What is the true scale of a TPU superpod? The true scale of a supercomputer isn't defined by how many processors you can fit in a room; it’s defined by how many processors can effectively think as one. This distinction is the driving force behind Ironwood, our seventh-generation Tensor Processing Unit (TPU), and the architecture of the TPU superpod. On paper, the scale of an Ironwood superpod is impressive: it scales up to 9,216 chips in a single interconnected domain. But the true scale lies in the fabric that binds them. In an Ironwood superpod, we use a breakthrough Inter-Chip Interconnect (ICI) networking technology running at 9.6 terabits per second. This allows those 9,216 chips to communicate with near-instant latency, accessing a massive 1.77 Petabytes of shared High Bandwidth Memory (HBM). So, how do you connect nearly 10,000 chips without creating a cabling nightmare? The answer lies in the Optical Circuit Switch (OCS). ◾ Ironwood clusters are built from building blocks called "cubes" (64 chips each). ◾ To scale beyond a single cube, multiple cubes are connected using an Optical Circuit Switch (OCS) network. This is a dynamic, reconfigurable optical network that connects entire cubes, allowing the system to scale from a small "pod" (e.g., a 256-chip Ironwood pod with four cubes) to a massive "superpod" (e.g., a 9,216-chip system with 144 cubes). ◾ This OCS-based topology is key to fault tolerance. If a cube or link fails, the OCS fabric manager instructs the OCS to optically bypass that unit and establish new, complete optical circuits swapping in a designated spare. → Hardware is only half the equation. The true scale of a TPU superpod is unlocked by a co-designed software stack for faster, more efficient outcomes. Thanks to deep integrations across the stack—from data-center-wide hardware optimizations to open software and managed services—Ironwood TPUs are our most powerful and energy-efficient TPUs to date.