Anthropic now runs Claude itself on Google Cloud — managed option goes GA with day-zero agents, skills and code execution

• Google Cloud announced the general availability of an Anthropic-managed deployment option for Claude on October 23, 2026 (post by Michael Gerstenhaber, VP of product management for Gemini).
• Anthropic operates the stack as the data processor on Google Cloud infrastructure, under Anthropic's Commercial Terms — authentication stays on Google Cloud IAM, billing on the Google Cloud invoice.
• Day-zero platform features on the new option: Claude Managed Agents (beta), skills, code execution, computer use, Files API, web search, and an MCP connector (beta), per Anthropic's announcement.
• Available models named: Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 — day zero on both options; Anthropic-managed currently runs on US and Global endpoints.
• Both options carry the same pricing and draw down eligible Google Cloud commitments.
• Google-managed keeps VPC Service Controls, customer-managed encryption keys, data residency in the US/EU/APAC, and BigQuery + Model Armor integrations; new features arrive there after Google integrates them.

Google Cloud announced the general availability of an Anthropic-managed deployment option for Claude Models on Google Cloud, sitting alongside the existing Google-managed option. Under the new route, Anthropic operates the stack as the data processor on Google Cloud infrastructure, under Anthropic's terms of service — and new Claude Platform features arrive the day they launch, instead of whenever Google finishes integrating them. The post is dated October 23, 2026.

The economics are deliberately boring: same frontier models on day zero, same pricing, same Google Cloud billing. Per the announcement, "the difference comes down to who acts as your data processor and whether you need access to the new Claude platform features when they launch." The trade is legal and compliance-shaped: Google's own comparison lists VPC Service Controls, customer-managed encryption keys and US/EU/APAC data residency only under the Google-managed column.

What the Anthropic-managed option includes

Anthropic's own announcement the same day says the offering carries the full set of Claude Platform features, with new features and betas shipping the same day they go live on the native Claude Platform. Named in the two posts: Claude Managed Agents (beta) for running long tasks without managing infrastructure, skills, code execution, computer use, agent skills, advisor strategy (beta), a Files API, citations, web search and web fetch, an MCP connector (beta) and prompt caching.

Operationally, it stays inside Google Cloud's house: teams authenticate with their existing Google Cloud IAM credentials, keep Cloud Audit Logging, and bill through a single Google Cloud invoice, with usage drawing down eligible Google Cloud commitments. Anthropic says the option is currently available with US and Global endpoints, and names Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 as available — new models hit both deployment options on day zero.

What stays Google-managed, and who should keep it

The Google-managed option continues as the compliance-first route: Google Cloud stays the data processor under customers' existing agreements, with VPC Service Controls, CMEK, and pinned data residency in the U.S., EU and APAC, plus native hooks into BigQuery and Model Armor. New Claude Platform features reach it only after Google integrates them into its own infrastructure.

That's the honest split: anything that needs VPC perimeters, customer-held keys or pinned residency stays Google-managed; agent workloads that want Managed Agents, skills or code execution on launch day are candidates for Anthropic-managed — but under Anthropic's Commercial Terms, not the Google Cloud agreement they already hold. Metarix's read is blunt on the procurement point: changing the data processor is a procurement event, so legal and security sign-off should come before the first API call, not after.

The fine print worth a procurement call

The blog's author is Michael Gerstenhaber, vice president of product management for Gemini at Google Cloud (via WebProNews), and his framing is that the decision is about the processor, not the technology. One asymmetry to note: Google's post claims the governance features — VPC Service Controls, CMEK, regional residency — only for its own option, and makes no such claim for the Anthropic-managed one. Don't assume parity until Anthropic's documentation says otherwise.

Why the timing matters

Skills aren't just a feature here — they're the subplot. This lands the same week Anthropic said it is handing its Agent Skills standard to the Agentic AI Foundation, the Linux Foundation group that also looks after MCP, announced at AGNTCon + MCPCon North America in San Jose (reported October 22–23), alongside Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba — taking the foundation to ten projects. So Anthropic is simultaneously open-sourcing the skills standard and shipping skills as a day-zero feature on Google Cloud, which is a neat bit of staging: the standard goes neutral, the implementation goes to market.

For builders, the week reads as one story: Okay News notes the Anthropic-managed service was published October 23 with the same commitment-drawdown mechanics as the Google option. If you've been calling Anthropic's API directly because you needed managed agents or code execution, Google Cloud just removed the excuse — the only question left is whose data-processing terms your contract allows.