NORBERT BIEDRZYCKI

The AI Capital Concentration: Is the Market Creating a New Generation of Technology Giants?

The AI Capital Concentration: Is the Market Creating a New Generation of Technology Giants?

More than half of all venture capital in the world now goes to a single technology. Within that technology, an ever smaller number of companies takes most of the money. That is not a market. It is a gravitational field.

According to OECD data published in February, AI companies attracted approximately $258.7 billion of venture capital in 2025, around 61 percent of all VC investment globally. Five years ago, when I first started writing about AI as an investment category, the comparable figure was around a fifth.

But the headline number hides the more important story. The capital is not spread across a broad ecosystem of startups. It is concentrated in a remarkably small group of companies. OpenAI raised $40 billion in a single round led by SoftBank in 2025. Anthropic raised $13 billion in September. xAI, Mistral and a handful of others raised multibillion rounds. A meaningful share of the global venture total was absorbed by fewer than ten companies.

Circular capital

What concerns me even more is the structure of some of these deals. In the second half of 2025, a series of agreements linked the biggest players in a closed loop. Nvidia announced plans to invest up to $100 billion in OpenAI, which in turn will buy Nvidia hardware. OpenAI signed a reported $300 billion cloud contract with Oracle, which buys Nvidia chips to fulfil it. AMD offered OpenAI warrants on its own shares in exchange for a large purchase commitment.

Each of these deals can be justified individually. Taken together, they create a system in which suppliers finance their customers, customers commit to spending they have not yet earned, and valuations are supported by commitments rather than cash flows. I have spent enough years in boardrooms to know that when the same dollar appears on several balance sheets, risk tends to be underestimated.

Why concentration happens

There is a rational logic behind the concentration. At the frontier, AI has extreme economies of scale. Training the most capable models requires tens of billions of dollars of compute. The best researchers want to work where the biggest clusters are. Customers prefer providers that will still exist in five years. Distribution through existing platforms favours those who already have hundreds of millions of users.

Public markets show the same pattern. A handful of technology companies now represent around a third of the value of the S&P 500. Their combined capital expenditure, mostly on AI infrastructure, reached hundreds of billions of dollars in 2025, and plans for this year point higher still.

Monopoly or ecosystem?

The fundamental question is whether this concentration will persist. There are forces pushing in both directions.

Toward concentration: the cost of frontier training, the importance of distribution, and the advantages of controlling compute, models and customer relationships in a single company.

Toward dispersion: rapidly improving open models, falling inference costs as demonstrated by DeepSeek last year, regulatory pressure in Europe and the US, and the simple fact that most enterprise value will be created in applying AI to specific problems, not in building general models.

What this means for investors outside the core

For most funds, competing for allocation in the largest AI rounds is neither possible nor wise. The more relevant question is how to position around the giants: which businesses benefit from their competition, which are vulnerable to being absorbed by them, and which have assets the giants cannot replicate.

For institutional investors, there is also a portfolio risk question. If a large share of your public equity, private equity, venture and infrastructure exposure depends on the same AI thesis and the same handful of companies, your diversification may be much lower than it appears on paper.

Investment question: Are we witnessing the formation of a new generation of global technology monopolies?

Read next

  1. AI won’t run out of customers. It will run out of power and time to repay its debt

    Many people read the sell-off in data center stocks as the first crack in the AI bubble. I look at the same numbers and see something else. The demand is there, and it is substantial. The trouble lies elsewhere: in the construction itself, in access

  2. DeepSeek and the Repricing of AI: Does Building Bigger Models Still Make Sense?

    The Business of AI series, written in 2026

  3. The AI Gold Rush: Are Investors Paying for Technology or Future Expectations?

    The Business of AI series, written in 2026