Jetson T4000 for Edge AI — Specs, Sizing & Deployment Fit
Last updated: September 2026
The lower-power entry to the Thor generation: roughly 1200 FP4 TFLOPs and 64 GB in a 40–70 W envelope (vs the T5000’s 40–130 W), at ~$2,999 — the same module price as the AGX Orin 64GB.
Planning Takeaway
The lower-power entry to the Thor generation: roughly 1200 FP4 TFLOPs and 64 GB in a 40–70 W envelope (vs the T5000’s 40–130 W), at ~$2,999 — the same module price as the AGX Orin 64GB. New long-horizon designs should weigh its newer architecture against AGX Orin’s far more mature deployment ecosystem.
Specifications
| Spec | Value |
|---|---|
| Compute | ~1200 FP4 TFLOPS sparse (1536-core Blackwell GPU, 64 fifth-gen Tensor Cores; MIG support) |
| CPU | 12-core Arm Neoverse-V3AE, up to 2.6 GHz |
| Memory | 64 GB 256-bit LPDDR5X, ~273 GB/s |
| Power | 40–70 W configurable (vs 40–130 W on the T5000) |
| Video encode (vendor spec) | Up to 6× 4Kp60 or 24× 1080p60 H.265 |
| Video decode (vendor spec) | Up to 4× 8Kp30 or 10× 4Kp60 H.265 |
| Form factor | Jetson AGX module (100×87 mm, 699-pin) |
| Availability | Orderable since CES 2026 |
| Est. planning capacity | Not yet modeled — EdgeAIStack stream benchmarks pending |
| Software | JetPack 7.2 (Ubuntu 24.04, CUDA 13, June 2026) — current; JetPack 6.x (Ubuntu 22.04) — legacy track |
| Indicative price | ~$2,999 (module, 1KU+, as of mid-2026) |
Sizing Fit
Aimed at generative-capable edge nodes that do not need the T5000’s full 130 W envelope: FP4-quantized mid-size LLMs and VLMs, and future-proofed vision nodes on the Thor software line. EdgeAIStack stream benchmarks for this module are not yet published — sizing below leans on the Thor platform guides.
- Thor platform decision framework — Comparison
- Edge LLM runtime stack — Guide
- Which VLM fits which Jetson — Guide
Decision Framework
Choose the Thor T4000 if:
- You want the Thor generation (NVFP4, current software line) without the T5000’s power and price.
- Your roadmap includes FP4-quantized generative models that AGX Orin’s Ampere silicon cannot accelerate.
- You are designing new hardware anyway, so AGX Orin’s carrier/thermal maturity matters less.
Choose a sibling instead if:
- Jetson AGX Orin 64GB — you deploy this year on proven carriers, need the dev kit path, or run classical vision — same ~$2,999.
- Jetson AGX Thor T5000 — you need 70B-class models, 128 GB, or maximum stream density.
Video decode & encode (datasheet)
Hardware codec limits for the Jetson Thor T4000 as NVIDIA publishes them at 70 W. Streams per module at 1080p30 and 4K30, rated decoder throughput, and NVENC streams at 1080p30. Full profile tables, sources and class-D interpolation: decode capacity for the Jetson Thor T4000.
| Codec | Decode 1080p30 | Decode 4K30 | Decode MP/s | Encode 1080p30 |
|---|---|---|---|---|
| H.265 | 39 | 9 | 2460 | 25 |
| H.264 | 40 | 10 | 2520 | 22 |
| AV1 | 24 | 6 | 1560 | — |
| VP9 | 29 | 7 | 1840 | — |
NVIDIA · DS-11945-001 v1.5 §2.2 Multi-Standard Video Decoder, Table 2-3 (T4000), 70W Mode · verified 2026-09-07 · Jetson Thor Series Modules Data Sheet · class A · preliminary per NVIDIA
Frequently Asked Questions
How much does the Jetson T4000 cost?
The T4000 module lists at ~$2,999 (1KU+ pricing) as of the July 2026 NVIDIA repricing — identical to the AGX Orin 64GB module. It has been orderable since CES 2026.
Jetson T4000 vs AGX Orin 64GB at the same price — which one?
T4000 for new long-horizon designs that want Blackwell-generation FP4 compute (~1200 FP4 TFLOPs) in a lower power envelope; AGX Orin 64GB for near-term production on its mature ecosystem of carriers, thermal designs, and field-proven software. Same ~$2,999 module price, different risk profiles.
What are the T4000’s power draw and stream capacity?
The T4000 is configurable between 40 and 70 W (vs 40–130 W on the T5000), with vendor-spec video decode up to 10× 4Kp60 H.265. EdgeAIStack stream-capacity benchmarks for the T4000 are not yet published, so treat stream sizing as directional and validate against vendor documentation.