NVIDIA announced two new modules built on the NVIDIA Thor architecture, the T3000 and T2000, which the company said "enable mass-market robotics and edge AI applications at scale" (Source: NVIDIA blog, 2026-07 · 原文支持).
According to NVIDIA's blog, the Jetson T3000 combines an NVIDIA Blackwell GPU, an eight-core Neoverse Arm CPU, 32GB of LPDDR5X memory, 273GB/s of memory bandwidth, and 25 GbE connectivity. The Jetson T2000, by contrast, is described by NVIDIA as bringing the Thor architecture to "a broader range of edge AI systems," offering 400 FP4 teraflops of compute and 16GB of memory as an entry point for developers building visual AI agents, autonomous mobile robots, and industrial manipulators.
| Module | Compute | Memory | Connectivity | CPU/GPU |
|---|---|---|---|---|
| T3000 | up to 865 FP4 TFLOPS | 32GB LPDDR5X | 273GB/s bandwidth, 25 GbE | Blackwell GPU + 8-core Neoverse Arm CPU |
| T2000 | 400 FP4 TFLOPS | 16GB | — | Entry-tier Thor architecture |
NVIDIA said the Jetson and IGX T3000 modules deliver 865 FP4 teraflops of AI compute in a form factor the company described as "roughly half the size and power of the T5000" (Source: NVIDIA blog, 2026-07 · 原文支持).
With the T3000 and T2000 added to the lineup, NVIDIA stated its edge AI platform now spans performance "from 70 TOPS to 2,000 teraflops," which the company said allows developers to address a wide range of edge AI workloads.
NVIDIA's announcement of the Thor-architecture modules frames a broader platform push: the company said the new modules extend Thor-based compute to mass-market robotics and edge AI applications. Within that platform, NVIDIA named seven companies it described as "leading" — 1X, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi, and Techman Robot — as building on the Jetson AGX Thor platform (Source: NVIDIA blog, 2026-07 · 原文支持).
NVIDIA's blog cited three examples of developers reducing memory footprints through software optimization rather than hardware changes. Humanoid robotics developers UBTech and Agile Robots, along with industrial solutions provider Connect Tech, reduced memory usage by up to 15GB, which NVIDIA said let them move from the Jetson AGX Orin 64GB module to the 32GB module. In smart retail, SandStar reduced memory usage by up to 4GB, enabling deployment on the Jetson Orin NX 8GB module instead of the 16GB configuration. In intelligent transportation, NoTraffic reduced memory usage by 30% on the Jetson TX2 NX, which NVIDIA said created headroom to add more AI capabilities without increasing hardware requirements.
| Company | Application | Memory Reduction | Hardware Outcome |
|---|---|---|---|
| UBTech, Agile Robots, Connect Tech | Humanoid / industrial robotics | up to 15GB | Orin AGX 64GB → 32GB module |
| SandStar | Smart retail | up to 4GB | Orin NX 16GB → 8GB module |
| NoTraffic | Intelligent transportation | 30% | More AI headroom on TX2 NX |
NVIDIA said it expanded its Cosmos 3 frontier open world foundation model family with a lightweight model compatible with NVIDIA Thor platforms. Cosmos 3 Edge is a 4-billion-parameter model that, according to NVIDIA, helps embodied systems "see the world, reason over it in real time, and predict and generate actions through on-device inference" (Source: NVIDIA blog, 2026-07 · 原文支持).
NVIDIA said developers using the open Cosmos framework can post-train Cosmos 3 Edge for specific embodiments and sensors in about a day, then deploy it on Jetson Thor for real-time vision analysis and on-device robot policy.
NVIDIA said developers can begin using T3000 emulation mode later this month with JetPack 7.2.1, while support for T2000 emulation mode will follow in a future release. NVIDIA also said the Jetson T3000 and T2000 modules themselves are scheduled to become available in Q1 2027.
| Milestone | Timing |
|---|---|
| T3000 emulation mode (JetPack 7.2.1) | Later this month (2026-07) |
| T2000 emulation mode | Future release |
| T3000 and T2000 module availability | Q1 2027 |
The timeline NVIDIA laid out shows a gap between software and hardware readiness: T3000 emulation via JetPack 7.2.1 arrives within the announcement month, T2000 emulation is deferred to an unspecified future release, and the physical T3000 and T2000 modules themselves are not scheduled until Q1 2027. At the same time, NVIDIA's memory-optimization case studies — UBTech and Agile Robots cutting up to 15GB, SandStar cutting up to 4GB, and NoTraffic cutting 30% — show developers already reducing footprints on existing Orin and TX2 NX hardware before the new Thor modules ship. Positioned against NVIDIA's stated platform range of 70 TOPS to 2,000 teraflops, the T2000 (400 FP4 teraflops) and T3000 (up to 865 FP4 teraflops, about half the size and power of the T5000) fill the mid-to-upper tier of that range, while Cosmos 3 Edge's 4-billion-parameter, roughly one-day post-training path is the software component NVIDIA points to for turning that hardware headroom into deployable robot policies.
NVIDIA said the Jetson T3000 and T2000 modules are scheduled to become available in Q1 2027.
According to NVIDIA, developers can begin using T3000 emulation mode later this month with JetPack 7.2.1; T2000 emulation mode support will follow in a future release.
NVIDIA said its edge AI platform now spans performance from 70 TOPS to 2,000 teraflops following the introduction of the T3000 and T2000 modules.
According to Intel's newsroom announcement, the company is investing €5 billion ($5.7 billion) to expand its Leixlip campus in Ireland, producing Intel Xeon 6 and next-generation Xeon processors on the Intel 3 node. Intel says the capital programme began earlier this year and builds on more than €30 billion invested in Ireland since 1989.
According to TechCrunch and VentureBeat, Meta released Muse Code — a beta terminal-based AI coding agent for large repositories — alongside the Muse Spark 1.2 model on August 5, 2026. Muse Code installs with a single command, runs persistent background agents, and is entirely proprietary, putting Meta in direct competition with Anthropic's Claude Code and OpenAI's Codex.
NVIDIA has moved Alpamayo 2 Super into commercial use, according to NVIDIA's official blog. The autonomous-driving reasoning model runs on Cosmos 3 Super Reasoner, holds three times the parameters of its 10-billion-parameter predecessors, and ships under the OpenMDW-1.1 license, which ITHome reports NVIDIA has extended across the entire Alpamayo model family for fine-tuning and commercial redistribution.