Google Puts Price Tag on Ironwood TPU Amid Nvidia Cost Pressure

Google has finally put a price tag on its seventh-generation Tensor Processing Unit (TPU), code-named Ironwood. The chip, which reached general availability in April 2026, was initially released without a public rate for three months. However, the company has now listed Ironwood at $12.00 per chip-hour in Iowa and $13.20 per chip-hour in London on its Cloud TPU pricing page.

The on-demand price is part of a tiered pricing system that also includes flexible capacity through Dynamic Workload Scheduler Flex-start pricing and committed-use rates for one- or three-year terms. In both regions, the committed-use rate for a three-year term is $5.40 per hour, while the on-demand rate in Iowa is higher than Nvidia's B200 chip.

The price of Ironwood is significant as it comes alongside a clearer picture of what the chip actually is under the hood. Each Ironwood chip carries 192GB of HBM per chip and has approximately 7.37 TB/s of memory bandwidth. Google's design reportedly lets Ironwood pods scale up to 9,216 chips in a single superpod.

This development comes as Nvidia faces pressure from its biggest customers who are being told to expect higher costs on next-generation Grace Blackwell and Vera Rubin systems due to climbing memory prices. The release of Ironwood's pricing information is seen as a strategic move by Google to make the economics visible, tier by tier, region by region.

It's worth noting that Anthropic negotiated a rate for Ironwood capacity at roughly $1.60 per TPU-hour, which is a fraction of the now-published $12.00 on-demand price. This highlights the spread between what Google will charge a marquee AI lab chasing training capacity and what everyone else pays without leverage.