The last few weeks have seen a dense succession of investments and strategic agreements between the main players artificial intelligenceAt the beginning of September OpenAI signed with Oracle infrastructures an agreement estimated at approximately 300 billion dollarsto purchase computing capacity over the next five years, sending the company's stock skyrocketing. A few days later, Nvidia announced an investment of up to $100 billion in OpenAI to build 10 gigawatt AI data centers, followed by expanded collaboration between OpenAI and CoreWeave for another 6,5 billion. In parallel, ecosystem has signed contracts worth 33 billion dollars with Nebius and CoreWeave, while Meta announced a multi-year agreement worth 14 billion with the latter. Confirming the climate of rush for investments, OpenAI has also signed an agreement with the manufacturer of AMD chips worth tens of billions of dollars that will allow it to reduce its current dependence on Nvidia. Finally, Google: has established a partnership with Oracle to bring AI to Gemini: on the group's cloud platform and will simultaneously invest £5 billion in the UK in research and data centres.
Overall, a chain of investments exceeding 500 billion of dollars, a sign of the immense effort by Big Tech to control the artificial intelligence infrastructure and ensure a long-term competitive advantage.
Cautionary signals: Low margins put analysts on guard
But as the euphoria of the insiders grew, an analysis published from The Information has dampened enthusiasm. According to the report, the profitability margins of the mega-contract between Oracle and OpenAI would actually be very slim, as a large portion of the expected revenues would be absorbed by energy, hardware, and maintenance costs. The news has fueled the skepticism of those analysts who see expectations prevailing in the current acceleration at the expense of real profits.
So far, he has observed The InformationAI has proven profitable especially for those who provide the physical infrastructure: chip and server makers like Nvidia, AMD, and Dell. “For many others, it's a bottomless pit. We already knew this was true for OpenAI, but a new example is Oracle, a software company whose rapidly growing cloud business is threatening the company's traditionally high profit margins.”
Secondo Martin Peers, the newspaper's respected deputy editor, in fact, there are still few signs that AI is a profitable business for those who use servers to sell apps or develop AI models. Programming assistants, for example, have shown that not have high margins, as other AI apps currently in circulation or still in development likely don't. Peers' comment is clear: "For years to come, we may look back on this period in the tech industry as one in which the entire industry was gripped by a collective delusion."
Circular agreements and the risk of an AI bubble
There is also a growing feeling that the entire sector is pushing its economic sustainability to the limit. Wall Street Journal feels the need for capital to build the AI infrastructure is already putting pressure on the funding sources traditional, resurfacing fears from tech bubble. For this reason, Nvidia's commitment to invest up to one hundred billion dollars in OpenAI is perplexing. These types of cross-trades or "circular agreements"—Nvidia sells chips to OpenAI, then invests in OpenAI, which uses those funds to buy more chips from Nvidia—are not popular in finance because they evoke sinister precedents. Before the dot-com bubble (2000) and the financial crisis (2008), in fact, there were several operations of this type, which apparently seemed like normal business relationships but concealed systemic fragilities.
“If a year from now we were at the point where there was an AI bubble that is erupted“, concludes the WSJ, "this agreement could have been one of the first warning signs". The fear is not so much that of an isolated speculative bubble, but of a systemic risk. If AI as a whole were to go into crisis and sink into one of those periods of stagnation that cyclically affect it, the entire ecosystem could be dragged down, with significant macroeconomic consequences.
Automation that doesn't innovate: economists' concern
While excitement, financial risk, and investment pressures are racing past analysts' monitors, the effects of the AI race have already begun to ripple downstream. According to Nobel Prize-winning economist Daron Acemoglu, the more money flows into AI, the more a growing number of companies think they need to follow suit.
Executives, pressured by boards of directors and shareholders, feel compelled to act quickly without yet truly knowing how to use AI effectively. The result isn't a profound redesign of processes, but the addition of chatbot while real problems remain unsolved and customers are dissatisfied. A trend that worries Acemoglu especially now that the AI's innovative drive is slowing down, while AI agents still exhibit learning and reasoning limitations that may take years to overcome.
The risk could be, concludes Acemoglu, to settle for mediocre automation rather than investing in real innovation"So-so automation, like self-checkout kiosks or automated menus, has cut jobs without providing any real productivity gains or real cost savings for the companies that have adopted them." This prospect would also not help the economy as a whole. Still sluggish, global growth depends today more than ever on quality technical progress capable of generating new markets and jobs.
When concentration blocks innovation
What is at stake is therefore structural and leads us to ask what the development model that is taking shape behind the AI race. And whether an ecosystem dominated by a few global players can foster the development of AI and progress in general, or, conversely, paradoxically risks becoming its greatest obstacle. Observing the history of technological cycles, a crucial factor in progress emerges: the alternation of phases of openness and stimulation of innovation with others of rationalization and consolidation.
The periods of great technological creativity They arise not from the concentration of resources, but from the proliferation of parallel experiments, from the opportunity for many to try different approaches, from the calculated risk of failure. Without this adaptability—on the part of businesses, industrial sectors, and institutions—innovation tends to stall. Carl Benedict Frey, an Oxford economist who deals with digital transition and economic development, in his recently published work: “How Progress Ends”, foresees the risk that this could happen right now in the United States. While the first discoveries – from transformers to the same Generative AI – were born from open experimentation in universities and small laboratories, today everything is in the hands of Big Tech.
Le startup They are acquired for the talent that works there, not for the projects they develop. Few venture to experiment in areas dominated by the big players. Those who control the platforms largely decide where innovation is directed. The institutional context, which at the beginning of the digital revolution had favored the plurality of research by financing it through public agencies and protecting it with antitrust policies, has now changed radically. Funding for basic research has shrunk, while acquisitions that would once have been blocked now proceed unhindered.
Herein lies the great risk for AI and innovation in general: when market structures impede the transition from cycles of consolidation to new waves of decentralized discovery, the economy loses its ability to generate the factors that should boost its growth. We find ourselves stuck in an "exploitation" phase—optimization of the known—without the possibility of returning to the "exploration" phase, from which true innovations arise.
The European paradox: regulating AI without driving innovation
In this scenario theEurope is in a singular position. While Silicon Valley Having concentrated power and capital, the old continent has chosen the path of regulation. First the GDPR, now the AI Act: Brussels seems to have given up competing in the AI race, limiting itself to setting limits and constraints. Yet, it could play a decisive role, if only it shifted its paradigm.
Its more decentralized and democratic structures than those of the United States, the tradition of independent university research and the fabric of innovative small and medium-sized enterprises could become a competitive edge precisely at the stage in which innovation needs to return to widespread experimentation rather than monopolistic concentration.
To do so, Europe should abandon its merely conservative posture. defensive – one that limits itself to regulating what others invent – and rediscover the ambition to be a protagonist. Not by concentrating resources in a few "national champions" on the American model, but by valorizing precisely what makes it different: the institutional polycentrism, university autonomy, an entrepreneurial culture less aggressive and more flexibleHistory has taught us that progress belongs not to those who accumulate the most power and capital, but to those who know how to maintain a vibrant ecosystem where innovation can continually regenerate itself. And progress, as Carl Benedikt Frey reminds us, is never inevitable: it can end.
