A third of this exam is Google's own offerings, and this section is the argument underneath them. Each claim has a concrete thing behind it, and the questions test whether you can name that thing.
The AI first claim
Google's position is that it has been building AI into its products for a long time, that the transformer architecture behind current models came out of Google research, and that the same models running its own products are what it sells. The business reading is that gen AI is not a recent addition to the catalogue.
The related claim is the ecosystem. Search, Workspace, Android, Chrome and Cloud all carry the same underlying models, so an organisation already using Google products meets AI where its people already work rather than in a new tool.
Enterprise ready
Google uses the phrase to mean five things, and each is answerable.
Responsible. Published AI principles, safety filters and the Secure AI Framework.
Secure and private. Data encrypted in transit and at rest, customer data not used to train Google's models, and controls over where processing happens.
Reliable and scalable. The same infrastructure that runs Google's consumer products, with published availability commitments.
Governed. Access control, audit logging and data residency configurable per workload.
The point of the phrase is the gap between a demonstration and a system a regulated business can put in front of customers. Most of that gap is these five things rather than model quality.
The infrastructure underneath
Data centres and ordinary cloud computing are the base.
GPUs are general purpose parallel processors, widely used for training and inference, and available on Google Cloud.
TPUs, Tensor Processing Units, are Google's own silicon designed for the tensor operations machine learning performs. Being purpose built for a narrow job is what makes them efficient at it, in performance and in energy.
AI Hypercomputer is the name for the whole arrangement, hardware, networking and software treated as one system rather than a pile of machines. The claim it supports is that training and serving at scale is a systems problem, not a chip problem.
The open approach
Google contrasts its position with a closed one, and it means three things. Open weight models such as Gemma that you can take elsewhere. Third party and open models offered alongside Google's own in Model Garden. And open standards for connecting agents and tools rather than a proprietary protocol.
The business argument is avoided lock in. You can move a workload, and you can mix models from several providers behind one platform.
Democratising development
The last claim in this section is that building with AI should not require a research team. Low code and no code tooling, pre trained models that need no training data, and APIs for common tasks such as speech, translation and document understanding.
For a product person this is the practically useful part. It decides whether a capability is a quarter of engineering work or an afternoon of configuration.