Product1 publisher3 min readPublished
Salesforce begins migrating select US customers to Google Cloud in Q4 2026
The expanded partnership also makes Salesforce capabilities callable from Gemini Enterprise with no connector to build, because both sides implement the open Model Context Protocol. Teams already on Agentforce get that part without new work.
The Product Desk · Product desk
What happened
- Salesforce and Google Cloud expanded their partnership at Dreamforce on September 15, 2026, connecting Salesforce's headless architecture to Gemini Enterprise so agents on either platform reason and act on the same data.
- Salesforce will begin migrating select customers in the United States onto Hyperforce on Google Cloud in the fourth quarter of 2026.
- SharkNinja is named as a joint customer pairing Gemini Live's real-time multimodal capabilities with Agentforce for consumer interaction over a product's life.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- cost The connector build between Agentforce and Google's agent window comes off the engineering plan, and the remaining work is scoping: deciding which of thousands of capabilities an outside agent may call. That lands on Salesforce admins, not the integration team.
- exposure Anything published through the headless layer is reachable by any model or agent that speaks the protocol. Permissions become the boundary that connector code used to be.
- decision A Q4 migration slot puts the infrastructure decision on Salesforce's calendar, likely before most teams have finished testing the agent interop that shares the announcement.
- precedent Because the interop is an implementation of an open standard, the next hyperscaler to implement the headless architecture can make the same no-connector claim without a separate Salesforce deal.
"Without custom integrations" is a specific engineering claim [2], and the specification behind it is the Model Context Protocol, implemented on both the Salesforce headless layer and Gemini Enterprise [3]. Salesforce puts the number of exposed capabilities in the thousands, callable by any model, agent, or interface [4]. It names three examples: pipeline health checks, account summaries, and case triage, across Sales, Service, Finance, and Advertising workflows [5][6]. Someone still has to decide which of the thousands a Google-side agent may call, and who signs off when one of them writes to a record.
"Enterprise AI only pays off when agents can understand a company's business context and act on it," said Kevin Ichhpurani, President, Global Partner Ecosystem at Google Cloud [7]. The enforcement the release describes is Tableau MCP, which applies governance, row-level security, and semantic models on every agent query [8]. An admin can point to that when a sales leader asks where a Gemini Enterprise answer got its pipeline number.
On the infrastructure half, "What changes here is the infrastructure underneath our customers, not the trust they've built with us," said Meir Amiel, President, Chief Trust and Infrastructure Officer at Salesforce [9]. Hyperforce on Google Cloud is already handling live production customer traffic [10], and migration of select US customers starts in the fourth quarter [11]. That quarter opens on October 1, sixteen days after the September 15 announcement [12]. Salesforce did not name the customers or publish a price.
One joint customer is named. Evan Gerber, a vice president at SharkNinja, said the company is "creating a new way for consumers to engage with SharkNinja throughout the life of a product" by pairing Gemini Live with Agentforce [15]. Teams that already use Google's models in Agentforce have part of the stack in place, since those models can power Prompt Builder and the Reasoning Engine [14].
The two announcements have very different reversal costs. Turning on the Gemini Enterprise interop is configuration on an open standard, and Google Cloud is one of the first hyperscalers to implement the headless architecture [13], which means a second implementer could offer the same surface. Moving production workloads to Hyperforce on Google Cloud is a contract and a migration plan.
The two do not depend on each other. For the interop, sort your Salesforce capabilities on two axes: read or write, and whether you would accept an agent outside Salesforce running it unattended. Read-and-unattended covers account summaries and pipeline health checks, and it can go into a pilot next week. A triage action that changes a case record is write-and-unattended, and it waits until the log shows which external agent called what.
What to watch
- Whether Salesforce names the select US customers or publishes pricing before the Q4 migration window opens.
- A second hyperscaler implementing the headless architecture would show how much of this interop comes from MCP and how much from the Google deal.
- Whether Salesforce ships admin controls for scoping which of the thousands of capabilities an external agent may call.