Google is the only major hyperscaler that doesn't pay Nvidia's margins for most of its AI compute, because it designs its own chips, in-house, all the way down.
Who they are
Google, the search engine that became the core of parent company Alphabet Inc., is one of four hyperscalers spending the most on AI infrastructure worldwide. What sets it apart structurally is vertical integration: unlike AWS and Azure, which primarily resell Nvidia-based compute, Google designs its own AI accelerators, called Tensor Processing Units (TPUs), and runs its own Gemini model family on top of them. CEO Sundar Pichai leads both Alphabet and Google.
What they do
Google Cloud rents out compute and AI models to outside customers, while Gemini serves consumers and enterprises directly, and the two businesses reinforce each other. At its Cloud Next 2026 conference in April, Google unveiled its 8th-generation TPUs, split into a training chip (TPU 8t) and an inference chip (TPU 8i) that links 1,152 chips into a single pod and delivers 80% better performance per dollar than the prior generation. Gemini is now processing more than 16 billion tokens per minute through direct customer API use, up 60% from the prior quarter.
How Google makes money
Google Cloud revenue hit $20.0 billion in the first quarter of 2026, up 63% year over year, the fastest growth of any major hyperscaler cloud business that quarter, and its backlog of signed-but-unfulfilled contracts nearly doubled quarter over quarter to more than $460 billion. Alphabet's total annual revenue exceeded $400 billion for the first time. CEO Sundar Pichai told investors plainly that Google is "compute constrained" and that cloud revenue "would have been higher" if it could build data centers fast enough to meet demand.
The bigger trend
Alphabet raised its 2026 capital-expenditure guidance to $180–190 billion, more than double the $91.4 billion it spent in 2025, funded partly by an $80 billion equity raise. Because Google owns both the Gemini models and the TPU chips they run on, it avoids paying Nvidia's hardware margins on a large share of its AI compute, a structural cost advantage AWS and Azure don't have to the same degree.
Whether the new TPU 8t/8i generation ships fast enough to relieve the "compute constrained" bottleneck Pichai described; how much of the record $460 billion cloud backlog converts into recognized revenue as capacity comes online through 2026 and 2027; whether rising depreciation from the capex surge weighs on margins before AI revenue catches up.
Related companies
Frequently asked questions
Alphabet Inc. is the publicly traded parent company, ticker GOOGL. Google is its largest subsidiary, encompassing Search, Cloud, YouTube, and Gemini. The names are used interchangeably here because Google represents the vast majority of Alphabet's revenue and AI infrastructure spending.
Google has designed its own Tensor Processing Units since 2015. Owning the chip design lets it avoid paying Nvidia's margins on a large share of its AI compute and optimize hardware specifically for its Gemini models, though Google still uses Nvidia GPUs for some workloads and customers.
CEO Sundar Pichai said Google could not build data centers fast enough to meet demand for its AI services in early 2026, meaning cloud revenue would have been higher if more capacity had been available, a supply problem rather than a demand problem.
This page describes public value-chain positioning for informational purposes only. It is not investment advice, and inclusion here is not a recommendation to buy or sell any security. Figures reflect public reporting as of mid-2026 and may have changed since.