Google's Machine: Gemini, TPUs and the Capex Wave
By AKA Studio
What’s inside
- NVDA NVIDIA: Google still buys accelerators at scale alongside its own TPUs; the capex line lands here first.
- VRT Vertiv: Power and cooling kit for the campuses Alphabet is building at record pace.
- ETN Eaton: Electrical gear feeding Google's buildout, where supply not demand is the stated constraint.
- AVGO Broadcom: Co-designs Google's TPU program — the most direct listed proxy for Alphabet's silicon bet.
- TSM TSMC: Fabricates Google's TPUs; Alphabet's chip ambitions run through Taiwan.
- ASML ASML: Every TPU generation Pichai promises depends on leading-edge lithography.
- MU Micron: Each TPU rack Google ships needs high-bandwidth memory alongside the logic die.
- CRWV CoreWeave: Alphabet explicitly plans to rent outside compute while supply-constrained.
- EQIX Equinix: Colocation and interconnection absorb hyperscale spill-over as Google adds regions.
- ANET Arista Networks: TPU pods sold on-premise to customers still need hyperscale-class networking fabric.
- CEG Constellation Energy: Google signed a long-term deal to buy 890MW of its nuclear power for AI data centers.
- GEV GE Vernova: Turbines and grid equipment gate how fast Alphabet's new campuses can energize.
- PWR Quanta Services: Interconnects and substations are the physical bottleneck Alphabet keeps flagging.
- GOOGL Alphabet: The subject itself: Gemini, Search, YouTube and Cloud in one line.
- APP AppLovin: AI-run ad auctions are the model Google is racing to keep owning across the open web.
Why it could work
If Alphabet's record capex converts into cloud revenue and Gemini keeps Search intact, the whole chain it buys from — silicon, power, cooling, launch — compounds with it.
What could break it
Capex that outruns revenue turned Alphabet cash-flow negative in the first half of 2026, and the market answered with its longest monthly losing streak since 2015.
For learning only. Not investment advice.