The days when AI deployment was financed with private capital are over.
To achieve the announced growth rates, the bond and equity markets will be solicited in proportions never seen before.
The excitement around the potential of artificial intelligence (AI) is not slowing down. The acceleration in investment is exponential and, by itself, explains a large part of current American growth. Over the period 2026-2031, Goldman Sachs calculates that $7.6 trillion could be invested to build the AI ecosystem. The telecom bubble — described at the time as one of the largest capital investments ever observed in such a short time — had mobilized a cumulative $500 billion between 1996 and 2000. AI would therefore be 15 times more important in absolute terms.
Just five American companies, commonly called hyperscalers, would spend $6 trillion alone over this period and will represent 80% of these investments. These are Amazon, Alphabet, Microsoft, Meta and Oracle. In 2027 alone, hyperscalers are expected to spend $1.1 trillion, 6-7 times more than 5 years ago, representing 3% of US GDP.
Until now, these companies have financed their capital expenditures with their annual cash surpluses, but their financial equation is rapidly deteriorating with important implications for the dynamics of the bond and equity markets.
In 2025, hyperscalers had an annual surplus (EBITDA – capex) of $200 billion, but this figure is melting quickly and will be close to 0 in 2026. If they wish to maintain the remuneration of their shareholders, these companies will therefore have to issue much more debt than before. Goldman Sachs calculates that they already represent 18% of net new issues on the investment grade market in the United States since the start of 2026. Such concentration destabilizes historical balances, because the weight of the technology sector in the investment grade market is currently only 4%. The concentration on a few companies linked to AI was already a source of fragility for major American equity indices like the S&P 500; it could also become so for the bond market.
One solution to limit the amount of new issues from hyperscalers is to reduce or cancel share buybacks, i.e. shareholder remuneration. This is already partly the case. These buybacks increased from 130 billion dollars in 2022 to 90 billion in 2025. At the same time, share issues for employee compensation have increased, halving the net effect compared to 2022. The natural support provided by these share buybacks is therefore disappearing.
This phenomenon is concomitant with an explosion in the supply of shares from the IPO market. IPOs from SpaceX, OpenAI and Anthropic are announced for 2026-2027 and could represent around $200 billion in new shares, four to five times the average annual size of the US IPO market across all sectors. Cash flows from hyperscalers and private capital are no longer enough: capturing savings from listed markets becomes essential, via more bonds and more shares in circulation.
The United States uses the supremacy of its financial markets to attract foreign investors and thus cheaply finance its AI infrastructure and strengthen its domination of the global economy. They are even relaxing the index rules so that passive management can quickly participate in these IPOs. This will guarantee them a considerable windfall of additional financing because the weight of the United States in global indices such as the MSCI World is close to 75%. Thus, asset managers who do not free themselves from benchmarks do not have the right to be wrong about the future profitability of AI because their capital allocation is de facto concentrated disproportionately on a sector and a country.
