AI Hardware Price Dashboard
Illustrative landed-cost dashboard (sales tax ~10%): specs, tables, and ~1-month price trends.
AI Hardware Price Tracker
Report 2026-10-16 PT Illustrative landed cost (sales tax ~10%) Single-file build — charts inlined as base64.
Summary cards
Best observed landed rows (sales tax ~10%) for Strix / Spark / Mac Ultra from today's price table; 5090 street from history anchors (MSRP is not street).
Best Strix Halo landed
$3,463.90
Framework Desktop (Ryzen AI Max+ 395 / Strix Halo)
Framework · list $3,149.00
in stock Lowest landed Strix row in today's table
Best Spark landed
$5,829.99
Gigabyte AI TOP ATOM (DGX Spark platform)
B&H · list $5,299.99
OOS FE/OEM with ConnectX-7 — illustrative landed (sales tax ~10%)
Best Mac Ultra 256GB landed
$10,448.90
Mac Studio M5 Ultra 256GB
B&H · list $9,499.00
preorder Single-box 256GB capacity class (prefer 256 over 96)
5090 MSRP landed (OOS)
$8,849.49
GeForce RTX 5090 32GB
Best Buy · list $8,044.99
OOS MSRP fiction on street — see history
Capabilities & clustering
Honest clustering story: only DGX Spark / GB10 OEMs ship official ConnectX-7 fabric. Strix is single-node (Ethernet demos only). Mac Ultra scales via TB5+MLX. 5090 has no NVLink / no VRAM pool.
Family matrix
| Family | Memory model | Typical power | Clustering | Software stack | Stand-out differentiator |
|---|---|---|---|---|---|
| DGX Spark / GB10 OEMs | 128GB coherent LPDDR5x unified (~273 GB/s) | ~240W adapter; SoC ~140W | YES — ConnectX-7 (2× QSFP ≤200 Gb/s): up to 3 direct / 4 via switch (NVIDIA Sync) | DGX OS, CUDA, NIM / AI Enterprise | Only class here with official multi-node fabric for capacity beyond one 128GB box |
| Strix Halo mini PCs | Up to 128GB LPDDR5x-8000 (~256 GB/s); ~96GB GPU-addressable on Linux | ~120–140W class (MS-S1 Max 320W PSU) | No / limited — no ConnectX; Minisforum markets Ethernet multi-box demos only | Windows/Linux; ROCm maturing; llama.cpp/Vulkan common | Best $/GB for single-node 70B–120B MoE; x86 + expandability (esp. Framework/MS-S1) |
| Mac Studio M5 | Unified 36–256GB+; Max ~460–614 GB/s; Ultra 1.2 TB/s | System up to ~480W continuous | No ConnectX; optional TB5 RDMA + MLX multi-Mac (Apple ~3× with 4 systems) | macOS, MLX, Core ML | Ultra 256GB is the capacity king without buying two Spark nodes; quiet appliance |
| RTX 5090 32GB | 32GB GDDR7 @ ~1.79 TB/s | 575W TGP | No NVLink; multi-GPU = PCIe only, no VRAM pool | CUDA 12.x / Windows-Linux gaming+AI stack | Highest tokens/s when the model fits in 32GB; street >> $1,999 MSRP |
| RTX 5090 AI Box / eGPU | 32GB GDDR7 desktop 5090 in TB5 enclosure | Integrated PSU (~850W class) + WATERFORCE AIO | No ConnectX/NVLink; GPU Selector ≠ pooled VRAM | CUDA / GeForce on Windows host | Plug-and-play external 5090 for TB5 hosts; bandwidth << PCIe x16 |
Clustering truth table
| Platform | Cluster fabric? | What scales? | Confidence |
|---|---|---|---|
| NVIDIA DGX Spark FE | ConnectX-7 QSFP | Model capacity across 2–4 nodes (e.g. dual-unit ~405B *capacity claim*); not a shared memory pool | High (NVIDIA docs) |
| PNY / ASUS GX10 / Gigabyte ATOM / Lenovo PGX / Dell Pro Max | ConnectX-7 retained on all researched OEMs | Same GB10 clustering model; OEM chassis/storage/support differ | High for ASUS/Dell/Lenovo docs; High-Medium for Gigabyte listings |
| Strix Halo (Framework/GMKtec/Beelink) | No | Single-node only | High |
| Minisforum MS-S1 Max | Ethernet demos only | Vendor dual/quad *application-level* demos — not ConnectX | Medium (vendor marketing) |
| Mac Studio | TB5 RDMA + MLX | Distributed inference/training across Macs | High (Apple WWDC/newsroom) |
| RTX 5090 | No NVLink | Parallel independent jobs; not pooled 64GB | High (NVIDIA GeForce specs) |
| RTX 5090 AI Box / eGPU | No ConnectX/NVLink | Single external 32GB CUDA node; GPU Selector assigns apps only | High (Gigabyte docs) |
Price table — list + illustrative landed (sales tax ~10%)
Best observed retail/MSRP row per product. Shipping $0 when free; otherwise landed = price + tax only and flagged.
| Product | Retailer | Channel | List | Ship | Tax (~10%) | Landed | Stock | Config | Link |
|---|---|---|---|---|---|---|---|---|---|
| Mac Studio M5 Max (base) | B&H | retail | $2,499.00 | unk | $249.90 | $2,748.90 | in stock | 18c/32g/36GB/512GB MHL64LL/A | Open |
| Mac Studio M5 Max (other configs) | B&H | retail | $2,799.00 | unk | $279.90 | $3,078.90 | in stock | M5 Max 36GB/1TB APMSM5M01 | Open |
| Framework Desktop (Ryzen AI Max+ 395 / Strix Halo) | Framework | retail | $3,149.00 | unk | $314.90 | $3,463.90 | in stock | Mainboard only Max+ 395 128GB | Open |
| GMKtec EVO-X2 (Ryzen AI Max+ 395 / Strix Halo) | GMKtec official | retail | $3,499.99 | $0.00 | $350.00 | $3,849.99 | in stock | 128GB + 1TB SSD / US | Open |
| Minisforum MS-S1 Max (Ryzen AI Max+ 395 / Strix Halo) | Minisforum | retail | $3,879.00 | $0.00 | $387.90 | $4,266.90 | in stock | 128GB + 2TB Max AI Compute Edition | Open |
| Beelink GTR9 Pro (Ryzen AI Max+ 395 / Strix Halo) | Beelink official | retail | $4,349.00 | unk | $434.90 | $4,783.90 | in stock | 128GB + 2TB Frost Silver / US | Open |
| Gigabyte AORUS RTX 5090 AI BOX 32GB | B&H | retail | $5,099.99 | unk | $510.00 | $5,609.99 | OOS | GV-N5090IXEB-32GD B&H GIAR5090AI32 | Open |
| Gigabyte AI TOP ATOM (DGX Spark platform) | B&H | retail | $5,299.99 | unk | $530.00 | $5,829.99 | OOS | ATAGB10-9001 128GB/4TB PCIe 4.0 | Open |
| Mac Studio M5 Max 128GB | B&H | retail | $5,399.00 | unk | $539.90 | $5,938.90 | preorder | 18c/40g/128GB/1TB APMSM5M16 | Open |
| Mac Studio M5 Ultra 96GB | B&H | retail | $5,499.00 | unk | $549.90 | $6,048.90 | preorder | 30c/64g/96GB/1TB MHL74LL/A | Open |
| Lenovo ThinkStation PGX (DGX Spark platform) | B&H | retail | $5,859.00 | unk | $585.90 | $6,444.90 | OOS | 30KL0002US 128GB/1TB | Open |
| ASUS Ascent GX10 (DGX Spark platform) | B&H | retail | $5,999.00 | unk | $599.90 | $6,598.90 | OOS | GX10-GG0010BN 128GB/1TB | Open |
| Dell Pro Max FCM1253 (DGX Spark platform) | B&H | retail | $6,286.58 | unk | $628.66 | $6,915.24 | OOS | GNFF4 / FCM1253 128GB/2TB | Open |
| NVIDIA DGX Spark (Founders / NVIDIA-branded) | NVIDIA Marketplace | msrp | $6,950.00 | unk | $695.00 | $7,645.00 | unknown | Founders Edition GB10 128GB/4TB | Open |
| HP ZGX Nano G1n (GB10) | B&H | retail | $7,699.00 | unk | $769.90 | $8,468.90 | OOS | HP ZGX Nano G1n 4TB B&H HECZ2V8UT | Open |
| GeForce RTX 5090 32GB | Best Buy | retail | $8,044.99 | unk | $804.50 | $8,849.49 | OOS | ASUS ROG Astral 32GB (BB SKU 13011710) | Open |
| RTX 5090 complete PC (prebuilt) | Newegg | retail | $9,299.99 | $0.00 | $930.00 | $10,229.99 | in stock | Skytech Blaze 5 R7 9800X3D / RTX 5090 / 32GB / 2 | Open |
| Mac Studio M5 Ultra 256GB | B&H | retail | $9,499.00 | unk | $949.90 | $10,448.90 | preorder | 30c/64g/256GB/1TB APMSM5U05 | Open |
| Mac Studio M5 Ultra 256GB 80-GPU | B&H | retail | $10,799.00 | unk | $1,079.90 | $11,878.90 | preorder | 36c/80g/256GB/1TB APMSM5U15 | Open |
| PNY NVIDIA DGX Spark (2-pack bundle) | Newegg | retail | $11,149.99 | $0.01 | $1,115.00 | $12,265.00 | OOS | PNY NVDGXSPARK2PK-KIT (prior single redirects) | Open |
Inventory snapshot
Inventory snapshot (best observed retail/MSRP row per product): 7 in stock, 4 preorder, 8 OOS. Most retailers expose status only (no shelf count).
7
In stock
0
Limited
4
Preorder
8
OOS
dgx_spark: 6 oos · mac_studio: 2 in stock, 4 preorder · rtx_5090: 1 oos · rtx_5090_egpu: 1 oos · rtx_5090_system: 1 in stock · strix_halo: 4 in stock
Numeric shelf/network qty shown in Stock when a retailer exposes it (0 product row(s)); otherwise status only. No invented counts. Public copy omits city/.
- GeForce RTX 5090 32GB @ Micro Center: OOS
- Gigabyte AORUS RTX 5090 AI BOX 32GB @ Micro Center: OOS
Trends
One-month time series by family, plus spotlight multi-product chart. Sparse SKUs stay sparse.
History tables
Anchors from price_history_import unless noted. Expand a product for dated rows; OEM Spark variants without public trackers are omitted (not fabricated).
Generated from 2026-10-16-sample-trends.md report + charts · Estimated sales tax (~10%) · Do not invent prices beyond DB snapshots · 2026-10-16 PT
