OpenAI is publicly planning to obtain Tensor Processing Units from Google for training and operating AI models. The ComputerBase note says financial details, chip count, and delivery schedule are not known, while OpenAI remains one of Nvidia's largest customers and seeks alternatives to cover demand and reduce dependence on a single supplier. The limit is a confirmed purchase; the measurement is a reported plan, and it does not sit inside the limit.

The desk reads this as a procurement hedge, not a measured replacement of Nvidia GPUs. The document proves intent to diversify, not a capacity shift, because no volume, price, or delivery window is attached. Until a chip count exists, the buyer's constraint remains Nvidia allocation, and Google's TPU demand stays unpriced for the fab.

Still unmeasured are the number of TPUs, the purchase price, the delivery window, the TPU generation, the training workload, and a contract value. The text also does not name a leaker, a deal, a Google Cloud region, or a model family. Without those figures, the page remains a standing note that can be corrected when a quantity appears.

The desk reads this as a procurement hedge, not a measured replacement of Nvidia GPUs. Until a chip count exists, the buyer's constraint remains Nvidia allocation, and Google's TPU demand stays unpriced for the fab.

The document leaves out the chip count, price, delivery timeline, and TPU generation.

Treat the report as a demand signal, not a capacity plan. No chip count, price, or delivery window changes the Nvidia allocation baseline.