AMD says the next generation of AI PCs should run more artificial intelligence directly on the device, reducing dependence on paid cloud services while giving users greater control over sensitive data.
AMD used its opening keynote at IFA 2026 in Berlin to argue that PCs are becoming powerful enough to handle AI tasks that previously depended on remote data centres.
Jack Huynh, AMD’s senior vice president and general manager of Computing and Graphics, said the company’s “Personal AI” strategy is built around three ideas: running more AI locally, keeping personal information under the user’s control and giving AI access to useful context about how someone works.
The pitch comes as AI assistants evolve beyond answering questions towards agents that can plan and carry out multi-step tasks.
AMD believes those workloads will make relying entirely on cloud computing increasingly expensive, particularly for developers and businesses using AI throughout the day.
What Does AMD Mean by Personal AI?
AMD’s Personal AI vision puts the PC at the centre of AI processing rather than treating it mainly as a window into cloud services.
Under that model, a computer would handle suitable tasks locally and reach the cloud only when extra computing power is required. AMD argues that keeping more work on the device could improve privacy because files, conversations and other sensitive information would not automatically need to leave the PC.
For users, the practical difference could be an AI assistant that works with local documents and applications without sending every request to an external server.
AMD is not proposing an entirely offline future.
Its strategy instead combines local and cloud computing, with the PC handling what it can and remote infrastructure providing additional capacity when necessary.
How Powerful Are AMD’s Local AI PCs?
AMD is increasing the amount of memory available to its Ryzen AI Halo systems so they can run significantly larger AI models without relying on a data centre.
AMD’s current Ryzen AI Halo platform offers up to 128GB of shared memory, enough to run some very large AI models locally. A newer version will raise that to 192GB, giving developers more room to run demanding AI tools without immediately turning to cloud servers.
Parameters are one measure of an AI model’s size.
Larger models generally require more memory to run, which is why many advanced systems have traditionally depended on expensive cloud servers.
At IFA, AMD demonstrated large AI models running directly on Ryzen AI Halo hardware and showed a forthcoming HP ZBook workstation using the platform. The company also highlighted Lenovo’s Ryzen AI Halo-powered ThinkCentre X desktop.
The focus is initially strongest on developers, creators and professional users rather than ordinary laptop buyers, but AMD’s wider argument is that increasingly capable local AI will eventually become part of mainstream PCs.
Why Is AMD Challenging the Cloud AI Model?
AMD says the economics of cloud AI become harder to justify as AI agents perform more tasks and generate larger amounts of data.
During the IFA keynote, Huynh presented an example in which processing 15 million output tokens a day through cloud services could cost around €300 daily, approaching €100,000 a year for a heavy user. Those figures were AMD’s own illustration and will vary substantially depending on the AI model, provider and workload.
Running a model locally removes per-token cloud charges, although users or businesses still have to pay for the hardware and electricity needed to operate it.
Microsoft is reinforcing that argument with Project Zenith, a Windows setup designed for high-performance developer PCs with at least 64GB of unified memory. Microsoft says supported systems will be able to run large models locally without metered cloud-token charges, with AMD Ryzen AI Halo powering the first Project Zenith device.
Are AI PCs Really Moving Away From the Cloud?
AMD is not alone in trying to shift more artificial intelligence onto personal computers.
Nvidia is preparing the first Windows PCs using its RTX Spark platform for October. Reuters reported that technology companies are increasingly pursuing on-device AI to reduce cloud-service costs while improving privacy and responsiveness.
That competition suggests the AI PC battle is moving beyond how many AI features manufacturers can advertise. Chipmakers are increasingly competing over how much useful AI work a computer can perform without repeatedly connecting to a remote server.
Cloud services will remain important for the largest models and demanding workloads, but AMD’s IFA message is that they should no longer be the starting point for every AI task.
The next test will be whether developers build useful applications around that local computing power and whether the benefits eventually reach mainstream laptops at prices consumers are willing to pay.
