AI Bubble and Military Bottleneck: A Systemic Crisis

The financial bets on the revolutionary promises of generative AI have soared to dizzying heights. Circular funding among industry giants is proliferating, while structural limitations are emerging regarding the reliability and economic value of large language models (LLMs). From one bubble to another, this new frenzy reflects the deeper disorganisation affecting Western economies in the deployment of capital and skills. In this respect, the simultaneous weakness in industrial capacity  among Ukraine’s backers reflects a systemic crisis. An opinion piece by Rémi Bourgeot, economist and engineer, Associate Research Fellow at IRIS.

While the world was waking up to the concrete potential of artificial intelligence with ChatGPT, the collapse of the Silicon Valley Bank in early 2023 triggered the onset of a financial crisis. Technology stocks were hit hard. Venture capital funds were blamed for their risky financial schemes, particularly in the cryptocurrency space, which was hit by a series of scandals.

These reservations were soon swept aside by a new wave of financial euphoria, this time centred on AI, but following similar patterns. Nvidia emerged as the big winner, with its graphics cards tailored to the requirements of giant neural network calculations. It effectively locked up the market with its proprietary platform, Cuda. The very notion of valuation ratios was overshadowed by the prospect of a radical transformation of human activity.

It comes as no surprise that the intrinsic limitations of LLMs were overlooked during the initial phase of euphoria. Beneath the sweeping reactions of both AI apologists and staunch detractors, a more nuanced perspective emerged from discreet commentators, combining a technical grasp of neural networks with a philological intuition about the strengths and the limits of the syntactic logic captured by LLMs.

OpenAI began by developing open, non-profit models, and its status remained hybrid for years. The prevailing idea was that LLMs would reach a qualitative tipping point, thanks to an explosion in size and compute resources. The confusing notion of AGI (artificial general intelligence) then served as a horizon for the most extravagant funding schemes.

However, by 2024, the technical achievements of companies like Mistral in France and DeepSeek in China, with incomparably more limited resources, began to cast doubt on the idea that model deployment required the trillions of dollars mentioned by Sam Altman at OpenAI.

The companies developing core AI models do not currently exhibit a real business model, beyond using investor funds to cover their expenses, particularly for the purchase of chips. On top of the issue of financial stability, the allocation of such resources to a particular technology must also be questioned. AI Pioneer Yann Le Cun has repeatedly emphasised the limitations of LLMs and called for efforts to be made on other types of models, which have been ignored by the bulk of investors. Instead, the bubble took on a new dimension, with massive funding from semiconductor companies like Nvidia to their own customers, like OpenAI.

This latest bubble raises questions not only about this very industry, but more generally about the way the economy is funded. It seems increasingly difficult for developed countries to sustain industrial momentum beyond waves of financial and institutional frenzy that suggest magical thinking, or sometimes even mass hysteria.

Meanwhile, the war in Ukraine highlights the limitations facing the Western industry in producing equipment. Production capacities for ammunition, armoured vehicles and electronic components have proven chronically inadequate to meet sustained and prolonged demand. Many factories capable of manufacturing critical components have been closed in recent decades. Supply chains are limited, often dependent on rare or offshore suppliers.

This situation reveals a systemic failure centred on insufficient production, which goes beyond the defence industry. It results from a lack of strategic planning, particularly in terms of financing, energy supply and skills deployment. Reviving production requires restoring complex industrial chains and long-term profitability models. Otherwise, even massive investments will have no effect.

Industrial strength does not come from stock market bubbles fuelled by the ecstasy of a post-physical digital nirvana. It requires careful interaction between businesses, research institutions and government agencies, based on long-term strategies and human skills. Behind the cutting-edge intellectual resources poured into LLMs, the bubble lays bare the erosion of industrial development strategies, exacerbated by failing educational systems and the relegation of scientific skills.

Nevertheless, in light of the manufacturing rout epitomised by Boeing, the US policy focused on redeploying manufacturing and controlling energy costs is showing tentative signs of improvement. This is the case even in semiconductors, with TSMC establishing operations. Although financial shocks hamper in-depth reindustrialisation, the country is ultimately managing to assert its dominance in the digital field.

The European Union, meanwhile, finds itself in a more precarious situation due to its technological retreat and the energy chaos stemming from Germany’s phase-out of nuclear power. By positioning itself as a faithful user of US technologies, it is undermining its industrial potential. In the dot com bubble of the late 1990s, Europe typically lagged behind during the upswing, endured the full brunt of the market crash, and ultimately failed to catch up on the technical front. In this respect, Ursula von der Leyen’s determination to cement the EU’s role as a digital and military vassal of the US for decades to come foreshadows a decline in European living standards and political dislocation.