Hybrid AI Explained: What Your Next Work PC Actually Needs

Buying a laptop used to involve a fairly manageable argument about RAM, battery life and whether you really needed the more expensive processor. Now there is an NPU to consider, an AI label on the box and the suggestion that your next spreadsheet might somehow need all of it. Before adding another acronym to the shopping list, it helps to ask a simpler question: where is the AI actually doing its work?

Hybrid AI combines processing on your device or nearby infrastructure with AI services in the cloud. The idea is to give each task a suitable place to run, rather than send everything to a remote server. That could mean handling a time-sensitive task on a laptop, keeping selected business information on a company-controlled server, and using a cloud service when a larger model or more computing capacity is needed.

We put the practical questions to Caleb Leung, General Manager, Systems and OEM at Ingram Micro, in a written interview. His answers followed the distributor’s August launch of The Great Migration, a program bringing together Microsoft and hardware partners around business device upgrades and AI adoption. The useful part for anyone buying the next work PC is understanding which capabilities will actually earn their place.

What Hybrid AI changes in everyday use

Using an AI website on a laptop does not, by itself, mean the laptop is running the model. Often it is simply displaying the interface while a remote system handles the request. Local AI moves that processing onto the device. Edge AI takes a similar approach closer to the source of the data, although the hardware might be a server at a shop or industrial site rather than the PC on your desk.

Leung expects early opportunities in everyday productivity and workloads that need to respond to a steady stream of information. He points to retail cameras analysing movement, stock and security events, alongside mining and agriculture systems processing equipment and environmental sensor data. Those are examples of where he sees potential, rather than evidence that every Australian business already has the technology in place.

The appeal is straightforward: processing nearby can reduce the need to move every piece of data over a network and can avoid some of the delay involved in a remote request. For a laptop user, a supported local model can also make selected features available without a live internet connection. Microsoft’s Windows AI documentation, for example, says Foundry Local can perform inference offline once the model has been downloaded and cached. That does not make every AI app on the laptop an offline app.

Ultimately, I see cloud, AI PCs and edge infrastructure working together, with workloads distributed according to privacy, governance, latency, cost and performance requirements.

Caleb Leung, Ingram Micro

Keeping data local still requires some homework

The privacy argument becomes more interesting when the material is a customer file, internal document or something a business would rather keep off a public service. Leung sees growing interest in Hybrid AI as organisations move beyond experimenting and start involving sensitive information and intellectual property. A controlled local environment can give them more choice over where that information is processed.

That choice needs to be checked in the actual application. A model might run locally while another part of the software synchronises files or sends information to an external service. Microsoft states that Foundry Local keeps inference inputs and outputs on the machine, but that is a specific implementation, not a promise covering every product described as an AI PC. The label on the lid cannot tell you where an app sends a document.

The same distinction matters when people talk about data sovereignty. Where information is stored, who can access it and which services receive it are separate questions from which chip performs a calculation. Businesses need to follow the whole path of the data. For personal files, our guide to cloud sync and local backups deals with another easily confused distinction: having a file in the cloud and having a dependable backup.

An NPU is useful when the software uses it

An NPU, or neural processing unit, is a processor designed for AI workloads. It sits alongside the CPU that handles general computing and the GPU that handles graphics and can also accelerate AI. In compatible software, assigning suitable work to the NPU can help use the system’s resources more efficiently. Microsoft’s guide to CPU, GPU and NPU roles explains that division of labour.

Leung recommends considering an NPU when refreshing a business PC fleet, along with sufficient RAM and overall performance. He also sees potential for better battery life, heat and fan noise when workloads are distributed appropriately. Those are possible benefits of a suitable hardware and software combination, rather than something we have measured here. A new processor does not automatically fix an inefficient application.

There is also no universal rule that local AI requires a brand-new NPU-equipped machine. Microsoft’s Windows ML supports execution across CPUs, GPUs and NPUs, with the available hardware and model determining what is practical. Some features have narrower requirements. Before paying extra, check the applications you intend to use, their supported hardware and the size of the models involved. A bigger AI performance number is not a substitute for compatibility.

And keep the ordinary laptop questions in the conversation. A good keyboard, useful ports, a suitable screen and enough memory still matter throughout the working day. Our look at choosing a main screen is a reminder that the least fashionable part of a setup can have a very obvious effect on how comfortable it is to use.

Start with the work, then choose the PC

One of the more useful parts of Leung’s advice comes before the hardware specification. He recommends starting with an AI strategy, identifying the applications a business expects to deploy, then working backwards to the capabilities needed. That is sensible advice for a smaller team too. Name the job you want to improve before choosing the machine that is supposed to improve it.

The goal is not to buy AI PCs simply because they are “AI PCs”; it is to ensure the next hardware investment remains capable as AI becomes a normal part of how we work.

Caleb Leung, Ingram Micro

Leung acknowledges the risk of investing without a clear use case, although he argues that preparing early can make sense where the additional cost is modest. There is a fair balance to strike here. If a fleet is already due for replacement, useful AI capability may be worth including in the comparison. If the current machines still meet the job and the relevant software runs in the cloud, the case for replacing them purely for local AI needs more substance.

A practical starting point is one defined workflow with an observable benefit: time saved, less data transferred, or a feature that remains usable away from a reliable connection. The result still needs to be checked, and people need to understand when to intervene. Leung identifies trust, governance, skills and uncertainty about return on investment as barriers to wider adoption. Buying hardware addresses only part of that problem.

Hybrid AI gives businesses more places to run the work. The value comes from choosing those places deliberately. When the next laptop salesman starts with the AI badge, asking which application will use it is a perfectly reasonable way to bring the conversation back to earth.

Zachary Skinner

Zachary Skinner is the editor of TechDrivePlay.com, where tech, cars and adventure share the fast lane.
A former snowboarding pro and programmer, he brings both creative flair and technical know-how to his reviews. From high-performance cars to clever gadgets, he explores how innovation shapes the way we move, connect and live.