Google Distributed Cloud (GDC) air‑gapped has been extended with Gemini‑driven enterprise search, the Gemini 3.5 Flash model, and NVIDIA Blackwell‑Ultra (HGX B300) hardware. The additions are generally available and aim to let teams deploy modern agentic workloads inside isolated environments while keeping data fully under control.
New AI capabilities for sovereign environments
Gemini on GDC enterprise search is now in preview. It provides a managed conversational interface that can ingest text, audio, and video, and returns summaries or answers without sending data outside the perimeter. The service uses Gemini embedding models for multi‑modality and includes built‑in retrieval‑augmented generation (RAG) that stays inside the air‑gapped zone.
Frontier and open‑weight models are also expanded. Gemini 3.5 Flash is generally available on GDC air‑gapped, offering ultra‑low‑latency inference, multimodal understanding, and long‑context processing. In addition, the platform now supports the open models Gemma and Mistral, giving engineers the option to match model characteristics to specific workloads on the same governed stack.
Infrastructure upgrades that enable high‑performance AI
The compute layer now includes NVIDIA HGX B300 systems built on the Blackwell‑Ultra architecture. These GPUs are positioned to serve Gemini and other models at production scale within the isolated environment.
Network throughput has been raised: virtual machines can now use up to 32 Gbps per VM, up from the previous 25 Gbps. This higher bandwidth reduces data movement bottlenecks between CPUs and GPUs during inference pipelines.
Managed Cloud DNS is generally available on GDC, exposing the same Kubernetes‑native APIs and CLI tools that teams use in public clouds, but without requiring external DNS infrastructure inside the air gap.
Security and operational considerations
Node‑to‑node traffic can be protected with hardware‑offloaded IPsec encryption capable of up to 100 Gbps. This meets high‑throughput encryption‑in‑transit requirements while preserving CPU cycles for workloads.
Integration with Elastic Security brings machine‑speed threat detection and response to the distributed environment. Practitioners should evaluate how Elastic’s alerts and policies map onto the GDC node topology and whether existing SIEM processes need adjustment.
Related CloudNinjas coverage: Google Cloud.
What This Means For Practitioners
- AI engineers can prototype or migrate Gemini‑based RAG applications without redesigning data pipelines for external clouds.
- Platform teams should inventory GPU resources and verify that the new Blackwell‑Ultra nodes are provisioned with the required drivers and runtime libraries.
- DevOps/SRE staff need to update networking configurations to leverage the 32 Gbps VM limit and ensure IPsec policies are applied consistently across nodes.
- Security engineers must validate that Elastic Security integrations respect the air‑gap boundaries and that DNS management does not re‑introduce external dependencies.
- Future evaluations should track the rollout of additional open‑weight models and any changes to the preview status of Gemini enterprise search.
