Google expands on-premises AI for sovereign data needs

Google is expanding its push into on-premises artificial intelligence through Google Distributed Cloud, targeting organisations with strict data sovereignty and compliance requirements.

The offering is aimed at enterprises and governments that want to keep sensitive data in local environments while using newer AI systems. It is positioned as a way to run AI workloads in private data centres or at the edge, rather than relying solely on public cloud infrastructure.

In its latest survey of more than 1,400 senior IT leaders, Google found that 48% were prioritising infrastructure with data residency controls to support compliance with local data security laws. The research also found that 52% of organisations now use a hybrid cloud approach to AI, combining on-premises systems with multicloud environments.

That reflects a broader shift in how large organisations are approaching AI deployment. Businesses and public sector bodies have faced a trade-off between using advanced AI services and keeping data within tightly controlled jurisdictions.

Google argued that three risks are shaping those decisions. The first is jurisdictional risk, as changing regulation, intellectual property concerns and foreign data access requests push organisations to keep data local. The second is economic independence, with some customers concerned that reliance on foreign infrastructure providers could expose critical services. The third is geopolitical risk, especially for operators of essential services seeking to reduce exposure to global disruption.

Sovereignty focus

Google Distributed Cloud is aimed at that segment of the market. The platform extends Google Cloud services into customer-controlled environments and is available in two deployment models.

One is an air-gapped version, fully disconnected from Google Cloud and the public internet. In that setup, Google says the system cannot be remotely shut down by the company. The other is a connected version, which uses a Google-managed software lifecycle while running on a customer's existing hardware.

The service provides an on-premises AI environment that includes infrastructure for AI workloads, access to Gemini models and open models, and inference services. The emphasis is on giving customers a way to build and run AI systems while keeping data under local control.

The announcement comes as cloud providers compete to address growing concern among regulated industries and public sector customers. Demand for AI tools has risen quickly, but many organisations still face limits on where data can be stored, processed and accessed.

Hybrid cloud has emerged as one way through that problem. By splitting workloads between local systems and remote cloud environments, organisations can keep sensitive data in one location while using external resources for other tasks.

The model has gained traction because it offers a compromise between control and access. Public cloud providers still offer broad access to AI models and computing resources, but local hosting remains attractive to customers that need to meet domestic rules on residency, privacy and operational resilience.

Market pressure

For Google, the on-premises and hybrid segment is strategically important. Amazon Web Services, Microsoft and other suppliers have all made similar moves to meet demand from customers in government, defence, healthcare, finance and critical infrastructure.

Those buyers often need formal guarantees around where data sits and how systems can operate if external connectivity is lost. Air-gapped systems in particular appeal to users that want complete operational separation from the public internet.

Google's reference to remote shutdown is likely to resonate in that context. Concerns about dependence on overseas providers have become more prominent as governments review digital infrastructure through the lens of national resilience and strategic autonomy.

The survey suggests compliance is no longer a niche issue in AI infrastructure planning. Nearly half of senior IT leaders in the study said data residency and related controls were now a priority, indicating that governance requirements are shaping architecture choices alongside cost and performance.

At the same time, the finding that more than half of organisations have adopted a hybrid cloud approach to AI suggests many customers do not see a single-environment strategy as workable. Instead, they are building a mix of local and external systems to balance regulation, security and access to newer AI tools.

Google framed that trend as evidence that on-premises AI no longer has to mean isolation from current model development. Its pitch is that customers can keep sensitive workloads close to home without giving up access to advanced AI services.