Announcement
MSI PRO MAX EDGE AI+ 11M: 4-Liter Ryzen AI Max+ 395, Now Taking Orders
MSI's four liter AI desktop is out. Ryzen AI Max+ 395, 128GB of unified memory and up to 96GB of it available to the graphics. Available to order now.

MSI shipped it. The PRO MAX EDGE AI+ 11M is a four liter desktop with a Ryzen AI Max+ 395 inside and 128GB of memory the graphics can actually reach. We are taking orders today.
This one took its time. MSI announced the design back in January, as the AI Edge. It turned up at Computex in June under the name it kept. The launch went official on July 24. The hardware never really changed. What changed is that you can buy one.
Four liters, and no power brick
97.5 x 188.4 x 248.5 mm, and 2.9 kg on the scale. Sandblasted aluminum with diamond cut edges, which is MSI's way of saying it will not embarrass you on a desk. The 300W supply sits inside the case, so the only cable running to it is the one going to the wall. Anyone who has kicked a laptop brick out from under a desk will care about that more than the spec sheet lets on.

The chip
Sixteen cores, 32 threads, 64MB of cache. The graphics are Radeon 8060S, and they carry no memory of their own. There is a separate neural engine on board worth 50 TOPS by itself. MSI quotes 126 TOPS for the whole chip, but that is three processors counted as one. Take it with salt.
The memory is the point
128GB of LPDDR5X, soldered down, and up to 96GB of it can be handed to the graphics. That is the whole reason this thing exists.

A discrete card will give you faster memory. It will not give you 96GB of it at this price, in this volume, on this power budget. Nobody's building that card. Not this year, not at this size.
Does it beat a rack of proper accelerators? Of course not. But what stops most people is capacity, not bandwidth. If you want a 120 billion parameter model running under your own roof, this gets you there. A graphics card does not.
What it actually runs
MSI published its own throughput figures. No test configuration came attached to them, so read them as directional.
Model | Memory used | Avg tokens/sec |
|---|---|---|
Gemma4 E4B | 9.6 GB | 50.51 |
Gemma4 12B | 7.6 GB | 23.24 |
GPT-OSS 20B | 13 GB | 44.36 |
Gemma4 26B | 17 GB | 51.85 |
Qwen 3.6 35B | 23 GB | 58.96 |
GPT-OSS 120B | 65 GB | 33.29 |
The last row is the one to look at. 65GB resident, 33 tokens a second, nothing leaving the building.
Two boxes, or four
Cluster them and the ceiling moves. MSI lists setups running up to 670 billion parameters across four nodes.
Model | Quantization | Cluster size |
|---|---|---|
GPT-OSS 120B | INT8 | 2 nodes |
Qwen3-235B-A22B-Instruct-2507 | Q4KM | 2 nodes |
Qwen3-Next-80B-A3B-Instruct | BF16 | 2 nodes |
Qwen3-235B-A22B-Instruct-2507 | INT8 | 4 nodes |
Qwen3-235B-A22B | UD-Q8_K_XL | 4 nodes |
DeepSeek-V3.1 (671B) | Q4_0 | 4 nodes |
Whether you need DeepSeek V3.1 sitting in your own building is a separate question from whether you can have it. Now you can have it.
Ports and the rest
Two 40Gbps USB-C ports, one front and one back, plus HDMI, DisplayPort and enough Type-A to go around. There is 2.5 gigabit ethernet, and an SD card reader up front. A physical switch steps the machine through three performance modes, no software needed. Four displays if you want them.

It stays quiet under load, which MSI leans on hard in the marketing. Two M.2 slots, Wi-Fi 7, Windows 11 Pro.

Rob's take
“MSI's new PRO MAX Edge has three fans, three heat-pipes, and a massive copper heat-spreader. Nobody else has put that much cooling on this chip. I'm expecting this will be the fastest AMD Strix Halo system you can buy.” |
Availability
The 128GB configuration is available to order now. Pricing is quoted per order, so call or write and we will get you a number the same day. Full spec sheet is on the product page.
sales@gpupartner.com • +1.626.88G.PU00
Sources
Source | What it covers | Link |
|---|---|---|
MSI | Official launch announcement, July 24 2026 | |
MSI | Product page, specs and throughput figures |
- MSI PRO MAX EDGE AI+ 11M
- Ryzen AI Max+ 395
- Strix Halo
- unified memory
- 128GB LPDDR5X
- 4 liter workstation
- local LLM desktop
- edge AI PC
- Radeon 8060S
- 96GB VRAM
- GPT-OSS 120B
- DeepSeek V3.1 cluster
- mini workstation
- on-device AI