Amazon Patents a System That Pauses an Unfinished Job and Resumes It on a Different Server

By Patentlyze Team · Updated Jul 24, 2026

What if a cloud server could freeze mid-task, hand its exact memory state to a completely different machine, and pick up right where it left off? That's the core idea in this Amazon patent.

See all 9 drawings from this filing ↓

How Amazon's pause-and-resume cloud workflows actually work

Imagine you're baking a cake and you need to stop halfway through to wait for the butter to soften. Instead of standing in the kitchen doing nothing, you take a detailed photo of everything on the counter, walk away, and when the butter is ready, someone else steps in, looks at the photo, and finishes the cake from that exact point. Amazon's patent describes exactly this kind of handoff, but for cloud computing tasks.

The system runs your task on one cloud server until it hits a waiting point, like needing a result from a database or an outside service. Instead of keeping that server idle and running up the bill, it saves a snapshot of where the task was at, shuts down, and sends back a reference called a handle. When the outside service finishes its work and sends a response, a second server picks up that handle, reloads the saved state, and finishes the job.

For you as a user or developer, the workflow still feels like one continuous process. Behind the scenes, though, Amazon is swapping in fresh servers only when they're actually needed, rather than leaving one server running and waiting the whole time.

How snapshots and handles hand off workflow state between servers

The patent describes a method for continuation workflows on Amazon's serverless computing platform (think AWS Lambda). The key problem it solves: long-running tasks that need to pause and wait for an external operation create waste, because the server holding the task sits idle but still costs money.

Here's the flow the patent lays out:

• A first virtual machine starts executing a workflow until it hits a point where it needs something from an outside entity, like a database query, an AI model call, or a third-party API.
• The system detects this waiting point, saves the full execution state as a snapshot, and generates a handle (a lightweight reference to that snapshot, not the full data itself).
• The first virtual machine is shut down or released. The handle and a resume request are sent back to the external entity.
• When the entity finishes its work and sends a response, it includes the handle. A second virtual machine loads the snapshot, restores the exact state, and continues the workflow from where it stopped.

This can chain across multiple virtual machines for workflows with many waiting steps. The final machine delivers the finished result back to whoever originally requested the workflow.

What this means for AWS serverless computing costs and reliability

The core issue this addresses is cost and efficiency in serverless computing, where customers pay per millisecond of compute time. Right now, a task that has to wait for a slow external call either burns money sitting idle or gets awkwardly re-architected by developers to avoid the problem. This patent describes a way to make the platform handle that complexity automatically, without the developer having to design around it.

For AWS, this could make Lambda-style serverless functions practical for longer, more complex workflows that today require a separate orchestration service like AWS Step Functions. If Amazon ships something like this, you'd get more flexible cloud applications at a lower cost without needing to manage the state-saving logic yourself.

The drawings

9 drawing sheets from US 2026/0211710 A1 · click any drawing to enlarge

SHEET 1 SHEET 2 SHEET 3 SHEET 4 SHEET 5 SHEET 6 SHEET 7 SHEET 8

Browse all the drawings 9 drawing sheets

× ← →

Editorial commentary on a publicly published patent application. Not legal advice.