Big Tech’s $725-billion AI binge is starting to pay off, report says
A report from Exponential View indicates that major tech companies' significant investments in AI infrastructure, including data centers, are beginning to pay off, with global AI sales exceeding depreciation costs for the second consecutive quarter. While the economics are holding, margins remain thin, with depreciation consuming over two-thirds of revenue. The report addresses the sustainability of hundreds of billions of dollars poured into chips and data centers to meet customer demand for AI.
A new report from research firm Exponential View reveals that the substantial investments by major tech companies in artificial intelligence infrastructure, including data centers, are beginning to yield returns, with global AI sales (excluding China) hitting $25 billion in Q1 2026 for hyperscalers and neoclouds. This figure surpassed the estimated $21 billion in depreciation costs tied to data center and chip investments for the second consecutive quarter, signaling potential economic sustainability for the AI boom. Azeem Azhar, founder of Exponential View, commented that "the economics are holding" despite razor-thin margins, as depreciation charges still consume over two-thirds of revenue, leaving a small buffer for other operational costs like power and labor.
The findings address a key concern about whether customer demand can justify the hundreds of billions of dollars poured into chips and data centers, with US tech giants like Meta, Google, Microsoft, and Amazon planning up to $725 billion in capital expenditures this year for AI infrastructure. The report also indicates that older chip models, such as Nvidia's H100, are retaining significant value and demand, with Amazon Web Services CEO Matt Garman confirming continued use of six-year-old Nvidia A100 servers. Furthermore, the analysis suggests a shift in user preference, with more developers opting for open-weight and Chinese AI models like DeepSeek, leading to a decrease in the share of tokens requested from Google, OpenAI, and Anthropic models on platforms like OpenRouter. This trend, according to Azhar, pushes leading foundation-model companies to offer additional services and lock-in to maintain premium pricing.