Where're All the AI Chips? (www.wheresyoured.at)

Recent revelations indicate that Microsoft may have overstated its AI data center capabilities, with only about 2 gigawatts of AI-specific capacity active, rather than the 12 gigawatts previously claimed. This discrepancy raises concerns about the company’s transparency regarding its GPU infrastructure investments, which amount to an estimated $50 billion. The majority of these assets, including NVIDIA's high-end H100 and H200 chips, appear to be sitting idle in warehouses or under construction, contradicting Microsoft's insistence that it has rapidly scaled up its AI infrastructure. This situation is significant for the AI and machine learning community, as it highlights the challenges and potential risks tied to the rapid expansion of AI technology. The apparent misalignment between capital expenditure and actual deployment of infrastructure could lead to a broader reckoning about the health and sustainability of the AI sector. With a market still buzzing from speculative investments in AI chips, such discrepancies challenge the credibility of capacity claims and fuel skepticism about whether the industry can meet increasing demand for computing resources effectively.