Who Reaches $100 Billion in TPU Revenue First, Alphabet or Broadcom?
This post may contain links from our sponsors and affiliates, and Flywheel Publishing may receive compensation for actions taken through them.
Alphabet (NASDAQ:GOOGL | GOOGL Price Prediction) and Broadcom (NASDAQ:AVGO) have both reported earnings, and the TPU is central to each story. Broadcom designs the chips and ships them. Alphabet runs them inside Google Cloud and now also sells TPU systems to outside customers. Both are racing toward $100 billion in AI chip sales, and each one measures the race its own way.
Google Cloud More Than Doubles as Broadcom’s AI Chip Sales Triple
Google Cloud revenue reached $30.71 billion[1], up 103%. At that pace, TPUs have become a product line that brings in revenue. CEO Sundar Pichai said, “We’re also seeing strong demand for our TPU systems across a wide range of customers.”[1] Cloud operating income rose to $11.12 billion from $3.59 billion a year earlier, so the build-out is already paying off. It is also expensive. Capex hit $51.36 billion[1], and free cash flow fell to -$9.07 billion.
AI semiconductor revenue at Broadcom rose to $16.70 billion[2], up 221%. Custom accelerators (XPUs) made up 73%[3] of that AI revenue. Production shipments of TPU version 8i for Google have also started. CEO Hock Tan said Broadcom plans to deliver “multi-tens of billions of dollars of TPUs annually over the next several years.”[3] Free cash flow came in at $13.67 billion[2], or 46% of revenue. That shows how profitable it is to be the supplier.
| Business Driver | Alphabet | Broadcom |
|---|---|---|
| TPU Role | Owns, operates and sells TPU systems | Designs and supplies TPU silicon |
| Key Customers | Enterprises, Gemini users | Google, Anthropic, OpenAI, Meta Platforms (NASDAQ:META) |
| Free Cash Flow | -$9.07B[1] | $13.67B[2] |
Broadcom Crosses $100 Billion First, but Google Keeps More of the TPU
Broadcom expects about $115 billion[3] in AI semiconductor revenue in fiscal 2027 and $230 billion in fiscal 2028. Tan added, “Our demand actually exceeds this outlook.”[3] Those totals cover six XPU customers plus networking, and Tan expects Anthropic to become Broadcom’s largest XPU customer in 2027. So Google’s share of Broadcom’s total gets smaller as the total increases.
Alphabet does not report TPU revenue separately, so any timeline is an estimate. Counting all custom accelerators and networking, Broadcom reaches $100 billion in 2027. Counting TPUs alone, Google gets there in 2028. It makes money twice: from selling the hardware and from renting compute through Cloud. Cloud backlog exceeded $460 billion[4] as of Q1, though that older figure has likely changed since. Still, without a official TPU figure, the 2028 date could easily be off.
Supply Limits and Funding Will Set the Pace
Tan named substrates, HBM memory and power as limits on how fast the chips get used. A Singapore substrate fab starting in fiscal 2027 should ease one of those constraints. Broadcom guided gross margin to about 73%[3] because XPUs carry lower margins. I will keep an eye on whether operating margin holds near 66%[3].
For Alphabet, watch the TPU 8i ramp and Cloud margins. Funding is the weak spot. The company raised $28.7 billion[1] in new notes and still has $27.6 billion available in its at-the-market equity program, which points to more dilution.
Why Broadcom Holds the Clearer Path in the $100 Billion Race
Broadcom has the cleaner path to the target, and it gets paid without building data centers. Over the past month its stock rose 1.41% while Alphabet’s fell 16.7%. I read that gap as investors concerned about Alphabet’s cash burn. Broadcom offers the more direct exposure to AI chip sales. Alphabet’s case rests on owning Gemini, Cloud and the chips and building on that advantage for longer (we reverse-engineered what the biggest tech winners looked like early and put the pattern in a free playbook here). I would change my view if Alphabet started reporting TPU revenue or got back to positive free cash flow.
Contact [email protected] for any questions or corrections.







