NVIDIA Unveils Jetson Thor-Based T3000 and T2000, Stepping Up to Expand into Robotics and Edge AI Markets
Published2026.07.20 11:51
NVIDIA has unveiled its next-generation Jetson platform targeting the robotics and edge AI markets. The company plans to support the development and deployment of humanoid robots, autonomous machines, and visual AI systems through new Jetson Thor-based modules and development tools.
NVIDIA announced on the 16th that it has unveiled new modules based on the Thor architecture, the 'Jetson T3000' and 'Jetson T2000'. Both products are designed for the large-scale deployment of robotics and edge AI applications.
NVIDIA explained that as humanoid robots and autonomous machines have recently expanded into actual industrial sites, there is a growing demand for high-performance, low-power computing platforms capable of running Foundation models in edge environments. In response to this, NVIDIA has introduced a new family of Thor-based products.
The Jetson T3000 is equipped with an NVIDIA Blackwell GPU, an 8-core Arm CPU, and 32GB of LPDDR5X memory. AI computing performance supports up to 865 FP4 teraflops and also provides 25GbE connectivity.
According to NVIDIA, the T3000 supports multimodal AI inference tasks, such as Large Language Models (LLM), Vision Language Models (VLM), and Vision Language Action (VLA) models, even in smaller form factors and low power consumption environments.
The Jetson T2000, unveiled alongside it, offers 400 FP4 teraflops of performance and 16GB of memory. It is targeted at various edge AI applications, such as visual AI agents, autonomous mobile robots (AMRs), and industrial robots.
NVIDIA also unveiled 'Jetson Agent Skills' to enhance development productivity. This feature was introduced as a development support tool that automates memory optimization, system configuration, and deployment tasks.
In addition, the lightweight model 'Cosmos 3 Edge' has been added to 'Cosmos 3', an open-world Foundation model for embodied AI systems. Developers can use this to optimize AI models for robots and autonomous machines for specific hardware environments.
NVIDIA announced that it plans to support the emulation of T3000 and T2000 performance using the Jetson AGX Thor developer kit.
Developers can use T3000 emulation mode through JetPack 7.2.1, which will be released at the end of this month. Emulation features for the T2000 will be available in the future.
NVIDIA announced that it plans to release the Jetson T3000 and T2000 modules in the first quarter of 2027.
관련뉴스
.jpg)
SKT and NVIDIA Collaborate on AI Cloud… First Launch in Korea in 2027
SK텔레콤이 엔비디아와 AI 전용 클라우드 인프라 구축에 나선다. 양사는 엔비디아 DSX 플랫폼을 기반으로 한국에서 AI 팩토리 운영 모델을 먼저 구축한 뒤, 이를 아시아 시장으로 확대할 계획이다. 이번 협력은 SK하이닉스의 HBM 공급을 중심으로 이어져 온 SK그룹과 엔비디아의 관계가 데이터센터 설계·운영과 AI 클라우드 사업으로 넓어졌다는 점에서 주목된다.
2026.06.08by 배종인 기자
.jpg)
Naver and Nvidia to Build Gigawatt-Class AI Factory
Naver is expanding the AI infrastructure of its Sejong data center 'Gak Sejong' based on NVIDIA's DSX platform and is promoting the long-term construction of a gigawatt-scale AI factory. This collaboration is a move to secure the computational foundation needed for HyperCLOVA X advancement, AI agent services, and physical AI model development. As demand for sovereign AI grows in the global market, the domestic AI infrastructure competition is expected to be reorganized around data center and cloud operation capabilities.
2026.06.08by 배종인 기자
SK Hynix and NVIDIA Jointly Develop Next-Generation Memory for AI Factories
SK하이닉스가 엔비디아와 AI 팩토리용 차세대 메모리 공동개발을 추진한다. 양사는 엔비디아 AI 인프라 로드맵에 맞춰 메모리 개발과 공급 협력을 강화하고, 반도체 설계·제조 과정에 AI 기술을 적용하는 협력도 확대한다. 이번 파트너십은 HBM 중심의 공급 협력을 AI 인프라, 퍼스널 AI, 피지컬 AI 플랫폼용 메모리와 자율 제조 영역으로 넓히는 데 초점이 맞춰졌다.
2026.06.08by 배종인 기자
Jensen Huang’s Bold Move for AI Alliance… Korean Industry Rapidly Emerges as Key Pillar of NVIDIA Ecosystem
젠슨 황 엔비디아 CEO의 방한 일정동안 SK, 네이버, LG, 두산, 삼성전자, 현대차그룹 등 한국 기업들이 메모리, AI 클라우드, 로봇, 전력, 소재를 잇는 엔비디아 생태계의 핵심 파트너로 편입되는 전환점을 만들며, 이번 협력을 실제 수익과 산업 경쟁력으로 증명하는 실행 속도가 어떻게 펼쳐질지 전세계의 관심이 주목되고 있다.
2026.06.09by 배종인 기자
NOTA Presents Two MoE Quantization Research Papers at ICML 2026 Workshop
Nota, an AI model optimization and lightweight technology company, will present two research papers on MoE model quantization at the ICML 2026 AdaptFM workshop. According to Nota, the accepted papers are DREAM-MoE and SRA-MoE, which focus on reducing performance degradation while executing large-scale AI models with less memory and computational resources. Following its achievements in MoE quantization technology at the NVIDIA Nemotron hackathon, the company will unveil related research results at this workshop.
2026.06.11by 명세환 기자
