PCMag editors select and review products independently. If you buy through affiliate links, we may earn commissions, which help support our testing.

Preorders for AMD's $3,999 Ryzen AI Halo, Its DGX Spark Competitor, Start in June

AMD is now ready for its developer platform to go head-to-head with Nvidia (and Apple).

 & Jon Martindale Contributor

Our team tests, rates, and reviews more than 1,500 products each year to help you make better buying decisions and get more from technology.

Our Expert
LOOK INSIDE PC LABS HOW WE TEST
65 EXPERTS
43 YEARS
41,500+ REVIEWS
(Credit: AMD)

AMD is looking to undercut Nvidia's flagship semi-professional AI developer hardware with its Ryzen AI Halo Developer Platform.

After previewing its Ryzen AI 400 series at CES 2026, AMD today unveiled the Ryzen AI Max PRO 400 Series processors for agent computers, and announced that the AI Halo will be available for preorder in June, starting at $3,999, or $700 less than Nvidia's DGX Spark. With 128GB of LPDDR5x, a 50 TOPS NPU, a 40-core RDNA 3.5 GPU, and 2TB of storage, AMD argues it can compete on performance, too.

"For agentic AI to move beyond the cloud, the PC needs to handle real-time tasks locally while responding quickly, protecting sensitive data, and supporting the memory demands of complex agent workflows," AMD says. "Our Ryzen AI Halo developer platform delivers these capabilities in an AMD-validated system, while AMD Ryzen AI Max PRO 400 Series processors bring the same architecture to commercial AI PCs and OEM systems ready for enterprise deployment."

Mac minis are sold out just about everywhere because everyone's buying them up for AI development—and reselling, of course. But while AMD compares itself to that iconic miniature system in its slides for the new developer platform, it's mostly gunning for Nvidia's DGX Spark with this release. As WCCFTech notes, AMD claims it can deliver up to 14% faster local AI performance with specific models, thanks to its powerful AMD AI Ryzen Max+ 395 CPU with 16 cores. It supports Windows and Linux, whereas the DGX Spark is Linux-only.

The hardware supports running local AI models with up to 200 billion parameters. There's Wi-Fi 7, 10Gbps Ethernet, and a TDP of just 120W. All that is packed inside a system that's just 6 inches square and less than 2 inches tall.

(Credit: AMD)

Despite framing its new hardware against Nvidia and Apple, AMD also clearly sees cloud AI computing as a major competitor. Indeed, online chatbots and AI tools are the most popular way professionals interact with AI now, so why run it locally when the up-front costs are in the thousands?

AMD estimates that customers using about 6 million tokens per day will incur cloud costs exceeding $770 per month. After three years, that's over $27,000. In comparison, AMD's Ryzen AI Halo is priced at $4,000 and costs $16 a month to run using typical energy costs. So, you're effectively saving money within six months, by AMD's calculations.

(Credit: AMD)

That won't be an equation that makes sense for every business, nor one that's even accurate when local power costs and specific work demands are taken into consideration, but it's a compelling argument. As models become increasingly expensive and AI developers hunt for profit, running local AI is likely to be far more attractive. If AMD can get its hardware into businesses while they're building the foundations of their AI endeavors, it stands a chance of staying there for several hardware generations.

(Credit: AMD)

Up next are the Zen 5 Ryzen AI Max PRO 400 Series processors. They combine AMD RDNA 3.5 graphics with an AMD XDNA 2 NPU, 192GB of system memory, and 160GB of VRAM. AMD says they're "designed for AI developers, engineers, and creators working across simulation, content creation, and data-intensive workflows." Expect them in devices from partners like HP and Lenovo in the third quarter.

About Our Expert

Jon Martindale

Jon Martindale

Contributor

Jon Martindale is a tech journalist from the UK, with 20 years of experience covering all manner of PC components and associated gadgets. He's written for a range of publications, including ExtremeTech, Digital Trends, Forbes, U.S. News & World Report, and Lifewire, among others. When not writing, he's a big board gamer and reader, with a particular habit of speed-reading through long manga sagas. 

Jon covers the latest PC components, as well as how-to guides on everything from how to take a screenshot to how to set up your cryptocurrency wallet. He particularly enjoys the battles between the top tech giants in CPUs and GPUs, and tries his best not to take sides.

Jon's gaming PC is built around the iconic 7950X3D CPU, with a 7900XTX backing it up. That's all the power he needs to play lightweight indie and casual games, as well as more demanding sim titles like Kerbal Space Program. He uses a pair of Jabra Active 8 earbuds and a SteelSeries Arctis Pro wireless headset, and types all day on a Logitech G915 mechanical keyboard.

Read full bio