AMD starts selling computer that can run top AI models locally, can bring down AI cost

AMD has launched the Ryzen AI Halo Developer Platform, a compact system designed to run large AI models locally. The pitch is lower cloud dependence, steadier costs and greater data control for developers and businesses.

In Short

• AMD's new AI PC runs massive AI models locally
• The system promises better privacy and lower cloud dependence
• Large AI workloads can now run on a compact desktop machine

Artificial intelligence is entering a new phase. Its impact is no longer limited to tech companies or office jobs; it is now reshaping the devices people use every day, from smartphones to laptops. As AI models become larger and more capable, companies are racing to develop chips that can run them more efficiently and at a lower cost. AMD is the latest to join that race with its new Ryzen AI Halo Developer Platform. It is a compact AI computer that has the ability to run powerful AI models locally, without relying on the cloud.

Ever since AMD unveiled the system in late May, it has attracted attention for one particular claim. The machine can run AI models with up to 200 billion parameters on a single system. That is a big deal because models of this size typically require expensive cloud infrastructure or rented GPUs housed in data centres. AMD says its new platform changes that equation.

A powerful AI machine that fits in a backpack

At the heart of the system is the AMD Ryzen AI Max+ 395 processor. According to AMD, it is the most powerful x86 chip in which the CPU and GPU share the same memory pool. The machine comes with up to 128GB of unified memory, allowing developers to work with large AI models locally. AMD says this gives developers enough headroom to build, test and run agentic AI applications on a single machine.

The amount of memory is what many people are talking about. For comparison, high-end graphics cards such as the RTX 5090 come with 32GB of video memory, while the RTX 4090 offers 24GB. AMD's system provides more than three times that amount in a compact chassis that can fit inside a backpack.

No cloud, no rented GPUs

For years, developers wanting to run large AI models have had to rely on cloud providers and pay recurring fees for computing resources. AMD is pitching the Ryzen AI Halo as an alternative.

The company says the machine can take developers from Linux prototyping and fine-tuning all the way to Windows deployment. It supports popular AI frameworks and tools such as PyTorch, vLLM, llama.cpp, Ollama, ComfyUI and LM Studio, while also being optimised for AMD's ROCm software stack.

The message is simple: no data centre, no cloud and no rented GPU.

Could it save money?

The machine is also sparking discussions about the growing cost of AI subscriptions. Today, developers often pay for multiple AI services, including coding assistants, chatbots and cloud-based AI tools. There is also a growing debate over the rising cost of AI for businesses as many AI companies shift towards usage-based billing, making expenses less predictable as usage increases.

AMD's argument is that if large AI models can run locally, users may not need to rely as heavily on cloud services. The 128GB version of the machine starts at around $3,999, and some developers argue that the upfront cost could eventually offset years of subscription expenses.

More importantly, local AI means users retain greater control over their data. Sensitive files do not have to leave the device, and developers no longer need to worry about usage limits or token counts every time they interact with an AI model.

Why this could matter

The biggest promise of the Ryzen AI Halo Developer Platform may not be raw performance alone. If AI models become powerful enough to run on personal computers, it could fundamentally change how people use artificial intelligence. Lawyers could process sensitive documents without sending them to the cloud. Developers could build applications without worrying about rising cloud bills. Startups could experiment with new ideas without paying for expensive GPU rentals.

In other words, AI could become more private, more affordable and more accessible.

- Ends