Executive summary
In the space of a few weeks, AI has helped create the world’s first trillionaire, the United States government has blocked its own allies from accessing its most advanced AI models, and serious proposals have emerged for Washington to take direct equity stakes in leading AI companies. Three firms, all American, now account for around nine in ten visits to the leading AI LLMs between them. Taken together, these developments point towards AI concentrating wealth and influence in a small number of hands, rather than spreading it broadly across society and across the world.
We believe quantum computing will accelerate and deepen the same dynamic rather than offset it. Quantum computing today is concentrated in an even smaller group of countries and companies than AI, and that group is unlikely to expand much before the technology becomes commercially and strategically important. At the same time, the timeline for quantum computing to matter is compressing, partly because today’s AI systems, themselves controlled by a handful of firms, are already being used to help solve some of quantum computing’s hardest engineering problems.
The export control playbook that the United States has applied to AI, most visibly in its recent treatment of Anthropic, is also being applied to quantum computing, and we expect other quantum capable countries to follow a similar approach given how directly quantum computing threatens existing cyber security and cryptography.
Within a decade, and quite possibly well before that, we expect a small number of countries, companies and individuals to control the most powerful computing technologies that have ever existed, covering both artificial intelligence and quantum computing. History offers very little precedent for the kind of voluntary sharing of power and wealth that would be needed to prevent this concentration translating into a permanently wider gap between those who hold this technology and those who do not.
1. The optimistic case, in brief
The “golden age” argument, as applied to AI, runs roughly as follows. AI is a general purpose technology, like electricity or the internet, and its costs will fall quickly while open source models continue to narrow the gap with the most advanced systems. On this view, AI could become the most widely shared technological advance in history, lowering the cost of expertise such as legal advice, medical diagnosis, education and administration for everyone, everywhere, including those who currently have the least access to such things.
A similar, and in some ways even bolder, version of this argument is now being made about quantum computing. Quantum computers, once they reach sufficient scale, are expected to transform fields such as drug discovery, materials science, logistics and climate modelling, with benefits that could eventually flow to any country or organisation able to access the technology, whether by owning it outright or by using it through the cloud. Advocates point out that earlier waves of computing, from mainframes to personal computers to mobile phones, did eventually spread far beyond the countries and companies that first developed them.
These arguments are reasonable as far as they go, and they are not without precedent. Mobile phones and the internet did spread more widely, and more quickly, than many observers expected forty years ago, including to countries that had no role in inventing either technology. However, a technology spreading geographically is not the same thing as the wealth, power and strategic advantage it creates spreading in the same way. The rest of this briefing focuses on that distinction, first for AI and then, in more detail, for quantum computing.
2. AI: the evidence so far
2.1 A handful of firms hold the keys
Most people, businesses and governments now access AI through tools built and run by a small number of companies. As mentioned, as of June 2026, the three largest AI chatbots, ChatGPT, Google’s Gemini and Anthropic’s Claude, account for roughly nine in ten visits to the leading AI assistants between them. This level of concentration is not new in the technology sector. Search and social media went through a similar process, but AI is different in one important respect. It increasingly sits close to decisions about information, infrastructure and national security, so whoever controls the platform has influence that extends well beyond the technology sector itself.
It is important to directly call out what some consider to be a serious counter-argument. By several measures it appears that the market is becoming less concentrated, not more: ChatGPT’s share of visits has fallen sharply over the past year as Gemini and Claude have grown, and open-weight models, most visibly DeepSeek but also Llama and Mistral, now approach the frontier at a fraction of the cost and can be downloaded, run locally and fine-tuned outside the control of any single provider. If that trend holds, the resource that determines access becomes cheaper and more widely distributed over time, which cuts against the idea of a widening divide.
However, the reason we still expect concentration to dominate is that the level that matters keeps moving. Frontier capability, the export controls described below, and the capital required to train and serve the largest models all rise together, so cheaper open models reduce dependence on any one vendor without changing who defines the frontier or who can be cut off from it. Diffusion at the application layer and concentration at the infrastructure and capital layer are not in tension; they are the same market maturing along two axes at once.
2.2 The trillionaire moment
On 12 June 2026, SpaceX listed on the Nasdaq in the largest initial public offering in history, raising $75 billion. Combined with his existing stake in Tesla, this took Elon Musk’s net worth past $1 trillion, making him the first person ever to cross that threshold. The headline is about one individual’s wealth, but the underlying point is broader. Markets are placing enormous value on a small number of companies, and people, at the centre of AI, computing and space infrastructure, and when that much value concentrates this quickly in so few hands, it becomes harder for everyone else to catch up, not easier.
2.3 AI as an instrument of state power
On 13 June 2026, the United States government ordered Anthropic to cut off access to its newest models, Fable 5 and Mythos 5, for all foreign nationals, including Anthropic’s own foreign employees wherever they are based, citing national security concerns. Anthropic complied, taking both models offline worldwide, while objecting publicly that the process behind the decision lacked transparency.
Two weeks later, on 25 June, The Trump administration asked OpenAI to limit the release of its new GPT-5.6 model to select government-approved partners before any wider launch, citing security concerns stemming from the model’s capabilities. OpenAIs CEO, Sam Altman told staff that this was the best path to getting 5.6 released, with a general release likely “a couple of weeks later.” He also said that he had made it clear to the Whitehouse that this is not the preferred long-term model, and OpenAI will work toward a more sustainable release approach.
Whatever the merits of these decisions, they demonstrate something significant. The US government can now decide, with immediate effect, who in the world gets access to the most capable AI available, which means access to AI is no longer purely a commercial choice made by companies but has become a lever of foreign policy, alongside trade tariffs and export controls on chips and weapons. Nathan Benaich, an AI investor at London-based Air Street Capital, put the point plainly in the Financial Times this week: “The most advanced AI is built by a handful of American companies, on American soil, under American law and what the rest of us are permitted to do with it can change on a Friday afternoon.”
2.4 The state and the company are becoming harder to separate
Earlier in June 2026, President Trump said his administration was exploring taking equity stakes in leading AI companies, so that, in his words, the American public “becomes a partner” in their success. Separately, a bill has been proposed in Congress for the US government to take a fifty per cent stake in leading AI firms. Neither proposal has been agreed, but the direction of travel is clear. For other countries and companies, it changes the calculation, because buying access to leading US AI is no longer simply a commercial transaction with a private company. It increasingly means doing business with an extension of the US government, one that has just shown it is willing to use AI access as leverage even against its own allies.
Sam Altman, OpenAI’s chief executive, has taken this logic a step further: in early July the FT reported that he has offered the US government a direct five per cent equity stake in OpenAI, a move that, if accepted, would make the US government a financial beneficiary of OpenAI’s commercial success and further blur the line between a private technology company and an instrument of US state power.
2.5 The UK and Europe admit they are behind
The UK government has launched a £500 million Sovereign AI Fund, live from 16 April 2026, explicitly aimed at reducing reliance on overseas AI infrastructure, although the UK remains heavily dependent on foreign owned computing power, chips and data infrastructure, so the fund is best seen as a first step rather than a fix. The European Union has gone further, and in early June 2026 the European Commission set out a Technological Sovereignty Package covering chips, cloud and AI, open source software, and AI’s role in energy. Over half of large Western European businesses say they plan to speed up work on data sovereignty and reconsider their reliance on non European cloud providers. These moves matter because of what they admit. Both the UK and the EU, wealthy and technically capable regions, are spending significant public money because they have concluded that dependence on US AI infrastructure is now a strategic risk, and if the UK and EU feel this exposed, the position of lower income countries, with far less capacity to respond, is considerably worse.
The episode has since reached the highest level of international discussion. Following the Anthropic ban, AI sovereignty featured on the agenda of the G7 meeting in the French Alps this month, confirming that the question of who controls access to advanced AI has moved beyond industry debate into the formal language of great power diplomacy. The scale of the challenge facing countries outside the United States is visible in the numbers: Anthropic’s most recent funding round, at $65 billion, was roughly equivalent to the total invested by the entire European venture capital sector across the whole of 2025. Against that backdrop, the gap between the leading AI powers and everyone else is not merely a technical question but a function of capital concentration on a scale that public spending programmes have so far struggled to match. Al Carns, the Labour MP who resigned as Minister for Armed Forces on 11 June 2026 over the government’s defence-spending plans, days before the Anthropic decision, captured the practical reality in terms that apply well beyond the UK: “We have the researchers, the universities, the standards. What we don’t have is the power stations to run the data centres, the planning system to build them, or the industrial base to make the chips. So the work happens here and the value lands somewhere else. We invent. Others build. Others decide. Then we read about it on Saturday morning.”
2.6 Developing countries face a widening gap with fewer options
Building or buying AI at scale requires three things in combination: capital, computing infrastructure and skilled people, and many developing countries have limited access to all three. As AI becomes more central to economic growth, public services and even military capability, countries that cannot compete risk becoming structurally dependent on a small number of AI suppliers, almost all American, and increasingly intertwined with the interests of the US government. This is not a new pattern in global affairs, but AI raises the stakes considerably, because the technology touches almost every part of the economy and the state, and because the gap in capability is widening quickly rather than slowly.
3. Quantum computing follows the same pattern, faster and from a narrower base
3.1 Where quantum computing exists today
The United States remains the largest single base, combining government funding through the National Quantum Initiative, the Department of Energy and the Department of Defense with private companies such as IonQ, Rigetti, D-Wave, PsiQuantum, QuEra, Infleqtion and Atom Computing. China has invested heavily through its National Laboratory for Quantum Information Sciences and the Chinese Academy of Sciences, with companies including Baidu, Origin Quantum and QuantumCTek, and has reportedly committed around $10 billion to its national quantum laboratory alone, alongside a much larger state venture fund covering quantum and other advanced technologies.
The United Kingdom has committed £2.5 billion over ten years under its National Quantum Strategy, supporting companies such as Quantinuum, Oxford Quantum Circuits, Universal Quantum, Riverlane, ORCA Computing and PQShield. France, Germany and the Netherlands form the core of a European cluster, home to companies including PASQAL, Alice and Bob and Quandela in France, IQM and planqc in Germany, and a Dutch national programme, Quantum Delta NL, backed by around 615 million euros. South Korea has committed roughly $2.3 billion towards quantum computing by 2035, and Japan announced semiconductor and quantum investment of around $7.4 billion in 2025, one of the largest national commitments outside the United States and China. Beyond this core group, a longer tail of countries, including Canada, Australia, India and Singapore, run smaller quantum programmes.
3.2 The timeline for quantum computing is compressing
The mainstream estimated timeline for when quantum computing might become genuinely useful increasingly points to around 2029 to 2030. The constraints are well understood. Today’s leading quantum machines have dozens of stable qubits, while error corrected systems that could outperform classical computers on commercially useful problems are thought to need somewhere in the region of a thousand. That is a gap of roughly a hundred times, but it is widely seen as an engineering problem on an actively pursued path, rather than one waiting on an unknown scientific breakthrough, which is part of why the 2029 to 2030 range deserves serious attention rather than being dismissed as hype. That said, the estimate should be held within context: scaling to error corrected, fault tolerant systems is not guaranteed to be linear, several hardware approaches remain in contention, and past quantum timelines have slipped more often than they have held. A slip of several years in either direction would not be surprising, and the argument that follows does not depend on the precise date, only on the direction of travel.
3.3 The AI and quantum computing loop is accelerating both
A frequently cited reason the timeline may compress is what we have previously described as a compounding loop between AI and quantum computing, though how much this has actually moved the timeline so far is asserted more often than it is measured. At present, the relationship is asymmetric. AI is already being used by leading laboratories to help solve some of quantum computing’s hardest engineering and error correction problems, effectively borrowing the pattern recognition and optimisation strengths of large AI models to speed up progress on quantum hardware. The reverse effect, in which mature quantum computers remove some of the computational ceilings that currently constrain AI training and inference, is expected to follow later, once quantum systems reach sufficient scale. The practical consequence is that the organisations best placed to apply AI to quantum research, which are largely the same small group of companies that already dominate AI, are likely to extend their lead in quantum computing too, reinforcing rather than disrupting the existing concentration of capability.
3.4 The export control playbook is already being extended to quantum computing
If the Anthropic episode showed how quickly the United States can restrict access to frontier AI, even for its own allies, there is good reason to think the same approach will apply to quantum computing, and to a significant extent it already has. In September 2024, the US government introduced export controls covering quantum computers, related components and associated software. From January 2025, US investors have been barred from investing in China’s quantum computing, quantum sensing and quantum communications sectors. In March 2025, the Trump administration added around eighty companies to its export blacklist, more than fifty of them Chinese, several with quantum links. Most recently, in January 2026, the Remote Access Security Act passed the House of Representatives by 369 votes to 22, and would extend these controls to remote and cloud based access to quantum computers, closing a loophole that had allowed China Telecom’s Tianyan quantum cloud platform to attract more than 37 million visits from over sixty countries.
Two things stand out about this timeline. The first is how early it has happened. These controls were put in place years before quantum computers are expected to be commercially useful, which suggests governments already see quantum computing’s value, and risk, for cyber security and code breaking as significant enough to justify restricting access now, well ahead of any civilian payoff. The second is the response from China, where the controls appear to have accelerated rather than slowed domestic development, with China’s quantum industry reportedly growing more than thirty per cent a year and reaching around 11.56 billion yuan in 2025, partly by building out its own supply chain for the components it can no longer easily import.
Given the cyber security stakes involved, and the precedent the United States has now set twice, on AI models and on quantum hardware, we would expect the United Kingdom, the European Union, Japan and South Korea, all of which have their own quantum programmes and their own cyber security concerns, to introduce comparable controls of their own over the coming years, whether in coordination with the United States or independently. The effect, in either case, is the same. It narrows further who gets access to the most capable systems, and it increases the likelihood that the world divides into separate technology blocs with different rules about who can buy, sell or even remotely access the relevant hardware and software.
4. How the divide gets enacted. Resources, security and political alignment
4.1 The Democratic Republic of Congo
In December 2025, the United States and the Democratic Republic of Congo signed what has been widely described as a “security for minerals” deal, giving the US preferential access to Congolese reserves of copper, cobalt, lithium and gold in exchange for security guarantees and significant influence over the country’s mining sector. The DRC accounts for around fifty-five per cent of the world’s known cobalt reserves and supplies more than seventy-four per cent of global cobalt, a material critical not only to batteries but to many of the components used in advanced computing hardware, including quantum computers. The deal required the DRC to amend its laws and potentially its constitution, and has since been challenged in court by Congolese lawyers and human rights groups who argue it was agreed without adequate domestic scrutiny and compromises the country’s sovereignty over its own resources.
4.2 Venezuela
Venezuela provided a more dramatic version of the same sequence. On 3 January 2026, United States forces captured Nicolas Maduro and transported him to New York to face narco-terrorism charges. His successor and former deputy, Delcy Rodríguez, initially pushed back against Trump’s public ultimatum but within weeks moved to comply with core US demands, reforming Venezuela’s hydrocarbon laws to open its oil sector to American companies and allowing the sale of Venezuelan crude, including the 30-to-50 million barrels of previously sanctioned oil Trump had demanded, with the proceeds settling in US-government-controlled accounts. The United States eased sanctions in parallel and has indicated it intends to use its position to direct future oil sales and limit the role of China and Russia in Venezuela’s energy sector.
Whatever view one takes of the circumstances of Maduro’s removal, the result was a swift realignment of a country’s natural resources towards the leading power, on terms set by the latter.
4.3 A pattern that connects directly to the technology divide
Taken together, the Democratic Republic of Congo and Venezuela illustrate a pattern that is likely to become more common as AI and quantum computing increase the strategic value of both computing power and the raw materials needed to build it. The countries best placed to offer security guarantees, market access, technology and political legitimacy are overwhelmingly the same countries that already lead in AI and quantum computing. The countries holding many of the critical minerals needed for advanced computing hardware, including cobalt, lithium, copper and various rare earth elements, are often those with weaker institutions, less negotiating leverage and, in some cases, leaders whose continuation in power benefits from external support regardless of how that support is obtained. The consequence is a transfer of strategic resources from those who hold them to those who can convert them into computing power, on terms set by the latter.
We expect the leading technology powers to secure access to these materials increasingly through bilateral deals of this kind, rather than through open markets, and we expect those deals to be one-sided and packaged together with the decisions about technology access and export controls discussed earlier. The resource-for-security pattern and the technology divide are not two separate phenomena. They are two sides of the same dynamic: a world in which power over the most transformative technologies, and control of the raw materials needed to sustain them, are concentrating in the same hands, with the terms of access for everyone else set by those who hold both.
5. Choosing a technology bloc. The dilemma for the rest of the world
A world of distinct technology blocs is not a future scenario but a direction that is already taking shape. What remains less examined, because it tends not to generate headlines in the countries at the centre of this process, is what bloc formation means for the much larger number of countries that are not leading in AI or quantum computing and do not have the resources to do so on their own terms. For these countries, the consolidation of technology power is not an abstraction. It is, increasingly, a concrete set of decisions about whose infrastructure to use, whose models to build on, whose data governance standards to adopt, and, as a consequence, whose political relationships to deepen. Countries that have historically managed to trade with all sides and commit to none are finding that technology alignment is making that position progressively harder to sustain.
China’s approach to AI in the Global South is deliberately different from the American one, and in ways that are proving effective. Where American AI companies offer powerful but expensive and proprietary products, subject to export controls and political override as the Anthropic episode showed, China is offering developing countries open-source models that can be downloaded, modified and deployed without expensive licences, without ongoing dependence on a cloud provider that can cut access, and without the risk of being shut out if geopolitical relations deteriorate. DeepSeek, China’s open-source large language model, costs around ninety four per cent less to run than comparable American models and has been actively distributed across Africa by Chinese companies, notably Huawei and Alibaba, putting AI within reach of businesses and public institutions that could not previously afford Western alternatives.
As The Wire China reported in June 2026, China is effectively “surrounding American AI from the South”, building an AI ecosystem in the Global South that blends open-source models, Chinese cloud infrastructure and Chinese technical partnerships in a pattern that echoes the Belt and Road infrastructure strategy that preceded it. The BRICS bloc has formalised this approach at a governmental level, adopting an AI governance declaration in July 2025 that explicitly frames digital sovereignty and the “right to development” as alternatives to Western AI governance frameworks, backed by a $5 billion New Development Bank fund for AI infrastructure across member states.
The appeal of this approach for developing countries is real, but it carries its own form of dependency, one that mirrors rather than dissolves the risks of the Western alternative. Open source models may be freely downloadable in principle, but using them at scale in practice typically involves the support of Chinese companies, the use of Chinese data centres, and data flows that remain accessible to the Chinese government. Huawei’s role in building 4G and 5G networks across Sub-Saharan Africa created an embedded dependency of exactly this kind, one that the US and its allies have spent years urging African governments to remove, largely without success. In practice, the choice available to most developing countries is not between dependence and independence but between two different forms of dependence, with different risks, different political relationships and different strings attached.
5.1 The South African example
South Africa sits at the sharp end of this dilemma, for reasons that make it the clearest illustration of a tension that many African countries face in some form. The South African government, led by the ANC, pursues a formally non-aligned foreign policy but has moved steadily closer to China and Russia in practice: it is a BRICS founding member and, together with Egypt, it proposed the BRICS AI Ethical Charter in 2024, explicitly positioning BRICS as an alternative AI governance framework to Western institutions.
In the technology infrastructure space, Huawei has deployed more than 2,800 5G base stations in South Africa, and MTN, one of South Africa’s largest mobile operators, has expanded its 5G coverage substantially in partnership with Huawei, while Huawei’s dominance of Africa’s 4G backbone, estimated at around seventy per cent of the continent’s infrastructure, means Chinese technology is already deeply embedded across the South African network. Removing it, even where Western governments have encouraged this course, would be a major engineering and financial undertaking with no straightforward solution.
At the same time, South Africa’s corporate economy, the most developed on the African continent, remains deeply embedded in Western financial markets and regulatory frameworks. South Africa is the largest beneficiary of the US African Growth and Opportunity Act, which gives South African exporters preferential access to American markets, and the prospect of losing that access has already been used as a lever of American political pressure. The Johannesburg Stock Exchange attracts significant Western institutional investment, and South Africa’s largest companies across banking, insurance, mining and retail maintain close ties to London, New York and European capital. The corporate picture is further complicated by the fact that alignment in this space is not straightforward: Naspers and its international holding vehicle Prosus, among South Africa’s best-known corporate names, hold a substantial stake in Tencent, giving one of the country’s most prominent corporate groups direct financial exposure to China’s technology sector while remaining listed on Western exchanges. This corporate world, which provides the tax base, skilled employment and foreign investment the South African economy depends on, has significant practical reasons to remain aligned with Western standards, even as the government’s foreign policy pulls in a different direction.
The arrival of AI and quantum computing as instruments of geopolitical leverage is sharpening this existing tension in ways that will become harder to manage over time. A South African government that continues to deepen AI and technology ties with China through BRICS, embeds Chinese AI models and infrastructure in its public sector, and aligns with a Chinese-led data governance framework, will increasingly create friction with the Western corporate world it depends on for investment and growth. Conversely, a South African corporate sector that aligns firmly with Western AI providers and standards may find it harder to operate in an environment where government policy and public infrastructure point in a different direction, and where the costs of a future course correction would fall heavily on the private sector. The US has already signalled its willingness to use economic access as a lever of political influence over South Africa, through the repeated uncertainty surrounding AGOA renewal and the broader tensions in the relationship during the Trump administration. When technology alignment adds another dimension to that pressure, the question for South Africa becomes how long a country can sustain a government that is functionally non-aligned and a corporate sector that is functionally Western, and often subject to Western regulation, before the gap between those two positions becomes untenable.
This is not merely a question of political preference but a practical problem about how technology systems work. A country that allows Chinese AI models, Chinese cloud infrastructure and Chinese data governance standards to become embedded in its public services, universities and corporate supply chains will find it progressively harder to pivot towards Western standards later, even if political circumstances change, because the investment already made in systems, skills and relationships does not translate easily across technology blocs. The same is true in reverse. This path-dependency problem means that the first significant alignment choice forecloses many of the subsequent ones, and that a country which drifts into alignment by default rather than making a coherent choice is likely to end up with systems that serve neither bloc well, at real cost to the efficiency and competitiveness of its economy. For South Africa, where the corporate world and the government may be pulling in genuinely opposite directions, the risk is not just of choosing poorly but of failing to make any coherent choice at all.
South Africa is far from alone in this position. Many African countries have benefited substantially from Chinese infrastructure investment through Belt and Road and its successors, even where their corporate sectors and financial systems retain close Western ties. DeepSeek and Huawei’s AI offerings are already out-competing American alternatives in parts of Africa where cost and access matter more than geopolitical alignment, and a sustained period of American disengagement from the continent has allowed Chinese technology infrastructure to become embedded in ways that will not easily be reversed. What makes South Africa the sharpest illustration of the dilemma is the particular combination of a government more explicitly non-aligned, a corporate economy more explicitly Western-integrated, and a technology and financial sector sufficiently developed that its alignment choices carry real and immediate economic consequences. As AI and quantum computing sharpen the stakes of technology alignment over the coming years, South Africa may emerge as an early signal of how the rest of the continent navigates a choice that many African countries cannot indefinitely defer.
6. A decade-long sorting, with little precedent for a different outcome
There is a useful, if imperfect, historical comparison in the way the world has handled nuclear weapons. The Treaty on the Non-Proliferation of Nuclear Weapons, signed in 1968, did not prevent the spread of nuclear weapons entirely, but it did help to establish and preserve a recognised hierarchy of nuclear armed and non-nuclear states that has lasted for more than fifty years, with only a small number of additions, namely India, Pakistan, North Korea and, widely believed but never officially confirmed, Israel, despite enormous diplomatic effort to prevent even these. The lesson is not that international agreements achieve nothing. It is that once a small group of states establishes a decisive advantage in a strategically important technology, that advantage tends to persist for a very long time, even when the rest of the world actively organises to close the gap.
AI and quantum computing differ from nuclear weapons in one crucial respect, which makes us think the resulting divide could be more durable, not less. Nuclear weapons have essentially no civilian economic value, so the incentives to control their spread have been almost entirely about security. AI and quantum computing, by contrast, have enormous civilian economic value on top of their security implications, which means the countries and companies that hold a lead have a double incentive to protect it: the security incentive that applied to nuclear weapons, and a commercial incentive to capture the economic returns for themselves and their shareholders. Where nuclear non-proliferation was driven mainly by governments managing a shared risk, the AI and quantum divide is being driven by governments managing a shared risk and by companies pursuing what are, by any historical standard, extraordinary financial returns. At present, these two sets of incentives point in the same direction, towards restricting access rather than widening it.
This is why we think the idea that the benefits of AI and quantum computing will simply spread to the rest of the world over time, as the “golden age” theory assumes, requires a degree of voluntary sharing of advantage for which history offers very little precedent.
Where genuine technology transfer has happened at scale in the past, for example under the Marshall Plan after the Second World War, it has generally served the strategic interests of the country doing the sharing, in that case helping to rebuild European allies as a bulwark against the Soviet Union, rather than reflecting altruism for its own sake. It is not obvious what equivalent strategic interest would lead the United States, China or any other leader in AI and quantum computing to deliberately narrow its own advantage over the rest of the world, particularly in a period when, as the Anthropic episode showed, the prevailing instinct is to tighten access even towards long-standing allies rather than loosen it.
That question has been brought into sharp focus by Sam Altman, OpenAI’s chief executive, who proposed in an opinion in the Financial Times on 1 July, a US-led international forum to set global AI standards, certify compliant countries and companies, and govern access to advanced AI. Altman invokes the International Atomic Energy Agency as his model. The analogy is more revealing than he may have intended. The IAEA was established by the nuclear powers to manage access to a technology that only they possessed, on terms largely set by them. The Non-Proliferation Treaty that followed enshrined a two-tier world: nuclear states that kept their weapons, and non-nuclear states that agreed not to acquire them in exchange for access to civilian nuclear energy under supervision.
If the IAEA is the template for AI governance, the outcome is a formal hierarchy in which the current AI leaders set the rules, certify compliance, and decide who gets access. That is not a solution to the divide we are discussing. It is the divide given institutional form and international legitimacy. Altman’s proposal acquires a further dimension when set alongside the fact that he has simultaneously offered the US government a five per cent equity stake in OpenAI. A company partly owned by the US government, operating within a US-led governance forum that certifies which countries and companies get access to advanced AI, is not straightforwardly distinguishable from a state instrument. For any country that has watched the US revoke access to Anthropic’s models overnight, the governance architecture Altman describes is not reassurance. It is a precise description of the world this paper has been analysing.
7. Divides reinforce divides, and the timeline is compressing
As we have argued in earlier briefings, AI does not on its own create populism or sectarianism, but it can deepen the conditions that produce both, and the same is true of quantum computing. A widening gap between technology “haves” and “have nots”, between countries and within them, adds to an existing sense that the gains from economic and technological change are not shared fairly. That sense of unfairness is already a powerful force in politics across many Western democracies, visible in debates over immigration, trade and national identity.
A country, a region or an individual that feels left behind by AI and quantum computing is also more likely to feel left behind more broadly, and to support parties that promise to look inward, protect what they have, and resist further integration with the rest of the world. At the same time, people who believe they have little or no meaningful opportunity in their home country increasingly look for ways to access opportunities in the countries that “have”, accelerating the pressures and tensions already visible in debates about immigration, both legal and illegal. And so the two sides of the same problem feed on each other, creating more, not less, social division.
What AI and quantum computing add to this picture is speed. Previous general purpose technologies, electricity, the internet, even nuclear power, took several decades to diffuse from their points of origin to the rest of the world, and that slow diffusion gave societies, institutions and international rules time to adapt, even if imperfectly. AI and quantum computing appear to be compressing a similar transition into a single decade, and almost certainly less, because, as discussed above, AI is itself accelerating progress in quantum computing, which in turn is expected to remove some of the constraints currently limiting AI. This compression matters because it leaves far less time for trade rules, international institutions or domestic political systems to adjust before the gap between leading and lagging countries becomes difficult to close.
Combine this with the resource for security pattern, and the likely shape of the next decade becomes clearer. We would expect to see a small number of technology and security blocs forming around the leading powers in AI and quantum computing, each with a dominant country or small group of countries at its centre, and a much larger periphery of countries whose access to advanced technology, and whose ability to sell their own natural resources, depends increasingly on their alignment with one bloc or another. This is, in some respects, a more technologically charged version of patterns of great power influence that have existed for centuries, but it is happening on a timeline measured in years rather than generations, and with higher stakes, given how central computing has become to both economic growth and military capability.
8. What to watch over the coming months
- Whether the Remote Access Security Act, or similar measures, are enacted into law in the United States, and how cloud providers and other quantum capable countries respond, particularly whether the United Kingdom, the European Union, Japan or South Korea introduce comparable controls of their own.
- Whether the United States extends quantum related export controls to allied nationals in the way it has just done for Anthropic’s AI models, which would confirm that the “no exceptions for allies” approach to AI access is becoming the default approach to quantum access too.
- Whether the discussion of US government equity stakes in AI companies progresses from exploration to action, and whether any similar proposal emerges for quantum computing firms, several of which, such as IonQ, Rigetti and PsiQuantum, are already publicly listed or close to it.
- Further resource for security deals beyond the Democratic Republic of Congo and Venezuela, particularly involving countries with significant reserves of the critical minerals and rare earth elements used in advanced computing hardware.
- How China responds to further export controls, including whether its accelerated investment in domestic quantum and AI supply chains continues to outpace the intended effect of the controls, and whether this hardens a longer term split between US aligned and China aligned technology blocs.
- Whether countries like South Africa are forced into explicit technology bloc alignment choices, and how the tension between government ideology and corporate economic interest resolves, given that the answer may provide an early signal of how the rest of Africa navigates the same question.
- Whether the £500 million UK Sovereign AI Fund and the £2.5 billion UK National Quantum Strategy, together with the European Union’s Technological Sovereignty Package, begin to show measurable results, or whether the gap with the United States and China continues to widen despite this spending.
9. In closing
AI has, within a short period of time, demonstrated how quickly a transformative technology can concentrate wealth and become an instrument of state power.
Quantum computing is following the same path, but starting from an even narrower base of countries and companies, arriving on a timeline that AI itself is helping to compress, and already subject to export controls years before the technology is commercially useful, controls that other quantum capable countries are likely to follow given the stakes for cyber security.
Meanwhile, the way this divide is being enacted in the wider world is already visible in the Democratic Republic of Congo and in Venezuela, where access to security and political legitimacy has been traded for access to natural resources, often with leaders whose continuation in power depends on the deal itself. We see no reason why quantum computing, given its importance for cyber security and code breaking, would be excluded from this pattern, and every reason to think it will be drawn into it even more directly than AI has been.
As Section 5 of this paper sets out, the consequences of bloc formation are already arriving for countries in the Global South, most visibly in South Africa, where the technology alignment question is sharpening an existing tension between a government with one set of political loyalties and a corporate economy with another. Many countries across Africa and beyond face versions of the same dilemma, and the speed at which AI and quantum computing are developing means that the window for making a coherent choice is narrower than it might appear.
Set against all of this, the “golden age” theory requires a degree of voluntary sharing of power, wealth and technological advantage for which history offers very little precedent, and for which we can identify no obvious strategic incentive on the part of those currently holding the advantage. Absent such a precedent, our central expectation remains that within a decade, and quite possibly considerably sooner, a small number of countries, companies and individuals will control the most powerful computing technologies ever created. The rest of the world’s access to those technologies, and to the security and economic benefits that come with them, will depend increasingly on what they are able to offer in return, if anything.
The risks seem clear, and the need for concerted and focused attention and intervention by society more broadly ever more vital.