The information that OpenAI proposed to the American government the acquisition of a stake of approximately 5% in the company opened a discussion that goes far beyond the limits of yet another typical business agreement.
For the first time, one of the leading artificial intelligence companies is reportedly considering a model of direct state participation in a technology that is evolving at paces without historical precedent since the Industrial Revolution.
The proposal highlights a deeper question: has the development of artificial intelligence entered a phase where even the strongest companies struggle to finance on their own the massive investments required?
And if so, will governments gradually be called upon to assume a role not only as a regulator, but also as a strategic investor, in a bailout program similar to those for banks during the 2008 to 2009 period?
It is noted that China has formed a hybrid technological ecosystem of private and state-owned companies developing competitive language models, with the cost being covered by the colossal trade surplus of 1.2 trillion dollars that it achieved in 2025.
The explosion of costs
The creation of modern large language models (LLMs) no longer resembles software development as we knew it a decade ago.
Artificial intelligence has transformed into one of the most capital-intensive sectors of the global economy.
Training an advanced model requires hundreds of thousands of specialized processors (GPU), ultra-modern data centers, massive amounts of electricity, specialized fiber optic networks, and highly specialized teams of engineers.
The cost does not stop at training.
Every query submitted to a large model requires computational power, a fact that drastically increases operational expenses.
In contrast to traditional software platforms, where serving millions of users has a relatively small variable cost, AI models continuously consume greater computational resources.
The numbers behind the revolution
Over the last two years, investments have acquired unprecedented dimensions.
Microsoft has committed tens of billions of dollars to OpenAI and the development of Azure AI infrastructure.
Meta invests tens of billions annually in artificial intelligence infrastructure, aiming to develop open models such as Llama.
Google is constantly expanding its data centers to support Gemini.
Anthropic has secured massive capital from Amazon and Google.
Elon Musk's xAI is also investing in large-scale infrastructure while concurrently seeking additional financing to expand its computational capabilities.
The common denominator is one: all companies need ever more capital.
The numbers changing the economy of AI
The largest companies in the sector have entered a capital expenditure race that resembles the construction of national energy or railway networks more than software development, on a scale reminiscent of the Industrial Revolution of the 19th century.
Microsoft announced that its investments in AI infrastructure are expected to reach approximately 80 billion dollars in a fiscal year, primarily for new data centers and artificial intelligence equipment.
Meta increased its capital expenditure budget for 2025 to 64 to 72 billion dollars, with a large part directed toward the development of supercomputers and AI facilities.
Amazon, through AWS, also invests tens of billions of dollars annually in cloud and AI infrastructure, while it has invested approximately 8 billion dollars in Anthropic.
Alphabet (Google) has announced investments exceeding 75 billion dollars for data centers, TPU processors, and the development of Gemini models.
OpenAI, although not publicly traded, has a valuation approaching 300 billion dollars while continuing to seek new capital for the expansion of its computational infrastructure.
Anthropic is valued at over 60 billion dollars, while Elon Musk's xAI has also reached a valuation of tens of billions of dollars within a minimal amount of time.
In total, the investments announced by the large technology groups for artificial intelligence infrastructure already exceed 300 billion dollars annually, without calculating private investments in startups, state subsidies, and the energy infrastructure required to power the new data centers.
The paradox of valuations
The stock market image of the companies is equally impressive.
1) Nvidia: market capitalization over 4 trillion dollars.
2) Microsoft: approximately 3.8 to 4 trillion dollars.
3) Apple: approximately 3 trillion dollars.
4) Alphabet: over 2 trillion dollars.
5) Amazon: over 2 trillion dollars.
6) Meta: approximately 2 trillion dollars.
The six largest technology companies now represent a market capitalization exceeding 16 to 18 trillion dollars, a percentage unprecedented in the history of American markets.
Despite the massive valuations, liquidity needs remain enormous.
Most companies raise capital through:
1) issuance of new shares,
2) convertible bonds,
3) corporate bonds,
4) bank credit lines,
5) strategic investors (Microsoft, Amazon, SoftBank, etc.).
Thus, the issue is not that AI companies already have excessive debt, but that capital needs are increasing faster than their cash flows.
If this pace continues, markets will have to absorb hundreds of billions of dollars of new investments every year.
The critical shortage of infrastructure and the shift in the economic model
The biggest problem is no longer just capital.
Companies face constraints in accessing advanced processors, power from energy grids, and suitable facilities.
Modern artificial intelligence units require power equivalent to small cities. In several regions of the United States, investments have already begun in new substations, electricity transmission networks, and even small modular nuclear reactors (SMRs), aiming to power the new data centers.
Thus, artificial intelligence ceases to be solely a technological product and transforms into a matter of national infrastructure and national security.
Why the American state is interested
The potential state participation in OpenAI cannot be examined independently from the geopolitical competition with China.
Washington now treats AI in a manner corresponding to how it treated nuclear technology, the internet, or semiconductors in the past.
Artificial intelligence is considered critical both for economic growth and for defense, cybersecurity, intelligence gathering, the development of autonomous weapon systems, and scientific research.
A potential participation of the American state in cutting-edge companies could ensure greater influence over their strategic decisions without requiring direct nationalization.
This is not a bailout in the typical form, but what could it mean?
So far, there are no indications that OpenAI or other large AI companies are on the verge of financial collapse, but the existence of a massive bubble in valuations that is not supported by fundamental magnitudes, as well as the borrowing of enterprises from private entities (private credit), is universally acknowledged.
Conversely, they continue to attract significant private capital and to be valued at particularly high levels.
However, their business model is characterized by a paradox: the more the use of their services increases, the more the need for new investments in computational power and energy infrastructure increases.
This creates a perpetual cycle of financing needs.
A new model of industrial policy?
The history of the United States shows that the state has intervened repeatedly in strategic sectors.
The creation of the internet was based on public funding through DARPA.
The development of the American aerospace industry relied on government contracts.
The CHIPS Act mobilized tens of billions of dollars to strengthen domestic semiconductor production.
During the banking crisis of 2008, the TARP program prevented the collapse of the financial system through state intervention.
These cases differ significantly from one another, yet they demonstrate that when a sector is considered critical for national security or economic stability, state involvement is not an exception.
Increased state participation is not without risks.
On one hand, it can offer stability and access to long-term capital.
On the other hand, it raises questions of competition, concentration of power, and state influence on the development of technology.
Furthermore, the close interconnection of AI companies with government agencies may reinforce concerns regarding the use of artificial intelligence for military or surveillance purposes.
The big question
The proposal for participation of the American state signals that the discussion has shifted: from how AI will be developed, to who will finance the constantly increasing infrastructure it requires.
If the cost continues to increase at current rates, private investments may not suffice to support the next phase of the technological race on their own.
In such an environment, it is not impossible that we will see more forms of cooperation between the state and the private sector, either through investments, through tax incentives, or through shared infrastructure.
What is certain is that artificial intelligence ceases to be exclusively an affair of Silicon Valley.
It is gradually transforming into a field of national strategy, industrial policy, and geopolitical competition. The discussion about OpenAI is perhaps the first sign of this new era.
Bank bailout plans and the future cost
The historical experience of the financial crisis of 2008 shows that the total losses of a sector differ significantly from the final fiscal cost of state intervention.
According to the International Monetary Fund (IMF), total write-downs of financial assets of American origin amounted to approximately 2.7 trillion dollars, of which approximately 1.6 trillion burdened the banks.
Despite the magnitude of these losses, the net fiscal cost of the TARP program was ultimately limited to approximately 31 billion dollars, as the largest part of the capital channeled into banks was returned to the American public.
A corresponding hypothetical framework can be applied to the artificial intelligence sector.
Today, the largest investments are made by Microsoft, Amazon, Alphabet, and Meta.
The total capital expenditures (CAPEX) of these four companies increased from approximately 224 billion dollars in 2024 to approximately 413 billion dollars in 2025, while analyst estimates converge that they will fluctuate between 600 and 700 billion dollars in 2026, with the largest part concerning investments in AI infrastructure.
Most analyses estimate that during the period 2025 to 2030 the investments of hyperscalers in AI will exceed 5 trillion dollars.
Based on the above, a hypothetical scenario of an AI market collapse can be formed.
If, after a period of overinvestment, demand proved significantly lower than expectations, large write-downs on data centers, specialized processors (GPU), power purchase agreements, and software would be probable.
Assuming that 30 to 40% of the already realized investments in AI infrastructure would be obsolete, total accounting losses could approach 1.0 to 1.5 trillion dollars within a horizon of a few years.
This magnitude is comparable to the losses suffered by American banks during the crisis of 2008, although it would primarily concern fixed assets and not financial products.
If an approach analogous to TARP were applied, where the state would acquire shares or convertible bonds instead of directly subsidizing losses, then the required financing could fluctuate between 200 and 400 billion dollars. From this amount, a significant part could be recovered in the future through sales of stakes, dividends, or repayments, resulting in the final net fiscal burden being noticeably smaller than the initial disbursements.
The key difference compared to the crisis of 2008 is that systemic risk does not originate today from bank balance sheets but from the concentration of massive investments in digital infrastructure and computational power.
Therefore, even if a bubble in artificial intelligence led to losses comparable to those of the financial crisis, state intervention would likely be smaller as a percentage of total losses but would constitute a terrifying blow to the US capital market.
Millions of micro-investitors will lose their savings and their jobs, while large pension funds will also be destroyed in an economic turmoil with multiple times the consequences of the financial crisis of 2008 to 2009.
It is no coincidence that the tech bros, the leaders of the largest companies, wished to join the closed system of power of Donald Trump after his return to the White House in 2025, nor is the, no longer strange, manifesto of Palantir regarding the connection of national security with the development of AI.
www.bankingnews.gr
Σχόλια αναγνωστών