Oracle warns: AI boom could collapse due to infrastructure risks

What happened?

Oracle, one of the tech giants most aggressively investing in artificial intelligence infrastructure, has included in its annual report (Form 10-K) a section detailing the risks that could derail the entire AI boom. According to Gizmodo, the company warns that its massive investments in data centers, which exceed $40 billion planned for this fiscal year, may not generate expected returns if AI demand fails to materialize or if supply, energy, or regulatory issues arise. The document, filed with the SEC, explicitly states that "demand for AI services may not materialize at the expected level" and that "construction and operation costs of data centers could exceed estimates." This is not an isolated warning: in its Q3 2024 earnings call, Oracle's CFO Jeff Epstein had already noted that capital expenditure would double this fiscal year, but returns depended on enterprise AI adoption.

Why is this important?

This warning is not a mere regulatory compliance exercise. Oracle, as a cloud infrastructure provider for companies like NVIDIA and OpenAI, has a privileged view of the ecosystem. Its analysis points out that the supply chain for critical components (GPUs, cooling systems, transformers) is fragile, and that electricity availability for data centers is already straining grids in several regions. According to the annual report, Oracle mentions "disruptions in the global supply chain that could affect the availability of specialized hardware" and "restrictions in energy supply at key locations." Additionally, it notes that real AI demand could be overestimated, leading to an investment bubble. Historical context: recall the dot-com bubble, where overinvestment in infrastructure without real demand led to a collapse. Between 1995 and 2000, telecom companies invested over $500 billion in fiber optics and network equipment, only for much of that capacity to remain unused when the bubble burst. Although AI has tangible applications, the current pace of spending echoes that era: according to IDC, global spending on AI systems will reach $154 billion in 2024, up 27% from the previous year, and is expected to exceed $300 billion by 2027. However, a 2023 Goldman Sachs report warned that only 10% of companies had implemented AI at scale, suggesting a gap between investment and actual adoption.

Market consequences

If the risks materialize, the consequences would be severe: companies like Oracle, Microsoft, Google, and Amazon, which have committed hundreds of billions to data centers, could face massive losses. For example, Microsoft has announced plans to spend over $50 billion on AI infrastructure in the coming years, while Google has allocated $30 billion in 2024 alone. AI startups that rely on cheap computing access would see costs rise or face restrictions; according to a Bernstein analysis, the cost of renting a cluster of NVIDIA H100 GPUs has surged 40% in the past year due to scarcity. Chip manufacturers (NVIDIA, AMD, Intel) would also be affected if demand contracts: NVIDIA, which reported record revenue of $22.1 billion in Q3 2024, could see a slowdown if hyperscalers reduce orders. On the positive side, this warning could lead to more rational investment and a focus on energy efficiency and diversification of energy sources. For instance, modular data centers and nuclear power are already being explored to reduce reliance on the electrical grid.

What should readers know?

• Oracle's warning is not a prediction of imminent collapse, but a wake-up call about real risks that need to be managed. The company has a history of conservative reporting, but in this case the risks are shared across the sector.
• Energy and component shortages could slow AI deployment, not stop it. The IEA estimates that data centers will consume 4% of global electricity by 2030, up from 1% today, requiring massive investments in grids and generation.
• Investors should be cautious: AI infrastructure profitability is not guaranteed. A McKinsey analysis suggests only 60% of AI projects generate a positive return, and many take years to achieve it.
• Companies using AI should diversify suppliers and consider more efficient models, such as edge inference or using older GPUs for less demanding workloads.
• Regulation (e.g., emissions laws, chip export restrictions) could exacerbate risks. The U.S. CHIPS Act and export restrictions on GPUs to China are already affecting the supply chain, and new European AI regulations could increase compliance costs.

"Oracle's warning is a worst-case scenario for the whole AI boom" — Gizmodo

In summary, Oracle's report is a reminder that even tech giants recognize the AI path is not linear and is fraught with obstacles. The key will be how the industry responds to these challenges: whether an orderly correction or a crash occurs will depend on the pace of real adoption, efficiency innovation, and geopolitical stability. For now, Oracle's warning serves as a healthy counterbalance to the excessive optimism that has dominated the sector.