AI is the biggest reallocation of capital in a generation. Here’s why that should concern all of us

AI has led to a fundamental restructuring of economic power, raising questions that may shape the future of democracy itself.

---

This year, global spending on artificial intelligence will reach $2.6 trillion – a 47% increase from a year ago. PwC analysis suggests AI adoption could boost global GDP by 15% over the next nine years. The last time the world saw growth at that scale was during 19th century industrialisation.

These are not abstract figures. They describe a fundamental restructuring of economic power. It is likely this restructuring is already reshaping the environment you work in. And it raises a question that should be at the centre of every conversation about democracy and public trust: who controls this infrastructure, and who is accountable for what it does?

The concentration problem

Today, five American technology companies are collectively worth more than $20 trillion. Nvidia alone is valued at $5.3 trillion – more than the annual GDP of every country on the planet, with the exception of China and the United States.

In June, SpaceX completed the largest IPO in history, raising $75 billion. On its first day of trading, its market cap reached $2 trillion, making Elon Musk the world’s first trillionaire. OpenAI and Anthropic are planning to follow suit, each with initial valuations expected to exceed $1 trillion.

Nearly three-quarters of today’s financial investment is flowing into a handful of companies. According to the Federal Reserve Bank of St. Louis, investment in AI software and data centres accounted for 39% of US GDP over the last four quarters. Governments are grappling with the near-impossible task of keeping pace. Saudi Arabia plans to spend $100 billion on data centres and AI companies. The EU has committed €50 billion, though only a fraction is currently secured.

Put yourself in the shoes of a political leader managing significant national debt, under pressure to fund healthcare, education and pensions. Would you divert vast sums to national AI infrastructure? Relinquish control to private firms? These are nearly impossible choices – especially when the level of private capital flowing into AI is one no government can match.

The governance gap

The result is that AI governance is fragmented, contested and dangerously behind the pace of adoption.

In the US, the priority remains innovation over legislation. Earlier this year, the White House asked Meta, Nvidia, Oracle and Google to join a Technology Advisory Council – placing the people building AI in the same room as those called to regulate it. Weeks later, an Executive Order that would have introduced federal safety reviews before any new AI model could be released was scrapped. The reason? Tech leaders had convinced the President it would slow innovation.

In Europe, the EU AI Act – the world’s first comprehensive AI legislation – came into force this August, imposing transparency obligations and penalties for serious breaches of up to €35 million, or 7% of annual revenue. In China, AI development is strictly controlled by the state, with compulsory labelling of AI-generated content. The government simultaneously plans to embed AI across every aspect of its economy by 2035.

Meanwhile, countries like Brazil attract tens of billions in data centre investment but develop no foundation models of their own. Sub-Saharan Africa lacks sufficient electricity supply, digital infrastructure, and regulatory capacity. The digital and income gap between rich and poorer countries is about to widen even further.

The accountability question

SpaceX won contracts worth a combined $6.5 billion to build the Golden Dome missile defence shield; a further $1.6 billion contract from the Pentagon; and another $1.8 billion to create a network of spy satellites capturing images of activities on the ground nearly anywhere in the world.

Meanwhile, Anthropic found itself in a legal dispute with the Pentagon over a $200 million military contract, insisting it didn’t go far enough to stop the technology being used in mass surveillance or to power autonomous weapon systems. The US government designated Anthropic a national security risk and blacklisted its products. Anthropic is suing. In the meantime, OpenAI struck its own deal with the US Department of Defence.

But the risks go beyond intentional misuse. Earlier this year, Anthropic discovered that Mythos 5.1 had capabilities so advanced they could be used to create cyberattacks, exploit cyber vulnerabilities, and potentially help build bioweapons. The model was released only to a limited number of partners, sending banks, tech giants and governments scrambling.

Weeks later, an OpenAI agent hacked an Australian government website, accessing private Medicare data. And then Jacob Coxon, a British researcher, quit Anthropic citing significant safety concerns. “The people building AI,” he said, “are convinced that AI could kill all of us by the end of the decade.” Four days later, Anthropic CEO Dario Amodei broadly agreed, calling for the industry to slow the pace of development.

The reaction was swift. A public gathering in Washington brought together socialist veteran Bernie Sanders and right-wing populist Steve Bannon, on the same stage. Both were calling for better control of AI. The White House branded AI safety fears a “massive hoax.”

What are the full capabilities of the tools being created? Is your pension financing military-grade technology without your consent?

The questions we need to answer together

The World Economic Forum projects AI will transform 86% of businesses in the next four years, creating 170 million new roles while making 92 million jobs redundant. But the data points to a question that surprises most people: not which professions will be most affected, but where. According to the World Bank, it is the wealthiest countries that will face the most disruption from job automation; more than triple the numbers impacted in low and middle-income countries.

If AI companies bear no legal responsibility for what their models can do – hacking a bank, stalling air traffic control, actions for which people today face criminal charges – what type of oversight would be appropriate? What happens to the concept of criminal negligence?

These are not hypothetical questions. They are the questions that will determine which future we build.

What this means for your sector