Google Launches Agent Compute Engine to Slash AI Agents' Idle-Time Costs with MicroVMs

Google Tackles AI Agents' Idle Infrastructure Problem with New Service

Google launched Agent Compute Engine on Thursday, positioning it as a solution to what the company sees as a major barrier to AI agent adoption: idle infrastructure costs1

1

. The new compute service suspends inactive AI agents and restores them on demand while preserving their execution state, addressing a critical challenge as enterprises scale their agent deployments2

2

.

AI agents often spend extended periods waiting for instructions, tool responses, or human approvals. While enterprises can reduce costs by shutting down underlying compute resources, restarting agents typically requires rebuilding the execution environment and restoring context, which adds latency1

1

. Google's infrastructure service optimized for hosting AI agents addresses this trade-off through hardware-isolated microVMs that can be suspended and resumed on demand.

MicroVMs Deliver Security and Efficiency for Cost-Efficient AI Agent Infrastructure

The choice of microVMs is designed to reduce overhead associated with conventional virtual machines, which typically take longer to boot and require more resources to initialize1

1

. Google says the technology combines the hardware efficiency of containers with the security of traditional virtualization2

2

.

Unlike containers that share the host operating system's kernel, microVMs run their own kernels, providing stronger isolation boundaries for agents executing AI-generated code or handling untrusted inputs1

1

. This architecture limits the potential impact of compromised agents. Google refers to Agent Compute Engine's implementation of these environments as sandboxes, which host the applications, data, and user instructions that agents need while containing cybersecurity guardrails that block malicious activity2

2

.

The microVMs running on Agent Compute Engine appliances have limited access to storage resources, preventing hackers from reading or deleting data if they take over one of the sandboxes2

2

. Google verifies appliance firmware integrity using custom Titan cybersecurity chips, while the appliances' CPUs feature mitigations against side-channel attacks that analyze physical state data to glean customer information2

2

.

How Agent Compute Engine Reduces Idle-Time Costs and Cloud Spending

The ability to suspend and resume an agent's computing environment swiftly helps reduce cloud spending, particularly for long-running coding and research agents that hold state for hours or days, according to Amit Kumar Jena, AI development head at IT consulting firm Kanerika1

1

. Traditional cloud compute wasn't built for long-running agents that activate thousands of times and otherwise sit idle, noted Manoj Chandra Jha, principal analyst at Nord-IQ Research1

1

.

As agents move from pilots to fleets of thousands, idle compute costs when using traditional cloud infrastructure can become a major constraint, increasing expenses by a significant margin1

1

. Agent Compute Engine detects when a sandbox goes idle and shuts it down to prevent unnecessary hardware usage, then quickly reactivates it when needed2

2

. This approach lets a single physical node host orders of magnitude more living agent microVMs than conventional VMs2

2

.

Developer Benefits and Checkpointing Capabilities

For developers, the ability to hold agent state while suspending and resuming tasks reduces engineering overhead for maintaining long-running operations1

1

. "It lets developers stop building their own save-and-resume layer. A lot of code that exists only to survive the pause goes away," Jena explained1

1

.

Google claims Agent Compute Engine's checkpointing mechanism goes beyond simply suspending an environment between tasks by preserving memory and execution state, allowing agents to resume from where they left off rather than reconstructing context and restarting execution1

1

. The service couples this with features integral to agents like cloning, snapshot/forking, rewind, and rollback that allow systems to operate efficiently without resource duplication2

2

.

Trade-offs: Vendor Lock-in, Storage Costs, and Maturity Concerns

The new compute service comes with trade-offs, particularly vendor lock-in, operational complexity, and uncertain cost savings. Agent Compute Engine is closely integrated with Google Cloud, potentially making it difficult for enterprises to move agent workloads to another provider or run them across multiple platforms without reworking infrastructure, according to Pareekh Jain, principal analyst at Pareekh Consulting1

1

.

Storage costs may rise because each microVM requires disk space to retain its working state, and at scale, storing thousands of persistent agent environments could add significantly to infrastructure bills, making it harder for finance teams to predict net savings1

1

. Potential cost savings depend on how agents use compute resources, with continuously active, compute-intensive workloads likely seeing smaller gains1

1

.

The service isn't yet in preview, meaning enterprises would be betting on Google's roadmap rather than proven production performance1

1

. Jena noted that AWS has run a microVM per session in AgentCore since 2025 and stops billing CPU while agents wait, while Microsoft's hosted agents in Foundry suspend between turns and keep the filesystem1

1

.

Enhanced Security Through Agent Gateway Integration

Agent Compute Engine stores login credentials outside the sandbox hosting an agent, preventing hackers from stealing sensitive data if they compromise the sandbox2

2

. When an agent needs to log into an application, Agent Gateway retrieves relevant credentials and submits them on the agent's behalf2

2

.

Google will roll out Agent Compute Engine in coming months alongside a new version of Agent Gateway2

2

. The updated gateway gains model routing capability to send each agent request to the model that can process it most efficiently, plus a tool for regulating inference costs2

2

. Token management provides control over token consumption across agents and models, allowing enterprises to attribute spend, enforce budget guardrails, and establish a single source of truth for token metrics2

2

.