AMD is betting that developers won’t want to keep paying for cloud AI compute. At CES, the company introduced its Mac Mini-sized Ryzen AI Halo PC, and now it’s putting a price tag on the plan: preorders start in June at $3,999, paired with Ryzen AI Max 300 CPU
AMD’s pitch is simple, and it comes with a dollar amount: stop renting AI power from the cloud and run it locally.
At CES, the company introduced the Ryzen AI Halo PC, a Mac Mini-sized system built to handle AI work on-device. Today, AMD followed through on the promise with pricing and a roadmap. The Ryzen AI Halo will start at $3,999 with Ryzen AI Max 300 CPUs, and preorders are set to begin in June. AMD also laid out what’s next with a future Ryzen AI Max 400 model powered by new chips.
The price is not meant to feel “consumer-friendly.” AMD is positioning the Halo as a direct alternative to high monthly AI computing fees—complete with break-even math aimed squarely at developers. The company points to a scenario where $773 a month is spent to use 6 million daily AI tokens. In that case, AMD says the Halo could pay itself off within six months. For heavier use. AMD cites a $2. 253 monthly bill for 18 million daily tokens. saying its $4. 000 Radeon R9700 Pro GPU could break even within three months.
AMD also doesn’t hide who it’s trying to take customers from. The Ryzen AI Halo is aimed at competing with NVIDIA’s DGX Spark AI PC, which is priced at $4,699 after launching at $4,000. The comparison is sharp in the details.
NVIDIA’s DGX Spark AI PC can only run Linux. AMD’s Ryzen AI Halo can run either Windows or Linux, because it’s powered by an x64 chip. AMD also argues its hardware mix gives developers more options: the Halo includes a 50 TOPS NPU plus a Radeon GPU with 40 compute units. while the DGX Spark leans entirely on NVIDIA’s Blackwell GPU for AI work. Both systems come with 128GB of unified system memory. which AMD calls essential for running large models—and which it also notes is more memory than what’s available in a Mac Mini or Mac Studio. both of which are popular with AI developers.
Under the hood, AMD’s Ryzen AI Max 400 chips are intended to extend that local-AI pitch even further. AMD says the Ryzen AI Max 400 lineup will be led by the AI Max+ Pro 495: a 16-core chip with a 5.2GHz boost speed. a 55 TOPS NPU. and Radeon 8065S graphics. These chips are rated to support up to 192GB of unified memory. with AMD saying that allows for 160GB of GPU VRAM. AMD also positions the AI Max 400 as an incremental step. saying the AI Max 400 family is only slightly faster than the AI Max 395. which has a 5GHz CPU boost clock. AMD hasn’t provided comparison benchmarks yet.
The company says Ryzen AI Max 400 chips will be available in the third quarter of 2026.
There’s an underlying bet running through AMD’s announcements: if you can amortize the cost of a local AI box quickly enough. the cloud stops looking like the default—and the developer’s spreadsheet starts looking like a sales pitch. AMD’s Halo pricing and projected break-even timelines are built to make that shift feel practical, not aspirational.
AMD Ryzen AI Halo Ryzen AI Max 300 Ryzen AI Max 400 AI Max+ Pro 495 Radeon R9700 Pro DGX Spark local AI NPU unified memory preorders in June CES 2026
So basically a $4k computer to not use the cloud… cool I guess.
I don’t get it, $773 a month for tokens?? Like are they charging by word now? If it pays off in 6 months why isn’t everyone buying this already.
Windows OR Linux doesn’t matter if developers still need the same stuff behind it. Also the math sounds kinda made up—“break even in 3 months” with 18 million tokens like ok sure. I feel like GPU prices and power costs gonna ruin that.
NVIDIA can only run Linux? That’s wild, AMD wins for sure then. But $3,999 is not exactly “local AI for everyone” lol. I saw something about Radeon R9700 Pro and now I’m thinking of those old Radeon drivers that never worked right… does anyone know if this thing actually runs stable or it’s just for developers who already live in Linux terminals?