presentofai

Google receives TPU v8i production shipments from Broadcom

TL;DR

Broadcom is shipping Google's TPU v8i production chips, confirming that the invisible custom-silicon supplier now drives more AI revenue than most companies generate in total.

What happened

  • Google is receiving TPU v8i production shipments from Broadcom, marking a new generation of custom AI accelerators for Alphabet's data centers.
  • OpenAI's first-generation Jalapeno accelerator is also shipping from Broadcom, and Meta is set to receive three generations of MTIA accelerators by end of 2027.
  • Broadcom's fiscal Q3 AI semiconductor revenue hit $16.70B, up 221% year over year, with custom XPUs accounting for 73% of that total.
  • Overall Broadcom Q3 revenue rose 85.5% to $29.59B, driven almost entirely by hyperscaler custom chip demand.
  • Broadcom's Q4 AI revenue forecast is $21.7B, a projected 236% increase, with CEO Hock Tan saying actual demand exceeds even that outlook.

Why it matters

  • Three of the biggest AI spenders (Google, OpenAI, Meta) are simultaneously receiving or awaiting Broadcom-built custom silicon, concentrating critical AI infrastructure supply in one supplier.
  • Hock Tan's claim that custom XPUs run at "half the cost of a GPU" is a direct commercial threat to both Nvidia and AMD, and hyperscalers are clearly buying the argument.
  • Broadcom generated $13.66B in free cash flow, 46% of revenue, giving it financial durability that merchant-chip rivals cannot match at current margins.
  • The TPU v8i shipment signals Google is deepening its custom-silicon independence, reducing reliance on Nvidia for its own workloads even as it resells Nvidia hardware to cloud customers.
  • Broadcom's gross margin is forecast to slip to roughly 73% from 78% a year ago, a sign that custom XPU volume growth is compressing unit economics even as absolute profits soar.

What to watch next

  • Whether Broadcom's Q4 $21.7B AI revenue guidance holds: land, power, and substrate supply constraints are the stated risks that could delay shipments regardless of demand.
  • Google's TPU v8i deployment scale: if Alphabet reduces external GPU procurement in parallel, it would be the clearest signal yet that custom silicon is displacing merchant accelerators at the top of the market.
  • Broadcom margin trajectory into 2027: a continued slide below 73% would test whether the custom-XPU model is structurally less profitable than it appears, and could reset investor expectations sharply.

Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.