Broadcom saw its TPU compute shipments double between fiscal Q2 and Q3 2026, reaching nearly 2 million Nvidia H100 equivalents. While the total volume of physical chips increased by approximately 65% to 800,000 units, the overall computing capacity grew faster due to a shift toward higher-performance hardware.
The TPU v7 model became the primary component in the company's shipment mix during the third quarter. This version provides roughly 2.5 times the performance of the previous TPU v6e. Meanwhile, initial production has commenced for the TPU v8i, which is expected to deliver about 2.2 times the performance of the v7.
Broadcom reported that XPU shipments rose more than 3.5 times year over year in fiscal Q3 2026. These custom accelerators generated 73% of the company's AI revenue, which exceeded $16.7 billion. This revenue is derived from a base of six specific customers, including Google.
The long-term agreement between Broadcom and Google covers future TPU development and AI networking, with projected annual values in the tens of billions of dollars. Despite this growth, analysts note risks related to revenue concentration, as the company relies on a limited number of clients for its custom silicon business.
