What’s at stake in AI’s trillion-dollar gamble

A small group of tech giants, called hyperscalers (companies like Alphabet, Microsoft, Amazon, Meta, and Oracle), are investing massive amounts to build AI data centers. These facilities house powerful computers that run artificial intelligence models. In 2026 alone, these companies plan to spend about $750 billion on these centers, with total investments potentially reaching over $5 trillion by 2030. The spending is driven by the belief that AI will transform industries and economies. However, these investments are unprecedented in scale, comparable to the largest capital expenditures in history. The challenge is that these companies currently earn only $150 billion to $200 billion annually from AI, meaning their spending far exceeds their current revenue.

For the hyperscalers to justify their investments, they must significantly increase their earnings by 2030. According to finance professor Jessica Wachter from the University of Pennsylvania, these companies need to boost their productivity by a factor of 2.7 to break even, accounting for costs like capital expenses, interest payments, and depreciation of assets. This level of growth is ambitious, as it would require economic expansion similar to the US IT boom of the mid-1990s, but compressed into just a few years. If they fail to meet these profit goals, they risk falling behind on loan payments, which could lead to bankruptcy. Wachter warns that if productivity gains do not materialize, the current AI infrastructure buildout could become the largest misallocation of capital in history.

To fund their AI data center projects, hyperscalers are increasingly borrowing money, which adds financial risk. In 2026, these companies are expected to have negative free cash flow, meaning their operating cash flow will not cover their capital expenditures. For example, Alphabet reported a free cash flow deficit of $5.9 billion in its latest quarter, despite generating nearly $120 billion in revenue. Borrowing is expensive, and some investors are becoming concerned about the sustainability of this spending. If demand for AI services declines or fails to grow as expected, the companies will still be responsible for repaying their loans. This debt is also spreading through the economy via complex financial arrangements, exposing lenders, insurers, and pension funds to hidden risks.

The core of AI data centers relies on powerful chips called GPUs (Graphics Processing Units), which handle the heavy computational work for AI models. These chips are expensive, accounting for about 60% of the total cost of a data center. Their performance roughly doubles every two years, which drives rapid advancements in AI capabilities. However, this rapid progress creates a challenge: data centers built today may become obsolete within a few years. Owners will need to invest billions more to upgrade to newer GPU generations by the end of the decade to stay competitive. Without these upgrades, the data centers risk becoming stranded assets—expensive facilities that are no longer useful, often referred to as hulks in the article.

For the AI industry’s investments to pay off, AI must drive widespread productivity growth across the economy. Productivity growth means businesses produce more output with the same or fewer resources, leading to economic expansion. Currently, there is little evidence that AI has boosted productivity on a large scale. A survey of 6,000 senior business executives found that 90% reported no productivity increase from AI over the past three years. However, executives expect a modest productivity boost of 1.45% over the next three years, with US executives anticipating a 2.25% increase. If AI fails to deliver these gains, businesses may stop investing in AI, limiting revenue growth for hyperscalers and stalling the economic benefits of AI.

One way businesses expect to achieve productivity gains from AI is by reducing their workforce. Executives surveyed plan to increase sales and cut jobs simultaneously. This approach could lead to public backlash, as communities and workers may feel left behind by the economic transformation driven by AI. The article highlights that public sentiment toward AI could worsen if job losses are significant, potentially leading to resistance against further AI investments. This adds another layer of risk to the AI industry’s gamble, as local communities and policymakers may oppose projects that threaten employment or environmental concerns, such as the construction of new data centers.

The financing of AI data centers has become increasingly complex, involving multiple layers of financial arrangements. For example, Meta’s Hyperion data center in Louisiana, initially projected to cost $10 billion, has expanded to a $50 billion project. To fund it, Meta partnered with Blue Owl Capital, a private-credit firm, forming a joint venture called Beignet. This arrangement involves intricate lease agreements and guarantees, where Meta provides a residual value guarantee to cover the facility’s value if leases are terminated. The financing structure includes multiple subsidiaries and LLCs, making it difficult for outsiders to fully understand the risks involved. Such complexity spreads financial exposure across the economy, including pension funds and insurance policies.