Vertex AI availability in europe-west2

Google Cloud Build with AI Custom ML & MLOps
1

Hi,

We would like to use the latest gemini models for our work, however require any use to be within europes-west2 - at present only 2.5 flash is available. Is there any indication on when we will get other models? its not great for redundancy either only having one model to choose from

1 Like
2

Hi Neil ….

I completely understand the frustration regarding redundancy and compliance. Operating within specific regions like europe-west2 is often a non-negotiable requirement for high-reliability and regulated domains.

While Google typically doesn’t announce exact dates for model rollouts in specific regions, here are a few technical strategies we use at Whitecyber Data Science Lab to manage these “Single-Region” constraints:

  1. Check for “Standard” vs “Flash”: Usually, the Flash models are deployed faster across regions for speed, while the Pro models (1.5 Pro) might take a bit longer to stabilize in every specific sub-region. Keep an eye on the Vertex AI Model Garden daily, as updates often appear there before the official documentation is updated.

  2. Multi-Region Residency: If your compliance allows, check if you can use europe-west1 (Belgium) or europe-west4 (Netherlands) as a failover. These regions often get newer models earlier than west2 and still keep the data within the EEA/EU boundaries.

  3. Architecting for “Model Agnostic” Failover: We always recommend building your application logic to be model-agnostic. If europe-west2 currently only has one Gemini model, ensure your orchestration layer can gracefully handle a fallback to other available models in that region (like PaLM 2 if applicable, though legacy) or implement a “Queued Retries” strategy if the primary model hits a quota limit.

  4. Data Residency vs. Compute: Double-check if you can use “Data Residency” settings to keep the data stored in London, while the “Compute” (inference) happens in a nearby region. Some Vertex AI services allow this “split” to maintain compliance while accessing newer models.

Redundancy is indeed a challenge when only one model is available. We hope the Pro 2.0 or 1.5 Pro variants land in London soon to provide that much-needed architectural depth!

Keep Calm … :blush: