Where're All The AI Chips?

An extensive financial investigation argues that Microsoft and other hyperscalers have been misrepresenting AI data center capacity, using vague terms like 'gigawatts of capacity' that actually refer to total power or non-AI infrastructure rather than deployed, powered GPU capacity. Drawing on earnings calls, Bloomberg and Guardian reporting, and capex/depreciation analysis, the piece estimates Microsoft has spent $265 billion in capex since 2022 but only has around $50 billion of GPUs actually installed (roughly 2GW of AI-specific capacity out of a claimed 12GW total). Extending this analysis across Google, Amazon, Meta, Oracle, CoreWeave and others, it estimates roughly $200-390 billion worth of NVIDIA and other AI chips are sitting uninstalled in warehouses or unpowered data centers, meaning NVIDIA's revenue growth reflects speculative multi-year advance purchases rather than immediate deployment. The piece concludes that reported 'demand outstripping supply' is largely driven by OpenAI and Anthropic's contracts rather than broad market demand, and calls the situation evidence of a looming AI infrastructure overbuild.

Source: https://www.wheresyoured.at/wherere-all-the-ai-chips. 8 Sync News only summarizes and links out; content copyright belongs to the authors and original sources.