MINIX has launched the T4000 and T5000 Generative AI Mini Workstations.

These powerful and space-saving solutions are built for professional generative AI, local large language model (LLM) inference, content creation, on-premise enterprise deployment, and lightweight model training.

The desktops are powered by the NVIDIA Jetson AGX Thor series modules with flagship Blackwell architecture. As such, they deliver exceptional on-device AI horsepower in a small desktop form factor.

The build features durable metal and plastic chassis, plus twin turbo intercooler for sustained performance.

The new offerings are engineered for professionals, developers, creators, and IT teams, redefining edge and on-premise AI without bulky server hardware.

At the core of the T4000 and T5000 are NVIDIA’s cutting-edge compute platform:

• T4000: Up to 1200 Sparse FP4 TFLOPs AI performance
• T5000: Up to 2070 Sparse FP4 TFLOPs AI performance
• 1536-2560 Blackwell GPU with fifth-generation Tensor Cores
• Multi-Instance GPU (MIG) for parallel task efficiency
• NVIDIA PVA 3.0 dedicated vision processing engine

The workstations natively support smooth local inference for 7B-70B parameter LLMs. This makes private, low-latency AI accessible for businesses and creators.

In addition, the offerings feature high-core-count Arm processing and large, fast memories of up to 128GB DDR5 on 12-core or 14-core Arm Neoverse-V3AE 64-bit CPU.

Designed for professional workflows, the mini workstations also include enterprise-grade networking and flexible expansion:

• Dual 10GbE ethernet
• Wi-Fi 6E
• Bluetooth 5.3
• 2x HDMI 2.1 TMDS (4K@60Hz)
• 4x USB 3.2 Gen 1 Type-A
• 1x USB 3.2 Gen 2 Type-C
• 24V DC input, up to 200W max power

Ideal use cases for the MINIX T4000 and T5000 include local LLM inference, generative AI creation, on-device AI computing, and lightweight model training.

Lenovo has unveiled new Lenovo Hybrid AI Advantage with NVIDIA solutions designed to accelerate AI adoption, reduce time-to-first-token (TTFT), and deliver measurable business results across personal, enterprise, and cloud environments.

Building on the inferencing acceleration introduced at Lenovo Tech World, this next phase of Hybrid AI execution expands the solutions with device to data center to gigawatt-scale AI cloud deployments.

This enables real-time decision-making, operational efficiency, and intelligent automation across industries at global scale. The solutions boost productivity, agility, and innovation by enabling faster AI deployment.

The development comes as AI is seen moving from training models powering real-time decisions. Lenovo is prepared to address the demand for validated hybrid AI platforms built for production-scale inferencing, as organizations will need infrastructure to support such.

In fact, Lenovo’s Hybrid AI Advantage with NVIDIA are now delivering ROI in less than six months. The new inferencing-optimized ThinkSystem and ThinkEdge servers are being utilized for real-time inferencing across retail, manufacturing, healthcare, sports, and smart city scenarios.

The expanded portfolio includes:

• two Lenovo Hybrid AI platforms, featuring NVIDIA RTX PRO 6000 Blackwell Server Edition and Blackwell Ultra
• Hybrid AI inferencing starter platform with RTX PRO 4500 Blackwell Server Edition
• Lenovo ThinkAgile HX650a with Nutanix Enterprise AI and Nutanix Kubernetes Platform
• Lenovo Hybrid AI platforms with Cloudian

Bringing inferencing directly to professionals

Lenovo and NVIDIA are bringing AI from development environments to real-world production at a global scale. This is thanks to new Lenovo AI inferencing platforms with NVIDIA Dynamo and NVIDIA NIM.

Meanwhile, Lenovo AI Cloud gigafactory platforms are powered by NVIDIA Vera Rubin NVL72. Industry-specific agentic AI solutions are also built with NVIDIA Blueprints and software.

For consumers, there’s next-generation NVIDIA RTX Pro Blackwell-powered mobile and desktop workstations. These will be rolled out across the ThinkPad P14s Gen 7, ThinkPad P16s Gen 5, and ThinkPad P1 Gen 1 lineups.

ThinkStation P5 Gen 2 desktops, meanwhile, will get up to two RTX PRO 6000 Blackwell Max-Q GPUs. They will also have support for NVIDIA OpenShell.

For gigawatt-scale scenarios, the next-gen Vera Rubin platform accelerates deployment for hyperscale and sovereign AI cloud providers.

These fully liquid-cooled, rack-scale AI systems are engineered for faster deployment and dramatically improved token economics. They can achieve up to 10x higher throughput and up to 10x lower cost per token.

CIPTA Industrial Sdn Bhd steps onto the global stage with its European debut at CloudFest 2026. They introduced high-density AI infrastructure and edge-ready systems built for modern enterprise workloads.

Held at Europa-Park in Rust, Germany from March 23 to 26, the event marks the company’s first major international showcase under its own brand. Backed by InWin Development Inc., CIPTA positions itself as a new-generation EMS provider focused on AI, cloud, and enterprise systems.

At Booth R41, the company is highlighting two key platforms: the RG658 PRO GPU server developed with Phison, and the cubePRO edge workstation created in collaboration with Accordance.

Built for scalable AI workloads

Leading the showcase is the RG658 PRO, a high-density GPU server designed to handle large-scale AI training and inference without pushing costs out of reach for enterprises.

The system supports up to eight high-performance GPUs and integrates Phison’s Pascari aiDAPTIV alongside its PASCARI enterprise SSD lineup. This combination aims to improve data throughput, reduce latency, and streamline AI pipelines.

Thermal performance is a key focus. The RG658 PRO uses a dual-chamber design to separate heat zones, paired with up to 14 high-speed PWM fans for sustained cooling under heavy workloads. Power delivery is handled by a 3+1 redundant configuration of 80PLUS Titanium PSUs, scaling up to 9600W.

The result is a platform built to scale AI deployments on-site while maintaining efficiency and reliability.

Edge computing without downtime

Alongside its GPU server, CIPTA is introducing the cubePRO, a compact edge workstation designed for environments where uptime and data integrity are critical.

The system supports up to four PCIe slots for GPU configurations, making it suitable for AI workloads at the edge. It also features high-capacity multi-SSD setups and optimized airflow for continuous 24/7 operation.

Through its partnership with Accordance, the cubePRO integrates the Disk Array ARAID M500 solution, enabling high-availability storage and data protection. This ensures uninterrupted performance for use cases such as industrial systems, remote nodes, and enterprise branch deployments.

The focus here is clear: bring AI processing closer to where data is generated, without sacrificing reliability.

Strengthening Malaysia’s role in AI infrastructure

CIPTA’s debut also reflects a broader shift in global supply chains. Operating from Malaysia, the company offers end-to-end services—from concept to production—along with flexible manufacturing cycles and cost-efficient operations tailored for Southeast Asia and international markets.

With access to InWin’s server chassis ecosystem and infrastructure solutions, CIPTA combines global platform capabilities with localized integration. The goal is to help enterprises deploy AI and cloud infrastructure faster while diversifying their supply chain footprint.

As demand for AI systems continues to grow, CIPTA is positioning Malaysia as a key hub for scalable, production-ready infrastructure.

Visitors can find CIPTA at Booth R41 during CloudFest 2026 in Europa-Park, Rust, Germany.