Illustration: a fortified vault door swung open, rows of chip trays sliding out on conveyors toward distant warehouses, while a large utility meter's dial spins beside a cloud of server racks, a rival conveyor of chips approaching from the far side. A generated image; it shows no data. Title: Where Alphabet's new AI money comes from: chips, cloud and the meter
AI-generated illustration
Article·AI Software

Where Alphabet's new AI money comes from: chips, cloud and the meter

The newest growth at Alphabet is not coming from search ads. It comes from Google Cloud consumption and from selling the TPU chips Alphabet used to keep for itself.

The read in plain words

Alphabet's new money is coming from places that look more like an infrastructure supplier than an ad business. Google Cloud is growing fast, and part of that growth comes from selling chips Alphabet used to build only for itself.2,1

The ad engine is still there. Google Services covers ads, Search, YouTube, Android, Chrome, Google Play, Google Maps and consumer devices.7 But the fastest-moving lines in the latest reports sit on the cloud side, and that is where this piece looks for the new money.

That matters for how investors read the company. A business that sells compute, chips and models to other companies takes its revenue from customers' AI budgets, not just from advertisers.

Two kinds of cloud money

Google Cloud revenue grew 103% to $30.7 billion, and the research note puts that down to demand for AI infrastructure and solutions.2 That is the headline figure behind this article.

The revenue comes in two forms. Services revenue comes mainly from consumption-based fees and subscriptions for Google Cloud Platform, Google Workspace and other enterprise services.8 Product revenue comes mainly from selling TPU systems, which are Alphabet's own AI chips packaged as complete hardware.9

The two are worth tracking separately. Consumption fees rise as customers run more AI workloads on Google's servers. Product sales arrive when a customer buys TPU systems outright, so they can be lumpier and depend on a smaller group of large buyers.

The chip business that used to be internal

The biggest change in where the money comes from is the TPU. Alphabet is now generating billions of dollars in revenue by selling homegrown AI chips that were previously reserved for internal use.1

That is a new market for Alphabet, not a bigger slice of an old one. Chips that once only cut Google's own costs now bring in revenue from other companies, and they show up as a product line inside Google Cloud.9

Our view: this is the part of the story to watch most closely. Selling chips puts Alphabet up against other chipmakers and against the in-house chips its cloud rivals sell, so the line depends on whether customers keep picking TPUs over the alternatives.

Models drive the meter

The model roadmap feeds the consumption side. On its July earnings call, Alphabet said it had launched Gemini 3.6 Flash and 3.5 Flashlight, that Gemini 3.5 Pro was in testing and that pretraining had started for Gemini 4.10

The smaller Flash and Flashlight models are built for high-volume use, and high volume is what consumption-based billing rewards.10,8 If newer Gemini releases bring more workloads onto Google Cloud, that shows up as metered fees, not one-off sales.

So the model pipeline and the chip business support each other. TPUs run the Gemini models, and both pull spending toward Google Cloud.

The leaders in our data are chasing the same money

Meta Platforms leads our score, Microsoft is second and Amazon is third. Each is going after the same kind of AI revenue from a different starting point.

Meta is building the Meta Enterprise platform, led by CJ Desai, to serve high-margin inference demand (the cost of running trained models for users) through an API, Muse code and other enterprise services.3 Meta also plans to maximize capacity for 2026 and 2027 because of demand constraints.11 One research note flags a trade-off: Meta's custom MI450X GPU cuts scale-up bandwidth to 900GB/s against 1,800GB/s for the standard part, which the note says hurts GenAI performance.12

Microsoft has reorganized into two segments, Agents and Infra, and Devices and Consumer.13 Revenue from OpenAI accounted for nearly 11% of its total quarterly revenue.4 Microsoft 365 Commercial cloud revenue grew 17% in fiscal 2026, helped by higher revenue per user from Microsoft 365 Copilot and E5, while seats grew 6%.14 That is a subscription upsell, which is a different route from Alphabet's chip sales.

Amazon's path looks most like Alphabet's. Amazon Web Services posted higher-than-expected revenue in the third quarter.15 Amazon has multi-year, multi-gigawatt commitments for its Trainium chips from the two largest AI labs, and on its call Amazon said more and more companies are using Trainium, which it read as a shift in market share.5,6 For Alphabet, that makes Trainium the clearest competitor in our data for the same customers its TPU business is going after.

What to watch next

The useful question is not whether Alphabet is growing. It is which part of the growth repeats. Three signals in the next reports would answer it.

First, the split inside Google Cloud between consumption and subscription fees and TPU product sales.8,9 A rising services line would mean the chips and models are pulling more metered workloads onto Google's servers. Second, whether TPU sales keep reaching billions as Trainium wins more customers.1,6 Third, whether the Gemini roadmap moves from testing to launch on the timeline the company has described.10

If those three hold up, Alphabet's revenue would rest more and more on AI infrastructure as well as advertising. That shift is the story to follow quarter by quarter.

What would change our view

  • If Alphabet's TPU sales stopped reaching billions, the new-market part of this read would weaken.1
  • If Trainium's gains among AI labs and other companies came at TPU's expense, the chip line would face tougher competition.5,6
  • If Google Cloud growth slowed sharply from 103%, AI infrastructure demand would no longer be carrying the mix.2
  • If Gemini 3.5 Pro or Gemini 4 stalled in testing or pretraining, the consumption side would lose a driver.10

If you buy AI compute, what decides whether you choose Google's TPUs, Amazon's Trainium or standard GPUs for your next workload?

Answer on X

Sources

  1. 1Industry research note, research note, Oct 29, 2026.
  2. 2Industry research note, research note, Oct 28, 2026.
  3. 3Earnings call META FY2026 Q3, transcript, Oct 21, 2026.
  4. 4Industry research note, research note, Oct 29, 2026.
  5. 5Earnings call AMZN FY2026 Q2, transcript, Jul 30, 2026.
  6. 6Earnings call AMZN FY2026 Q2, transcript, Jul 30, 2026.
  7. 7Industry research note, research note, Oct 28, 2026.
  8. 8Industry research note, research note, Oct 28, 2026.
  9. 9Industry research note, research note, Oct 28, 2026.
  10. 10Earnings call GOOGL FY2026 Q2, transcript, Jul 22, 2026.
  11. 11Earnings call META FY2026 Q2, transcript, Jul 29, 2026.
  12. 12Industry research note, research note, Sep 14, 2026.
  13. 13Industry research note, research note, Oct 28, 2026.
  14. 14MICROSOFT CORP 10-K (filed 2026-07-29), filing, Jul 29, 2026.
  15. 15Industry research note, research note, Oct 30, 2026.

For information only. Not investment advice.