Jetson AGX Thor Price Guide 2026 — Dev Kit, T5000 Module & Total Platform Cost

Published: August 31, 2026 | Category: Buying Guide | QSCompute

The Jetson AGX Thor developer kit — NVIDIA's first Blackwell-generation physical AI platform — has been shipping since late 2025, but its price structure is still confusing buyers. Between the $3,499 developer kit, the volume-priced T5000 production module, the cost-optimized T4000, and a DGX Spark that NVIDIA quietly raised to $4,699 in February 2026, procurement teams need a clear breakdown. This guide gives you 2026 street pricing, module-level costs, and the total-platform math for moving Thor into production.

What each Thor SKU actually costs in 2026

NVIDIA sells Thor in three forms. The developer kit is the only one available at retail; the modules are quoted at volume through channels like QSCompute. Here is where pricing stands in Q3 2026:

SKUContents2026 PriceAvailability
Jetson AGX Thor Dev KitT5000 module, carrier board, 1 TB NVMe, PSU, heatsink$3,499Shipping; lead times stretching into Q4 2026
Jetson T5000 moduleProduction module (same silicon as dev kit)~$2,999 at 1,000-unit volumeQuote-based, allocation-driven
Jetson T4000 moduleLower-power Blackwell module for embedded buildsWell under T5000 (quote-based)Quote-based
Jetson AGX Orin Dev Kit (64 GB)Previous generation, still in production$1,999Readily available
NVIDIA DGX Spark (GB10)Desktop Grace-Blackwell system, 128 GB$4,699 (raised from $3,999 Feb 2026)Memory-supply constrained

The February 2026 DGX Spark price increase is the clearest signal of the market: DRAM and memory supply constraints are pushing NVIDIA pricing up across the Blackwell lineup. Thor kits have held at $3,499 so far, but module lead times are stretching into Q4 2026 — lock quotes early.

Dev kit vs production module: the real cost split

For evaluation and algorithm development, the $3,499 dev kit is the whole budget. For production, the module price is only the starting point. The T5000 module bundles the Blackwell GPU (2,070 FP4 TFLOPS), a 14-core Arm Neoverse V3AE CPU, and 128 GB of LPDDR5x — memory is on-module, so there is no separate DRAM line item. What you still need to budget around the module:

ComponentTypical 2026 costNotes
Carrier board (custom or off-the-shelf)$300 – $900I/O planning is critical — the dev kit has no PCIe slot
Storage$80 – $2001 TB industrial NVMe for production image
Power supply$60 – $15040–130 W configurable; sized for sustained 130 W
Enclosure + thermal$150 – $500Fanless designs must handle 130 W Blackwell thermals
Software/NVIDIA licensing$0 – $1,000+/yrISAAC GR00T, Holoscan, Metropolis per-seat costs

A realistic production BOM per unit lands between $3,500 and $4,800 excluding carrier-board NRE — roughly 1.8–2.4× the T5000 module price alone. That is still dramatically cheaper than a DGX Spark per deployed node when you need 40–130 W embedded power, and it is the only one of the three Blackwell systems designed for robotics and in-vehicle deployment.

Thor vs Orin vs DGX Spark: when the premium is worth it

Thor costs 1.75× an AGX Orin dev kit but delivers 7.5× the AI compute at 3.5× better energy efficiency. The decision matrix is workload-driven:

WorkloadBest buyWhy
Humanoid / VLA robot policies, 12+ camera perceptionAGX Thor2,070 FP4 TFLOPS and 128 GB run 70B-class quantized models on-device
AMR, AGV, drone compute at volumeOrin NX / Orin AGXProven supply chain, 10× lower module cost, 15–60 W
Desktop development, 1 PFLOP-class local LLM workDGX SparkGB10 at $4,699 — but it is a desktop box, not embeddable
Fixed-function AOI / NVR appliancesOrin NX / RK3588 / HailoThor is overkill for single-purpose inference

Rule of thumb: if the deployment needs a robot's brain (multi-modal perception + policy), Thor's premium pays off. If the workload is a fixed video or inspection pipeline, an Orin NX or Hailo-based kit at a fraction of the price wins on TCO.

Who should buy Thor in 2026

Buy the AGX Thor developer kit now if you are building humanoid or mobile manipulation systems, need 12+ camera real-time perception, or want to ship a Blackwell-class product in 2027 and need to start software bring-up today. Buy the T5000 or T4000 modules at volume through a channel partner to lock allocation before Q4 lead-time pressure. If your roadmap is fixed-function edge inference, skip Thor — Orin NX or Hailo keeps your BOM honest.

Need Thor modules or dev kits with confirmed allocation?

QSCompute sources Jetson AGX Thor developer kits, T5000/T4000 modules and complete carrier-board designs with real lead times — no placeholder quotes.

Contact: +86 137-1464-6179 | info@qscompute.com