At Google Cloud Next 2026, Google folded Vertex AI, Agentspace and Gemini Code Assist Enterprise into a single control plane called the Gemini Enterprise Agent Platform. Google insists it’s a rebrand, not a retirement: existing Vertex AI workloads, SDKs and billing migrate with no breaking changes, according to TheNextWeb’s coverage of the keynote. But “same APIs, radically expanded scope” is exactly the kind of announcement enterprise buyers should read carefully, because it’s a strategic tell about where Google thinks the next few years of cloud AI spend are going: agents, not chat.
What actually changed
The Model Garden now hosts more than 200 foundation models — Google’s own Gemini and Gemma families, open models like Llama, Qwen3 and Mistral, and, notably, Anthropic’s Claude family (Opus 4.7, Sonnet, Haiku) running natively in the same catalog, billed serverlessly with no separate infrastructure to provision. Google also split the product in two: the Gemini Enterprise Agent Platform (Agent Designer, Agent Engine Sessions, Memory Bank) targets IT teams building agents at scale, while the simpler Gemini Enterprise app, with a no-code Workspace Studio builder, targets business users directly. Pricing starts at $30/seat/month for the standard tier and $21/seat/month for the business tier, per VentureBeat’s reporting.
Underneath the product layer, Google’s Agent2Agent (A2A) protocol reached version 1.2 with 150 organizations now running it in production, not pilot. It’s governed by the Linux Foundation’s Agentic AI Foundation rather than by Google alone — and Microsoft, AWS, Salesforce, SAP and ServiceNow are all listed as running A2A in production. That’s worth sitting with: Google’s direct rivals are adopting its interoperability protocol even while competing head-on for the same platform budget.
A three-way platform war, not a model war
Google, AWS and Microsoft are now racing to be the control plane for enterprise agents, not just the model host. Google Cloud CEO Thomas Kurian used his keynote to frame rivals as “handing you the pieces, not the platform” — Google’s pitch is a single integrated stack: custom Ironwood TPUs (4.6 petaFLOPS per chip), the Gemini 3.x model family, the agent platform itself, and distribution through Workspace’s 3+ billion users.
That full-stack argument has to be weighed against market-share reality: Google Cloud sits at roughly 11% of the cloud infrastructure market versus AWS’s 31% and Microsoft’s 25%, even after reportedly growing about 50% year-over-year in Q4 2025 — the fastest of the three hyperscalers. AWS answered days later with Quick Suite, a browser-extension frontend to Bedrock AgentCore that plugs into Chrome, Outlook, Word and Slack with 15+ enterprise connectors, per VentureBeat. Microsoft’s competing bet is Foundry, renamed from Azure AI Foundry in January 2026.
What this means if you’re choosing a cloud AI vendor
The decision enterprises face isn’t really “which model is smartest” anymore. With Claude, Gemini, Llama and 200+ other models sitting side by side in the same Model Garden, model choice is becoming commoditized — you can call Claude Opus through Google’s own billing surface today. What still differentiates vendors is platform: governance tooling, agent interoperability, identity and security architecture (Google’s Agent Development Kit reached stable v1.0 with Model Armor for prompt-injection defense), and which ecosystem your data and workflows already live in.
That reframes the vendor-lock-in conversation. The model is increasingly portable — the same Claude model runs inside AWS Bedrock, Google’s Model Garden, and Anthropic’s own API. The new lock-in surface is the agent tooling wrapped around it: session memory, multi-agent orchestration, and now A2A interoperability, which — because it’s Linux-Foundation-governed and adopted across competing platforms — may end up mattering more to a buyer than platform loyalty itself.
- If your stack is already GCP/Workspace-native, this consolidation lowers integration friction — one billing surface, one control plane for models and agents.
- If you’re AWS- or Microsoft-native, cross-vendor A2A adoption means you can plan for agent interoperability across clouds instead of betting on a single platform.
- Either way, evaluate agent governance and security tooling (prompt-injection defenses, identity architecture) as closely as you evaluate the model roster — that’s where the real differentiation now sits.
The signal behind the numbers
Google cited case studies to back the pitch: Danfoss automated 80% of transactional email decisions, cutting response time from 42 hours to near real-time, and Suzano cut SQL query time by 95% using the platform’s agent tooling, according to TheNextWeb. Whether or not Gemini Enterprise is the platform you pick, the underlying trend is the one worth planning around: 89% of business teams already use AI agents in some form, running an average of 12 agents per organization. The platform war between Google, AWS and Microsoft is really a fight over who manages that sprawl for you — and that’s a harder, stickier sell than simply hosting the best model.