You're correct in saying AI training is one of the most computationally intensive tasks in commercial technology. Because it relies on distributed GPU clusters optimized for high throughput and memory bandwidth. The compute required to train the major AI systems has doubled approximately every 6 months. Training AI is computationally intensive because it involves performing numerous floating-point operations (FLOPs) to adjust billions of parameters across massive datasets. I suspect that's one area where they grossly missed estimating the compute requirements. But, that still doesn't explain the data center buildout, IMO.
However, consider this - the Depository Trust & Clearing Corporation (DTCC) is the institution provider that automates, centralizes, and standardizes the clearing and settlement of securities transactions for the global financial system. DTCC manages assets valued at approximately $114 trillion in its central securities depository subsidiary, DTC, as of June 2026. In terms of transaction volume, the DTCC processes over $2.5 quadrillion in securities transactions annually. Managing that transaction volume calls for a lot of computing power. DTCC has already begun live production trades using tokenized stocks, bonds, and treasuries, and most people do not even know it is happening. The data center buildout required to run this system is so enormous that it tells you something about the intended scale that no official will say out loud.
You're correct in saying AI training is one of the most computationally intensive tasks in commercial technology. Because it relies on distributed GPU clusters optimized for high throughput and memory bandwidth. The compute required to train the major AI systems has doubled approximately every 6 months. Training AI is computationally intensive because it involves performing numerous floating-point operations (FLOPs) to adjust billions of parameters across massive datasets. I suspect that's one area where they grossly missed estimating the compute requirements. But, that still doesn't explain the data center buildout, IMO.
However, consider this - the Depository Trust & Clearing Corporation (DTCC) is the institution provider that automates, centralizes, and standardizes the clearing and settlement of securities transactions for the global financial system. DTCC manages assets valued at approximately $114 trillion in its central securities depository subsidiary, DTC, as of June 2026. In terms of transaction volume, the DTCC processes over $2.5 quadrillion in securities transactions annually. Managing that transaction volume calls for a lot of computing power. DTCC has already begun live production trades using tokenized stocks, bonds, and treasuries, and most people do not even know it is happening. The data center buildout required to run this system is so enormous that it tells you something about the intended scale that no official will say out loud.