were largely seen through an economic and competitiveness lens. Policymakers in Brussels and member-state capitals have long recognized that the continent’s lagging domestic AI and wider digital technology industry present a serious challenge to its long-term competitiveness, democratic resilience and independence. The Draghi report, for example, pointed out that the divergence between US and EU productivity growth – and GDP – from the 2007–08 financial crisis onwards can to an extent be explained by the rapid expansion of the US tech industry, contrasted against the EU’s apparent inability to create any major new market-leading technology companies since 2000. The EU is not currently home to any hyperscalers and its AI champions struggle to access the levels of (especially late-stage) funding available in the US especially.
The lack of a competitive AI and digital tech industry has made European countries increasingly reliant on the US for their technology stacks, a dependency which in turn has made it difficult for European companies to compete.
This lack of a competitive AI and digital tech industry has made European countries increasingly reliant on the US for their technology stacks, a dependency which in turn has made it difficult for European companies to compete. Just three US-based companies (Amazon, Google and Microsoft) account for around 70 per cent of the European cloud storage market, for example. However, with transatlantic tensions increasing throughout 2025 and early 2026, these kinds of external dependencies are now widely perceived as a serious national (and continent-wide) security risk. A March 2026 poll found that 86 per cent of Europeans now consider the possibility of the US government suddenly restricting Europe’s access to critical technologies and digital services to be ‘plausible’ and something that ‘should not be ruled out’, and express preferences for decoupling.
This change in narrative was spurred by several events. The first arrived in February 2025, when the Trump administration threatened to turn off Ukraine’s access to SpaceX’s Starlink systems and briefly requested US commercial satellite imagery provider Maxar to disable Ukraine’s access, in a bid to convince Kyiv to sign a controversial deal on access to critical minerals. This episode raised immediate questions about whether Washington might use access to its technology solutions, even those owned by commercial providers, as a coercive tool in negotiations with other supposed partners in Europe, as well as the trustworthiness of especially military systems which rely on US software support and maintenance. The Trump administration’s actions against the ICC (see above), and its later threats to raise tariffs against the EU and the UK if they did not agree to relax their regulation of US tech companies, brought further momentum to the European push towards strategic autonomy.
Recent threats against the territorial integrity and sovereignty of Greenland, and rising questions about the US commitment to NATO, added further urgency to this debate, and its implications for Europe’s position to withstand coercive pressure or even a potential military incursion, when its tech systems are not under its full control. Several European governments have since 2025 expressed concern that, for example, military cloud access could be turned off during a conflict or that platforms like F-35 fighter jets supplied by the US could similarly be subject to ‘kill switches’ in the form of disruptions of software and maintenance support. The US CLOUD Act, which allows US law enforcement to compel US companies to disclose user data, even if physically stored abroad, is of particular concern.
These flashpoints have sparked serious private, civil society and government efforts across Europe to diversify its AI and technology stacks, reduce weaponizable external dependencies and build and drive adoption of its own solutions. Strategic autonomy in AI has become an area of strategic focus in European Commission agendas, and featured prominently during Commission president Ursula von der Leyen’s ‘State of the Union’ address in September 2025. Notable examples include the EuroStack initiative, which brings together a wide range of European defence and technology companies, to build and implement sovereign European solutions, especially in AI. German chancellor Friedrich Merz and French president Emmanuel Macron hosted a joint summit in November 2025 to put action behind European digital sovereignty, with both leaders stressing the vital importance of strengthening Europe’s own AI and defence tech systems. The Austrian army migrated its IT solutions from a US provider to a European alternative, while the Dutch Ministry of Defence announced in 2026 that it would build a homegrown, sovereign cloud in an explicit bid to reduce its reliance on the US. The French government is currently in the process of transitioning away from US conferencing services towards European-made, open-source alternatives. Dutch technology giant ASML, meanwhile, invested €1.3 billion in the French AI company Mistral AI, explicitly citing the need to enhance European strategic autonomy.
Such developments are still at an early stage. The depth of entanglement, plus the lack of a clear, shared vision for how to address the challenge among European governments and companies, mean that Europe is unlikely to wean itself of US technologies in the immediate term. Moreover, some of the dependencies – especially in the military realm – may take over a decade to replace. Similar concerns surround the continent’s increased dependence on Chinese tech solutions and supply chains.
The increasingly urgent and public discussion of strategic autonomy is nevertheless significant. An autonomous European AI sphere is becoming more tangible, especially as governments and companies become more willing to pay the increased costs involved. This is not a dynamic confined to Europe alone, but one especially visible in countries previously reliant on both the US security umbrella and relatively open, globalized supply chains.
Rhetoric and actions on sovereignty are also likely to become a growing source of friction between those smaller players and the current leaders in the AI race. As global efforts to decouple become more explicit, Washington and Beijing will likely become more concerned about the market positions of their respective technology champions.
How this trend can lead to multipolarization
More countries will pay the ‘sovereignty premium’ and embrace domestic solutions over the cutting edge
The push towards decoupling and developing sovereign AI stacks could see more countries favour domestic champions in their procurement and start to prefer products from national providers over frontier solutions from politically unreliable external markets. This combination of increased spending, industrial policy, the use of government and private sector purchasing power to shape the marketplace, and an embrace of ‘good enough’ solutions over those at the cutting edge of development may provide those markets that are currently lagging behind with a way to catch up with the leaders.
2025 provided some early evidence of this dynamic in action. For example, European satellite-maker Eutelsat’s OneWeb constellation had previously struggled to compete with large US-based market leader Starlink, but saw its stock price surge and order book expand, as European countries and others grew increasingly concerned that a dependency on US companies could be weaponized against them. Geopolitical reliability trumped immediate performance in this equation – as Starlink’s constellation remains by far the most mature LEO satellite solution. In October 2025, Greenland signed a deal with Eutelsat to provide connectivity to the island. Demand has also come from outside of Europe. Taiwan, which is investing in LEO satellite systems to provide back-up connectivity during a possible invasion, has also turned to the European alternative as one of the more geopolitically secure options. Similar evidence emerged in the European cloud storage market, where local providers have seen their market grow. Though some concerns about friction, scalability and availability of many of these alternatives remain, countries are likely to become increasingly willing to pay a ‘sovereignty premium’, and accept more expensive or less mature but geopolitically reliable solutions over the less secure technological cutting-edge.
These efforts may not stay limited to the markets embracing sovereignty. A more wholesale change in global market preferences could emerge, as also other markets start to prefer technologies from trusted, allied partners over those provided by countries that have shown themselves willing to weaponize their dominance in technology and supply chains for their own geopolitical ends.
Distrust between the US and its traditional partners may grow
European attempts to strengthen its own technology stack and defence-industrial base were received unfavourably by the US, with Secretary of State Marco Rubio stating that ‘any exclusion of US companies from European tenders would be seen negatively by Washington’. Since then, the Trump administration has increased the pressure on leaders in Canada, Europe and Asia to continue to favour US technology, including through coercive mechanisms – for example, the threat of tariffs.
While aggressive measures may work in the short term, they may also become an increased source of geopolitical tension, and are likely to spur further fragmentation and decoupling in the longer term. A case study of how overplaying a dependency may expedite decoupling is China. China has long had the ambition to become fully self-reliant across the technology sphere and beyond – driven by important historical factors, as well as increasingly far-reaching restrictions imposed on Beijing by the US and its allies on the country’s ability to access to high-end semiconductors, microchips and other strategic inputs critical for the development of frontier AI and other technologies.
In 2025, China’s apparent success in indigenizing its AI supply chains caused alarm in Washington, and resulted in the Trump administration further tightening restrictions on the export of semiconductors and other inputs. Though Chinese leaders recognize that these restrictions have hampered China’s AI development and hindered its ability to fully reap the rewards of military AI deployment and other use cases, the Chinese technology industry has made rapid progress both in finding ways around these curbs and in expanding its own semiconductor and chip production to compensate.
Experts disagree about whether US-led efforts to prevent China from accessing frontier AI will ultimately prove effective. The pace of development in chips used for training AI systems is non-linear. The speed at which performance and quality improves will limit China’s ability to catch-up, meaning that the gap between the AI computing resources that China’s AI companies are able to access and that available to the US and to its allies, is only set to increase.
While attempts to maintain a hold over others may be effective in the short term, such influence will diminish over time. This dynamic could then lead to a multipolar AI marketplace, where chokepoints are less geographically concentrated.
For some of these weaponized chokepoints in AI supply chains, alternatives will take many years to develop – with some analysts suggesting that for some it may not be possible at all. The most profound challenge is presented by ASML’s ultra-violet lithography (EUV) machines. This highly specialized technology is vital for high-end chip production and can credibly claim to be the most complex machine ever made. But ASML’s most advanced technology has been subject to US-led export restrictions since 2019, prompting state-funded efforts by China to develop its own leading-edge EUV machines. Despite the complexity of the task, scarcity can lead to rapid innovation: in December 2025, Reuters reported that China’s ‘Manhattan Project’ to build its own version of ASML’s machines had made significant technical breakthroughs.
Though it seems that this, and similar, reporting, may oversell China’s progress made to date, the lesson is clear: weaponizing a technological dependency will motivate those on the receiving end to try and free themselves. While attempts to maintain a hold over others may be somewhat effective in the short term, such influence will diminish over time. This dynamic could then lead to a multipolar AI marketplace, where chokepoints are less geographically concentrated. China again provides an example of this, as it is at risk of losing one of its own – in critical mineral processing – as a result of its own overzealous use of this coercive lever.
Trend 4: Growing concerns about a potential AI valuation bubble
The AI race is usually presented as a binary one between the US and China, not only due to the significant technological lead that existing companies in those countries already enjoy, but also the vastly larger amounts of spending and computing power that the US in particular is able to mobilize.
By some estimates, US-based companies attracted 75 per cent of total global VC investment in AI in 2025, with AI capturing 61 per cent of all global VC spending in that year. By contrast, the combined EU attracted only 6 per cent. The UK and China accounted for 5 per cent each. (China’s AI ecosystem is far more reliant on government funding than any of the other examples – of the estimated $98 billion invested in AI in China in 2025, around $56 billion came from government funding.)
AI infrastructure investment from different sources is also dominated by the US. China and the US especially also dominate AI infrastructure spending and availability. In mid-2025, around 75 per cent of high-performing GPU clusters were in the US and 15 per cent were in China, while other markets had much smaller shares.
No other market players or ecosystems are currently in a credible position to match the levels of US private sector spending and reproduce existing market fundamentals. A growing share of these investments, especially in the US, is being allocated to spending on infrastructure such as data centres, which many AI developers and investors argue is a necessary precondition for the development of frontier AI. If the AI race is considered as constituting three parallel races – the race towards the frontier (the development of the most cutting-edge AI), the race towards diffusion (spreading AI throughout the economy), and the race towards application (finding different use cases for the technology) – this spending ‘arms race’ primarily treats the first of those as the key objective. A market correction in response to growing concerns about the ability of AI investment to return a profit may revise some of these fundamental assumptions – moving the focus towards the rapid, often far cheaper, diffusion and application of the technology. With such a revision, the overall AI race could be significantly reconfigured.
Throughout 2025 and early 2026, major investment banks and market analysts issued warnings that the recent surge in AI company valuations and the vast amounts of AI infrastructure spending presented all the characteristics of a market bubble. These warnings claim that valuations and spending appeared increasingly divorced from the financial forecasts on the technology’s ability to generate returns – especially as monetizable adoption has yet to match expectations. In 2025 alone, the major hyperscalers – including Alphabet, Amazon’s AWS, Microsoft, Meta and Oracle – spent at least $300 billion on new AI infrastructure and computing hardware such as data centres, access to GPUs, chips and servers, as well as power and cooling infrastructure to enable energy-intensive AI training and inference. More recent projections suggest that this spending could reach $700 billion in 2026. New investments by hyperscalers and other US AI champions like OpenAI and Anthropic reached such heights in 2025 that they accounted for almost 2 percentage points of US GDP growth, and, by some estimates, may have kept the US economy from entering a recession.
While AI is an important growth industry in markets outside of the US (and much of the physical infrastructure rollout mentioned above is not confined to the US, even if many of the investors and owners are), levels of spending in most markets elsewhere in the world have been more modest (certain investors in the Gulf and large-scale manufacturers of AI-enabling goods in Asia excepted).
This US-led infrastructure ‘arms race’ could pose a significant risk of over-investment if demand slows or the technological promise of AI is not met. Throughout 2025 and early 2026, analysts expressed concern that the valuations of companies active in AI were overly optimistic, and that investors were pouring capital into AI firms based on future revenue expectations that were far from certain to ever materialize.
One particularly significant factor in this, with relevance to this paper’s discussions of the ever-closer entanglement between governments and AI companies, is the challenging economics of AI investment. While the technology itself continues to rapidly improve, and will likely have a profound societal impact, monetizing it remains a challenge.
Deutsche Bank released an analysis in 2025 that suggested the major AI players would need to make up a shortfall of around $800 billion by 2028 to recoup the investments made to date. AI companies’ total revenues in 2025, however, remained far lower. Persistent uncertainty and supply-chain shocks resulting from the ongoing war in Iran, which threaten to delay AI infrastructure build-outs, negatively impact chip manufacturing and put into question the long-term financial commitment of the Gulf Arab countries to fuelling global AI investment, similarly threaten to undermine assumptions behind the current AI market rally.
Predicting the potential outcomes and future trajectory of spending on AI is difficult. Current high levels of investment in AI infrastructure and solutions may persist, especially if new, higher-revenue applications or technological breakthroughs lead to mass, monetizable adoption. Nonetheless, concerns about a bursting of the AI ‘bubble’ are already affecting AI investments and have led investors to pivot to different sectors – including defence tech.
How this trend can lead to multipolarization
The need to find reliable sources of capital may lead to more ‘patriotic tech’ and government–private sector entanglement
The growing interest in the defence applications of AI, as well as its importance in geostrategic competition, could enable the current levels of spending to be sustained for longer, as the beneficiaries come to be seen as ‘too big to fail’ from a geostrategic point of view. AI companies – especially those focused on developing frontier AI models – already have a powerful incentive to promote their solutions as geostrategic tools. If commercial growth opportunities start to disappear, further securitizing the technology and turning to government becomes an increasingly attractive alternative.
AI companies are already trying to find new sources of revenue and new customer bases, in order to generate a sustainable return on the enormous investments already made. Dual-use applications of their solutions appear to be among those that companies are exploring, as discussed in previous sections.
The announcement of ‘Project Stargate’ – a now apparently dormant commitment to spend $500 billion on new AI infrastructure – on the first day of President Trump’s current term was similarly framed as a public–private partnership to keep the US in the lead on AI.
Although the military segment of the market remains small and is not currently able to offset the high levels of general AI capital spending, it is nevertheless growing. Deeper entanglement and growing mutual interdependence between AI companies and the government will also help further position these companies as strategically important. It may also lead governments to take over some of the infrastructural spending, should the market alone no longer be able to deliver. OpenAI CEO Sam Altman has frequently described his company’s mission as akin to a ‘Manhattan Project’, with winning the race for AGI as the single most geostrategically important objective for the US. The announcement of ‘Project Stargate’ – a now apparently dormant commitment to spend $500 billion on new AI infrastructure – on the first day of President Trump’s current term was similarly framed as a public–private partnership to keep the US in the lead on AI. The involvement of AI companies and AI infrastructure providers like Nvidia in the export-focused aspects of the US government’s AI Action Plan, which explicitly focuses on exporting the US AI stack to like-minded nations as a geostrategic instrument, also helps create commercial opportunities for these companies and can generate vendor lock-in in new markets.
This dynamic may further encourage the growing closeness between governments and the AI sector – which could expedite the trends towards patriotic tech, global distrust and sovereignty discussed in previous chapters.
The market may shift towards cheaper, open-source models if frontier models fail to deliver returns
Though no market will be immune from the wider economic fallout of a market correction in the US, not all AI ecosystems are equally exposed to this risk. China and, to an extent, the EU and certain middle powers like India, have placed an emphasis on the rapid diffusion of leaner, cheaper and frequently open-source models over the highly capital-intensive AI frontier. Races over technology have not infrequently been won by the countries best able to implement and diffuse an innovation, rather than those that initially developed it. The AI bubble popping could validate this point again.
China’s open-source AI players would be among the main beneficiaries of such a scenario, as already, their models do not significantly lag behind – and occasionally outperform – far more capital-intensive US models. Recent events demonstrate this dynamic in action. In the aftermath of the January 2025 launch of China’s DeepSeek R1 model – a relatively cheap-to-train, open-source AI model, which by most metrics was only modestly behind more expensive US ‘frontier’ models in performance – suggested that high-performing AI models could be developed without spending big – or even having access to the most advanced AI chips.
The ensuing brief market panic over DeepSeek suggests that any bursting of the AI market bubble could cause a more permanent shift towards cheaper, less capital-intensive innovation, which could make it possible for other actors to compete. Since then, many more even higher performing Chinese open-weight and open-source models have been launched.
Smaller countries may have a ‘second-mover advantage’
In the aftermath of the AI bubble bursting, more established AI players could end up the beneficiaries of a correction through the reduction of the competition, consolidation of their market share and ability to hoard cheap infrastructure, intellectual property and talent. This could make it even more difficult for others to catch up. The dynamic could, however, also move the market in the other direction.
Countries currently behind the leaders in the AI infrastructure rollout may find themselves with fewer stranded assets and access to cheaper computing power and other AI resources. This may allow them to avoid the costly mistakes made by others and integrate AI across their wider economies more efficiently and cost-effectively – what is known as a ‘second-mover advantage’. Previous infrastructure bubbles, such as the one that led to the dot-com crash in early 2000s, suggest that a widespread market collapse would leave a lot of stranded assets, unused infrastructure and resources, which other actors could then acquire at low cost.
Smaller competitors may also benefit from the greater availability of talent, with a lack of skilled AI engineers presently one of the main barriers to companies seeking to build or make use of the technology. Governments and militaries may also be able to bring more AI talent into their ranks. Such a reallocation could speed up dual-use AI development and deployment, and could further diversify (and securitize) the industry.
