Prepare and test your skills
Prepare and test your skills
Multi-cloud flexibility and interoperability is the ability to build a generative AI workload once and run it in Google Cloud, on-premises, or another cloud without rebuilding it. Google Cloud's open approach supports this ability with open software, leading machine learning frameworks, and choice of where models run. That combination keeps operations consistent across environments and prevents dependency on a single cloud provider.
Open standards are shared, non-proprietary technologies that let the same workload move between platforms. AI Hypercomputer (opens in a new tab) is a supercomputing system that combines performance-optimized hardware, open software, leading machine learning frameworks, and flexible consumption models. Because training, tuning, and serving use open software and familiar frameworks instead of a closed-only format, the workflow depends on portable tools rather than on one provider's proprietary stack. This emphasis on flexibility and choice lets customers build the AI platform that suits their needs and pursue better performance per dollar.
Deployment choice means an organization can place the same type of generative AI model in the managed cloud, in a self-managed cluster, or on-premises. Google Cloud extends AI-enabled cloud infrastructure from the cloud to on-premises deployments, so on-premises systems sit on one side of the boundary and Google Cloud services sit on the other, with the same models and containers able to move between them. Vertex AI Model Garden provides Google models such as Gemini and Imagen alongside select open-source software (OSS) models and Hugging Face text generation models.
The same open models can run in multiple destinations:
Operational consistency means the same operational practices work the same way in each environment. AI Hypercomputer co-designs compute, storage, networking, and software to optimize workloads across the entire AI lifecycle. That lifecycle moves in order:
Google Kubernetes Engine (GKE) carries this consistency into orchestration by managing high-performance, cost-effective training and serving with industry-leading scale. Features such as node auto-repair automatically replace unhealthy nodes, and horizontal pod autoscaling adjusts capacity, so the control plane manages health and scale the same way whether the cluster runs cloud-based training or large-scale serving.
Exam tip: Choose the open path when portability matters — an OSS or Hugging Face model that can deploy to Vertex AI or self-host on GKE avoids single-provider dependency.
Model choice is the ability to select from different generative AI models, and portability is the ability to run or move those models across environments without being tied to one vendor. Google Cloud's open approach supports both by offering first-party, third-party, and open-source models through managed or portable infrastructure. This approach helps leaders match models to performance and cost needs and reduce vendor lock-in.
Vertex AI (opens in a new tab) is Google Cloud's AI/ML platform for accessing, testing, tuning, and deploying generative AI models. Vertex AI provides Google's foundation models, such as Gemini and Imagen, and also gives access to other models through Vertex AI Model Garden. Vertex AI Model Garden is a model library where teams can discover, test, customize, and deploy select open-source models and assets. Third-party models can also be deployed to Vertex AI Model Garden or self-hosted on Google Kubernetes Engine (GKE) or Compute Engine, and models can be tuned for specialized tasks to improve performance.
| Model source | What it provides | Where it can run |
|---|---|---|
| Google models on Vertex AI | Gemini, Imagen, and other Google foundation models | Vertex AI with managed tools |
| Other models in Vertex AI Model Garden | Select open-source models and assets | Vertex AI Model Garden for discovery, testing, and deployment |
| Hugging Face text generation models | Open-source text generation models | Vertex AI or GKE |
| Self-hosted third-party or custom models | Third-party and custom-trained models | GKE or Compute Engine, with attached GPUs on Compute Engine |
Portability means a workload can run in more than one environment, so a team can change tools or move models without rebuilding everything. GKE supports portability and open standards because it is built on Kubernetes, which lets workloads run anywhere and helps avoid vendor lock-in. This portability is especially useful for organizations that already have Kubernetes expertise or a multi-cloud strategy. Vertex AI on Distributed Cloud extends a similar Vertex AI experience to secure on-premises, air-gapped environments by using the same tools, APIs, and models as Vertex AI on Google Cloud, which supports data sovereignty and compliance while keeping development familiar.
Choosing a deployment option means matching both the model and the place it runs to performance and cost requirements. Vertex AI is a fully managed, end-to-end platform that accelerates development and offloads infrastructure management, making it a good fit for teams focused on MLOps and rapid time-to-value. GKE is a customizable AI/ML platform for high-performance, cost-effective serving and training with flexible infrastructure, and it is the better fit when portability, open standards, or existing Kubernetes skills matter. Cloud Run is a serverless platform for containerized inference that can scale to zero, which suits event-driven applications and smaller models where cost efficiency is important.
| Option | Best for | Tradeoff or portability note |
|---|---|---|
| Vertex AI | Fully managed end-to-end platform for MLOps and rapid time-to-value | Less infrastructure control, managed by Google Cloud |
| GKE | Customizable AI/ML platform for high-performance serving and training at scale | Portable through Kubernetes, with more management responsibility |
| Cloud Run | Serverless containerized inference for event-driven apps and smaller models | Scales to zero, with less control than GKE |
Exam tip: Choose Vertex AI when you want a managed platform for model access and MLOps, and choose GKE when you need Kubernetes-based portability and open standards to avoid vendor lock-in.
Google's open-source contributions give the wider AI community shared building blocks, and Google Cloud turns those blocks into managed services without closing them off. Kubernetes, an open source container orchestration platform, and TensorFlow, an open machine learning framework, are the two best-known examples. Because anyone can use, inspect, and improve these projects, innovation spreads across many companies at once instead of staying inside one vendor, and Google Cloud customers inherit that shared progress.
Google Kubernetes Engine (GKE) (opens in a new tab) is a Google-managed implementation of the Kubernetes open source container orchestration platform, which means the control and automation come from Google while the underlying standard stays open. GKE runs containerized workloads of every kind, including AI and machine learning (AI/ML) applications, and supports both training large foundation models and serving inference requests at scale. Because GKE is built on an open standard rather than a proprietary one, the same workload definitions and operational skills carry over to other Kubernetes environments.
Google also contributes back to the Kubernetes ecosystem rather than only consuming it. Kueue is a Kubernetes-native job queueing system that manages resource allocation, scheduling, quota management, and prioritization for batch workloads. The Ray on GKE add-on brings an open-source framework for scaling distributed Python applications into the managed platform, letting teams run large-scale data preprocessing, distributed training, and online serving with minimal code changes.
AI Hypercomputer (opens in a new tab) is a supercomputing system that pairs performance-optimized hardware with open software, leading ML frameworks, and flexible consumption models. AI Hypercomputer is designed to boost efficiency across AI training, tuning, and serving. This combination matters because it lets teams keep using the frameworks and tools they already know instead of being forced onto a single vendor's stack.
Open models and open frameworks meet in the same place. Teams can choose a model from Vertex AI Model Garden, including models from Hugging Face, configure the deployment in GKE, and deploy it. Open models such as Gemma can also be deployed on GKE. TensorFlow Enterprise follows the same pattern, with customers such as Moloco running machine learning infrastructure on GKE and TensorFlow Enterprise.
Vendor lock-in is the risk that a customer becomes dependent on one provider's proprietary technology and cannot move without rebuilding everything. Open standards reduce that risk by keeping the interfaces and formats portable. Google Cloud recommends GKE when an organization wants portability and open standards, wants to run workloads anywhere with Kubernetes, and already has Kubernetes expertise or a multi-cloud strategy.
Portability is not limited to the public cloud. Google Distributed Cloud extends AI-enabled cloud infrastructure from the cloud to on-premises and edge deployments, so the same open foundation can run closer to where data is created. The tradeoff is between control and convenience, and Google Cloud offers a spectrum from fully configurable to fully managed:
| Service | Level of management | Best for |
|---|---|---|
| GKE | Google-managed Kubernetes with deep control | Custom pipelines, large-scale training, portability, and avoiding vendor lock-in |
| Vertex AI | Fully managed, end-to-end platform | Teams focused on MLOps and rapid time-to-value that want to offload infrastructure management |
| Cloud Run | Serverless, scales to zero | Event-driven applications and serving smaller models cost-effectively |
A collaborative ecosystem grows through participation, and Google Cloud's ecosystem includes partners that build on the same open foundations. Canonical, the publisher of the Ubuntu Linux operating system, works with Google Cloud to deliver trusted, secure open-source tools to millions of users. Other collaborations, such as those with NVIDIA, Datadog, and Broadcom, extend the same pattern into accelerated computing, monitoring, and cross-cloud network observability.
The community side is equally concrete. Google publishes open-source sample applications such as Online Boutique and Bank of Anthos to demonstrate cloud-native patterns, and GKE AI Labs offers experimental samples for accelerating AI/ML initiatives. Customers including Unity, HubX, and project44 build and serve their workloads on GKE, which shows the open foundation supporting production systems rather than only experiments.
Exam tip: Choose GKE when control, portability, and open standards matter most; choose Vertex AI when the priority is a fully managed platform with less infrastructure work.
Vertex AI is a fully managed end-to-end platform for MLOps and rapid time-to-value that offloads infrastructure management. Google Kubernetes Engine (GKE) is a customizable artificial intelligence and machine learning platform for high-performance, cost-effective training and serving that supports portability through the Kubernetes open standard. Cloud Run is a serverless platform for containerized inference that scales to zero for event-driven applications and smaller models where cost efficiency is important.
Google Cloud's open approach uses open software, leading machine learning frameworks, and deployment choice across managed cloud, self-managed clusters, and on-premises environments so the same workload can run without rebuilding. Because Google Kubernetes Engine is built on the Kubernetes open source container orchestration platform, workload definitions and operational skills carry over to other Kubernetes environments and help avoid vendor lock-in.
Vertex AI Model Garden provides Google models such as Gemini and Imagen alongside select open-source software models and Hugging Face text generation models for discovery, testing, customization, and deployment. The models can be tested, tuned, and deployed in Vertex AI, while Hugging Face text generation models can also run in Vertex AI or self-hosted on Google Kubernetes Engine, and third-party or custom models can be self-hosted on Google Kubernetes Engine or Compute Engine with attached graphics processing units.
An enterprise is planning its generative AI strategy across multiple business units. Some applications require complex multimodal reasoning, while others require high-volume, lightweight processing at minimal cost. The technology leadership team wants to ensure they can evaluate, deploy, and switch between Google first-party models, partner models, and open-source models without being locked into a single architecture.
Why does accessing this diverse range of models within Vertex AI Model Garden provide a strategic advantage for the organization?
It automatically converts prompt engineering instructions into BigQuery SQL queries to eliminate the need for API endpoints.
It forces all third-party and open-source models to share identical weights, parameters, and training datasets with Gemini.
It allows leaders to align specific tasks with the most cost-effective and performant models while mitigating vendor lock-in.
It removes all security and governance requirements so developers can deploy unverified open-source models without administrative approval.
Last updated