AI stock valuations: examining the risk of an earnings-driven correction

Historically, market cycles are often compared to past events, most notably the dot-com boom. While many investors point to current earnings as justification for high valuations, concerns remain about the long-term durability of those profits. The possibility of an “earnings bubble” – where prices and earnings both decline – is a significant risk.

During the dot-com era, stock prices surged ahead of actual profits, leading to inflated price-to-earnings (P/E) ratios. When earnings failed to meet expectations, prices plummeted. Current P/E ratios are not as extreme as those seen in the early 2000s, as earnings have also increased substantially.

However, three key factors raise concerns about the sustainability of AI-related earnings: the financial performance of hyperscalers, questions surrounding accounting practices, and the prevalence of circular financing.

Leading cloud computing companies, known as hyperscalers, have traditionally been strong generators of free cash flow. Recently, these companies have begun operating with negative free cash flow, meaning they are spending more than they are earning. This spending is largely directed towards semiconductor companies that produce the chips powering data centers, leading to increased earnings for those semiconductor firms. A slowdown in hyperscaler investment could trigger a significant drop in semiconductor earnings and stock prices.

Furthermore, some critics allege that hyperscalers are artificially inflating earnings by extending the depreciated lifespan of their servers and networking hardware. Nvidia has refuted these claims, asserting that customers depreciate GPUs over four to six years based on their longevity and usage.

Another concern is the practice of circular financing, where companies use financial leverage to invest in businesses that purchase their products. For example, Nvidia’s investment in cloud provider Nebius allows Nebius to immediately purchase Nvidia GPUs, which is then recorded as revenue for Nvidia. This creates a closed-loop system that can distort genuine market growth and make it difficult to assess the sustainability of earnings.

While AI holds transformative potential, past technological advancements, such as railroads and the internet, have also experienced speculative bubbles. To avoid a similar fate, hyperscalers must generate revenue that justifies their capital expenditures. Current projections estimate the AI revenue-to-capex ratio will be around 4% in mid-2026, highlighting the considerable gap that needs to be closed.