Published: July 17, 2026 | Category: Buying Guide | QSCompute
NVIDIA's Thor platform — announced at GTC 2025 with a staggering 2,000 TOPS of INT8 compute — is reshaping procurement timelines across the edge AI industry. If you're budgeting for Jetson Orin deployments in H2 2026 or planning 2027 rollouts, you need to understand what Thor changes and what it doesn't. This guide cuts through the speculation with architecture-level comparisons, realistic pricing projections, and a decision framework for when to buy Orin now versus waiting for Thor.
Thor is not a simple Orin successor. It's a unified architecture designed to replace three separate product lines — Jetson (edge AI), DRIVE (autonomous vehicles), and parts of the Grace-Hopper portfolio — with a single Blackwell-derived compute platform.
| Specification | Jetson AGX Orin | NVIDIA Thor (Projected) |
|---|---|---|
| GPU Architecture | Ampere (2048 CUDA + 64 Tensor) | Blackwell (next-gen Tensor Core) |
| INT8 TOPS | 275 TOPS | 2,000 TOPS (~7.3×) |
| CPU Cores | 12× Cortex-A78AE | Grace-derived ARM Neoverse (est. 20–24 cores) |
| Memory | 32/64 GB LPDDR5 (204.8 GB/s) | Est. 64–128 GB LPDDR6 (400+ GB/s) |
| Process Node | Samsung 8nm | TSMC 4nm (Blackwell silicon) |
| Power Envelope | 15–60W configurable | Est. 45–150W |
| Functional Safety | SIL 2 (IGX variant) | SIL 3 target (ISO 26262 ASIL-D) |
| Connectivity | PCIe Gen4, 10GbE | PCIe Gen5/6, NVLink-C2C, 25GbE TSN |
| First Shipments | Available now | H2 2027 (automotive first), H1 2028 for edge |
The leap from 275 to 2,000 TOPS is transformative — but it comes with a higher power ceiling. Thor modules targeting 45W configurations (fanless-capable) may deliver ~800–1,000 TOPS, with the full 2,000 TOPS reserved for actively-cooled 150W implementations. For the industrial edge, the practical uplift is closer to 3–4×, not 7×.
One of the biggest questions from procurement teams: will Thor make Orin cheaper, or should I budget extra for Thor?
| Module | Q3 2026 Street Price | Thor Equivalent (Est.) | Expected Thor Price |
|---|---|---|---|
| Orin Nano 4GB | $149 | Thor Nano (2028) | $199–$249 |
| Orin Nano 8GB | $249 | Thor Nano Plus (2028) | $349–$399 |
| Orin NX 8GB | $449 | Thor NX (2028) | $599–$699 |
| Orin NX 16GB | $649 | Thor NX Plus (2028) | $849–$949 |
| AGX Orin 32GB | $1,299 | Thor AGX 64GB (H2 2027) | $1,799–$2,199 |
| AGX Orin 64GB | $1,899 | Thor AGX 128GB (H2 2027) | $2,499–$2,999 |
| AGX Orin Industrial | $2,799 | Thor Industrial (2029+) | $3,500+ (late availability) |
Key takeaway: Thor is not a "cheaper Orin." NVIDIA positions each generation up-market. Orin pricing will hold steady through 2027 as the volume workhorse, with Thor commanding a 40–60% premium for the extra compute. If your application runs comfortably on Orin today, buying now locks in known costs. If you absolutely need 2,000 TOPS for multi-model LLM + vision pipelines at the edge, budget for Thor — but expect to wait until mid-2028 for non-automotive volume availability.
Buy Orin now if:
Wait for Thor if:
NVIDIA has confirmed source-level compatibility between Orin and Thor — CUDA code written for JetPack 6 will port to Thor's Blackwell-derived GPU. The hardware interfaces change (new connector pinout, higher power draw, different thermal solution), but the software continuity is strong. QSCompute recommends buying Orin for current projects and planning Thor evaluation units for Q4 2027, with volume deployment in late 2028.
Jetson Orin — All Modules In Stock
Nano 4GB from $149 · AGX Orin 64GB from $1,899 · Industrial variants available. Volume pricing for 10–1,000+ units. Same-day shipping from Hong Kong.
Contact: +86 137-1464-6179 | info@qscompute.com