Rafay Turns NVIDIA Omniverse DSX Blueprint into One-Click Digital Twin Service

Standing up an AI cloud usually means months of stitching together GPUs, Kubernetes, and AI services from different sources, piece by piece. Mirantis k0rdent AI, validated under NVIDIA's AI Cloud-Ready initiative, collapses that into one sovereign layer, from bare metal to running models. • Months of setup become days • One platform for GPUs, Kubernetes, and AI services • Your hardware, your data residency, your rules, with multi-tenant isolation • Run the GPUs you choose, NVIDIA, AMD, or Intel, anywhere you deploy See everything k0rdent AI does, from metal to model: https://buff.ly/NHiQgqx #Mirantis #k0rdentAI #NVIDIA #AIFactory #MetalToModel #AIInfrastructure

QumulusAI Expands AI Compute Capacity with NVIDIA Blackwell GPUs. Read more on our website: https://lnkd.in/gKfdkCjS QumulusAI has purchased 1,632 NVIDIA Blackwell B300 GPUs to expand its high-performance AI compute capacity and support growing customer demand. The expansion includes 204 NVIDIA HGX B300 systems and additional NVIDIA RTX PRO 6000 Blackwell GPUs to support AI training, inference, and visualization workloads. Mike Maniscalco, CEO of QumulusAI, highlighted the company’s focus on delivering flexible and scalable AI infrastructure at speed. #AIInfrastructure #NVIDIA #ArtificialIntelligence #CloudComputing #ITTechPulse #BestPlaceToWork

Fresh inventory just landed. 🚀 The wait is over. We now have NVIDIA H200s, H100s, and RTX 6000 Pros available and ready to deploy. ✅ Competitive pricing ✅ Flexible contract terms ✅ Monthly or long-term billing options ✅ Fast deployment Whether you’re training foundation models, scaling inference, or expanding GPU capacity, we’ve got the infrastructure to get you moving, without the long lead times. If AI compute is on your roadmap, let’s talk. #AI #GPU #NVIDIA #H200 #H100 #RTX6000Pro #AIInfrastructure #CloudComputing

Ever wondered what a GPU actually looks like? I recently got a chance to see NVIDIA’s GB200 NVL72 hardware and it’s quite different from what most of us imagine when we hear the word “GPU”. It’s not just one GPU. It’s 72 powerful Blackwell GPUs working together inside a single rack to process massive AI workloads. Think of it as a highly coordinated team of 72 specialists working on the same problem at the same time. The key is not just having 72 GPUs but making sure they can communicate with each other extremely fast. So when you see a ₹25–30 crore rack you’re not really looking at just a computer, you’re looking at a building block for an AI supercomputer. #AI #NVIDIA #GB200 #NVL72 #Blackwell #AIInfrastructure #DataCenters #GPUs #GenerativeAI #Technology

With all hyperscalers and AI labs building in house GPUs, NVIDIA has a chance to do the funniest thing ever. Cut their 75% margins and sell directly to consumer, this could make DGX Spark worth just as low as 1000$ 

Hyperscalers.com AI GPU Market Update for July / August Demand for enterprise AI GPUs continues to outpace supply across enterprise AI GPU platforms, with lead times extending across most product families. Current lead times: • H200 NVL – ~32 weeks • RTX PRO Blackwell 96GB – ~22 weeks • HGX B300 – <18 weeks August outlook: - Customers delaying purchasing decisions until August should expect longer lead times and higher pricing. - Demand continues to accelerate while manufacturing capacity remains constrained. - There is a growing risk that available production allocations will be exhausted over the next 2–3 months. Forward outlook: - If current demand trends continue, lead times approaching 12 months for many enterprise GPU platforms are increasingly likely. - Pricing pressure is expected to remain upward throughout the remainder of the year. Our recommendation: secure GPU allocations early to lock in current pricing and production capacity before availability tightens further. Contact Hyperscalers for current pricing and allocation availability. #AI #GPU #NVIDIA #EnterpriseAI #DataCentre #Hyperscalers

Saw this post about the Kimi K3 2.8T model and it got me thinking. Kimi K3 2.8T is MASSIVE. So massive that a single NVIDIA DGX B200 (even at FP4) can't run it. You need GPUs like GB300 NVL72, B300, or M1355X. Why? => Because each of those GPUs has 288 GB of memory. That's a lot. Now here's the interesting part one way to make it run on B200s is by connecting multiple nodes together using WideEP. 👉 What is WideEP? WideEP (Wide Expert Parallelism) lets the model spread different "experts" (parts of the model) across multiple machines. 👉 Think of it like teamwork: ◆ Model is split ◆ Each node handles a chunk ◆ They communicate and work together ◆ Final output comes as one The catch? B200 gives only 400 Gbit/s bandwidth between nodes. NVL72 gives 18x higher bandwidth. That's a huge difference. My take ◆ This is the future of running crazy large models. ◆ Not just raw GPU memory, but also how well GPUs talk to each other. ◆ Bandwidth is the hidden hero here. #AI #LLM #KimiK3 #NVIDIA #GPU #DistributedSystems #AIInfrastructure #MachineLearning #SoftwareEngineering #GenerativeAI #MLOps #SystemsEngineering

DeepSeek is raising API prices after its ultra-cheap V4 Flash 0731 model reportedly triggered more demand than its servers could handle. The model costs just $0.14 per million input tokens and $0.28 for output, but DeepSeek’s estimated 20,000 NVIDIA H100 GPUs are struggling during busy periods. Users have reported slower response speeds, turning a bold price war move into a major capacity problem. Download Voru app and read top AI & Tech stories in 5 minutes. Follow us (@vorumedia) to keep up with the latest AI & Tech news. Source: Wccftech

The Trump administration declared that Moonshot's Kimi K3 model was distilled from Anthropic's Fable and further trained on Nvidia's GB300 GPUs in Thailand. According to a report, Moonshot "developed a sophisticated internal platform to conduct large-scale distillation against US models, allowing them to switch between multiple methods of access to avoid detection quickly", with the AI startup already owning some Nvidia GB300 servers and being able to access additional units via Thailand. #AI #Nvidia #Anthropic #Moonshot #AIStartups #ExportControls #TechPolicy #US #TechNews #BusinessNews https://lnkd.in/djrGNvre

This is a great direction because customers get insights for how their workloads will perform in real life—not theoretical benchmarks!