vCluster

At what point does a GitOps hub become the bottleneck when managing 15,000 Kubernetes clusters in a Hub-and-Spoke architecture? That question came from a real kubara community request, so we built a test model try to predict the behavior instead of guessing. TL;DR: We tested GitOps fleet management with Argo CD, vCluster, kubara, and Project Sveltos. In our setup, Argo CD’s application controller started hitting OOM kills at around 15k–20k cached objects per hub. Hydrated manifests helped, tuning helped only partially, and Sveltos handled addon-style rollout patterns at a fraction of the memory: around 2 GB compared to around 21 GB with Argo CD. The main lesson: at very large scale, architecture matters more than tuning. And combining GitOps engines is a real multiplier for what you can achieve with Open Source. Cost was also a key part of the test: running this at full target scale would have cost roughly €1.2M. With support from STACKIT, and by using vCluster to design realistic peak scenarios, we brought the cost down to around €120K while still testing 1,000 clusters and up to 5,000 heavy Applications. More in the blog post: https://lnkd.in/esMf5Qxd --- P.S. This is an honest reflection based on weeks of work. We simply wanted to share with the community what we tested, what we learned, and what we saw along the way.

RAISE Summit is almost here! Our team will be at booth 15D in the Le Nôtre suite across both days. Whether you're evaluating Kubernetes tenant isolation, figuring out how to scale AI workloads without spinning up full clusters, or just want to connect — come find us. We're also presenting in the Foundation Track stage July 9 at 12pm with Bola Malek (Product Lead, Baseten) and Nikita Vdovushkin (Product Director, Token Factory, Nebius) about AI Infra for Multi-model and Multi-Cloud Scale. Paris, July 8–9. See you there. #RAISE2026 #Kubernetes #AIInfrastructure #PlatformEngineering #AIInference

🔥Panel Alert: On July 9 at 12pm CEST our very own Lukas Gentele will be joined by Bola Malek from Baseten and Nikita Vdovushkin from Nebius to talk about what it actually takes to run inference at production scale across clouds and models. if you are attending RAISE Summit in Paris, don’t miss this session. #AI #Inference #Infrastructure #MultiModel #MultiCloud

Still buzzing from KubeCon + CloudNativeCon India? We definitely are! Team vCluster had an incredible week connecting with the community, sharing ideas, and having meaningful conversations around the future of Kubernetes, AI infrastructure, and multi-tenancy. During the KubeCon keynote, Saiyam Pathak showcased a mini AI factory demo using HAMi to slice GPU resources on a DGX Spark. Each GPU slice was then made available to separate teams through their own isolated Kubernetes environments powered by vCluster. The idea was simple but powerful: give teams a managed Kubernetes-like experience while safely sharing expensive GPU infrastructure. Later, Hrittik Roy and Saloni Narang took the stage with “Phippy’s First Steps into Kubernetes,” walking through the journey from Minikube to vind(vCluster in Docker), and showing how different local Kubernetes environments fit different use cases. A huge thank you to everyone who joined the sessions, stopped by for conversations, and shared their thoughts with us. If you are building an AI factory and want to offer your teams or customers isolated, managed Kubernetes environments with vCluster, comment “Hi” or send us a DM. We would be happy to chat.

On July 9, our CEO, Lukas Gentele, will moderate a fireside chat in the Foundation track at RAISE Summit 2026. The topic is "AI Infra for Multi-model and Multi-Cloud Scale." Joining him on stage: - Bola Malek (Product Lead, Baseten ) - Nikita Vdovushkin (Product Director, Token Factory, Nebius) 12pm. Foundation Track. Le Carrousel du Louvre | Ada Lovelace stage. If you're at RAISE, this is worth the lunch break. #RAISE2026 #FoundationTrack #AIInfrastructure #Kubernetes #AIInference

Day 2 ops at AI Cloud fleet scale is brutal. Drift across tenant clusters. CVE patches that have to land everywhere by Tuesday. Manual rollouts that turn every change into a campaign. vCluster Platform v4.10 ships today with a new Argo CD integration that solves all three. Platform teams declare the full tenant stack in one vcluster.yaml. Argo CD reconciles it across every tenant, continuously. Internal systems ship in the same file as tenant apps, no dedicated cluster required. Works with self-hosted Argo CD, or with Akuity for teams that need an agent-based, enterprise-ready Argo CD. Production-ready, every tenant, every time. Read the docs: https://lnkd.in/dg_fTJHf

Platform v4.10 and vCluster v0.35 are live. This release focuses on GitOps-driven workflows, modern Kubernetes networking, and expanded infrastructure visibility across VMs, machines, and tenant clusters. -𝗙𝗶𝗿𝘀𝘁-𝗰𝗹𝗮𝘀𝘀 𝗔𝗿𝗴𝗼 𝗖𝗗 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 lets you connect Argo CD or Akuity once, then have Platform register clusters, manage Applications, and clean them up with the tenant cluster lifecycle. Teams can now declare apps directly in vCluster config. -𝗚𝗮𝘁𝗲𝘄𝗮𝘆 𝗔𝗣𝗜 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 is the forward path for exposing Platform, vCluster APIs, and tenant workloads, with support for `TLSRoute`, shared Gateways, and Gateway-backed sleep mode. -𝗩𝗶𝗿𝘁𝘂𝗮𝗹 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 surfaces VMs alongside bare metal Machines, giving teams a unified view of KubeVirt-backed infrastructure. It launches as an experimental feature. Also shipping: Helm chart secret references, improved project infrastructure visibility, external peer and Tailscale SSH improvements, clearer node claim status, UI polish, and more reliable private-node upgrades. Full changelog: https://lnkd.in/gYb8sZPM #Kubernetes #PlatformEngineering #vCluster #CloudNative #DevOps #GitOps

AI infrastructure doesn't start with Kubernetes. It starts with knowing what infrastructure exists, how it's deployed, and how workloads map to the underlying physical environment. Join NetBox Labs and vCluster tomorrow to learn how teams connect infrastructure data with Kubernetes to improve GPU utilization, increase visibility, and operate AI environments with greater confidence and control. Can't attend live? Register anyway and we'll send you the recording and slides. Register here: https://hubs.ly/Q04lJCdB0

We're heading to RAISE Summit 2026 as a Neural Sponsor in less than a month! July 8–9. Le Carrousel du Louvre, Paris. Find us at booth #15-D in the Carrousel du Louvre’s Le Nôtre suite (under street level). Come talk Kubernetes, tenant isolation, and what running AI at scale looks like in practice. Mark your calendars! Our CEO, Lukas Gentele, will host a fireside chat with panelists Bola Malek (Product Lead, Baseten) and Nikita Vdovushkin (Product Director, Token Factory, Nebius) - Topic: AI Infra for Multi-model and Multi-Cloud Scale - Thursday, July 9 at 12 pm We'll be in the Foundation track, where the infrastructure conversations actually matter. Compute economics, sovereign cloud, scaling AI workloads in production. That's the work we care about. #RAISE2026 #AIInfrastructure #PlatformEngineering #Kubernetes #vCluster

Our very own Saiyam Pathak is taking the Keynote stage again at Kubecon India. If you want to learn about AI factories, how to build one with a live demo, then you cannot miss this Keynote. He is also co-chair for this Kubecon India. Latest news - you can find him in his fancy jacket and ask him about vCluster!

Do join our very own Saiyam Pathak , who is taking the keynote stage at KubeCon + CloudNativeCon India. Learn more about AI factories and how to build one. I’ve heard there will be a live demo too! 🚀

SemiAnalysis built ClusterMAX, a 142-criteria framework for evaluating GPU cloud providers. vCluster is the only vendor named as a security requirement. The criterion: "𝘷𝘊𝘭𝘶𝘴𝘵𝘦𝘳 𝘰𝘳 𝘴𝘪𝘮𝘪𝘭𝘢𝘳 𝘪𝘴𝘰𝘭𝘢𝘵𝘪𝘰𝘯 𝘣𝘦𝘺𝘰𝘯𝘥 𝘤𝘰𝘯𝘵𝘢𝘪𝘯𝘦𝘳-𝘣𝘢𝘴𝘦𝘥 𝘰𝘯𝘭𝘺." Not a suggestion. A requirement. If you're building or operating a GPU cloud, this is what enterprise buyers will use to evaluate you. Security, lifecycle, networking, reliability, monitoring, and more. 142 criteria across 10 dimensions. We mapped how vCluster addresses each one. The results: 78 criteria covered today. 9 more on the roadmap. The full breakdown is at vcluster.com/clustermax, including a downloadable scorecard and a step-by-step guide to achieving ClusterMAX Gold.

AI infrastructure is breaking the old data center model. In this episode of Liftoff Keith Newman sat down with Lukas Gentele, CEO & Co-Founder of vCluster Labs, to unpack what it really takes to operate GPU infrastructure at scale in 2026. The companies that define the control layer for AI infrastructure will shape the next 20 years of computing just as VMware shaped the last 25. Lukas Gentele is building directly at that layer. If you’re a founder, operator, investor, or infrastructure leader, this episode will sharpen how you think about AI clouds — and where the real leverage sits. Keith Newman The GTM Firm Liftoff with Keith Watch now: https://lnkd.in/gQHqj6J2 Recorded Live at AI-Infra 5 in the Ignite Studios IgniteGTM

NVIDIA released DSX OS at GTC Taipei and it’s the playbook for building AI factories at scale. DSX OS is open-source, modular software built to run AI infrastructure from end to end: lifecycle management, scheduling, health automation, resiliency, tenant isolation, and platform services. One co-designed stack, from chips to grid. The numbers speak for themselves: 40% more GPUs running at peak efficiency within the same power budget. Lower token costs. More throughput without new hardware. vCluster is named in the DSX OS ecosystem sitting in the platform services layer offering tenant isolation and tenant clusters to DSX deployments. If you’re building or operating AI infrastructure, check out vCluster >> vcluster.com #NVIDIA #DSX #AIFactory #GPUs #GTC2026 #vCluster #AIInfrastructure #Kubernetes #vCluster cc Saiyam Pathak

Kubernetes 1.36: User Namespaces Are Finally GA And Why It Matters for Tenant Isolation. Our very own Saiyam Pathak did a deep dive on, the CVE map, the YAML, and where every isolation layer stops. Comment "usernamespace" and we will send you the link to the deep dive. #Kubernetes #Security #Isolation #vNode #NVIDIAScape

This June is going to be one of the busiest months for me, as I’ll be delivering several talks on some of the biggest stages in India, covering topics ranging from MCP and Kubernetes to GPU platforms! Very excited to be presenting alongside my fellow speakers and sharing ideas that are changing the way we think about compute, scale and capacity 🚀 If you’re attending any of these conferences, say hi, and let’s catch up! 👋

Just published the first part of my new blog series: Cilium Powered vClusters. Part 1 covers how to set up Cilium as the CNI inside a vind cluster, replacing the default Flannel and kube-proxy with eBPF-powered networking, step by step. If you work with Kubernetes and want to understand how Cilium fits into a vCluster setup, this is a good starting point. Read it here: https://lnkd.in/drw3mUPY #Kubernetes #vCluster #Cilium #CloudNative #Networking #eBPF

🤠 Howdy...We are at KCD Texas, Austin - May 15, 2026 today and vCluster is proud to be a Platinum Sponsor! If you are around and want to learn about vCluster, vMetal, vNode, go talk to our very own Diaa Khalil and John Barfield. #KCDTexas #Kubernetes #vCluster #CloudNative #CNCF #Austin #PlatformEngineering #DevOps #SRE #AI

Experience the vCluster stack for AI infrastructure, purpose-built to take you from bare metal to production AI workloads. With vMetal for automated bare-metal provisioning, vNode for secure tenant isolation at the container level, and vCluster for tenant clusters on any Kubernetes environment, our platform enables AI Clouds and inference providers that are fast to deploy, easy to manage, and built for strict compliance requirements. 📍 RAISE Summit 2026, Carrousel du Louvre, Paris 📆 July 8-9 👋 Visit us at booth #15D in the Le Notre hall https://lnkd.in/dSyDH_px #RAISE2026 #RAISESummit #AI #GPU #Cloud #Infrastructure #Inference

We are live at AI Infra Summit with the panel, “Building Hyperscaler-grade Neocloud” Duncan Ng Yujing Qian J. Jubran Lukas Gentele Vultr GMI Cloud IBM vCluster #AIInfra #GPU #Cloud #Infrastructure

🚨 Worried about Copy Fail exploits? Not if you're using vNode. vNode had you covered — another vulnerability contained! Our team verified today: Customers running vNode were fully protected, even without patching the host kernel. Zero additional action needed on their end. Protected against even the ugliest zero-day exploits. This isn't the first time. Last year, when Wiz Research uncovered NVIDIAScape — a critical vulnerability affecting NVIDIA container toolkits that put a massive number of GPU workloads at risk — vNode protected against that one too. Two major container vulnerabilities in under 12 months. Both affecting a huge swath of the industry. Both particularly dangerous for anyone building AI infrastructure or running bare metal workloads, where the attack surface is large and the stakes are high. The pattern is clear: the threat landscape for containerized AI infrastructure is getting more hostile, not less. That's exactly what vNode was built for. vNode operates as an additional isolation layer between container and host kernel: That means that vulnerabilities in the host kernel don't become exploitable inside the virtual nodes. Attackers break out of your container but are still stuck in our virtual node layer. Your data stays protected. Your customers stay safe — even on zero-day, before a patch exists. 🔒 If you're building AI infrastructure and you're not running vNode, you're missing a critical protection layer. We're currently in private beta and leading neoclouds like Nscale are already using our tooling. Beta spots are limited — and given what the last 12 months have looked like and what's ahead with Mythos, I wouldn't wait. 👉 Request access on the vNode website or drop a comment below to get in touch. PS: Full blog post with all details and reproducible steps coming tomorrow. #AIInfrastructure #ZeroDay #ContainerSecurity #vNode #CopyFail #BareMetalSecurity #CloudSecurity #GPUSecurity #AICloud #TenantIsolation

Excited to share that I co-authored a piece on the AWS Architecture Blog! The post walks through how we dramatically cut EKS testing environment provisioning time, from 45 minutes down to under 5, using Amazon EKS Auto Mode and vCluster. Beyond the speed gains, the approach reclaims significant engineering hours, frees up meaningful compute and memory through resource sharing, and consolidates virtual clusters onto a single shared host cluster. I genuinely had fun with the design choices and the writing process, turning a real engineering problem into something others can pick up and build on. I co-authored this post with Samuel Lefki , Keith Hodo , Sai Samala , and Sai Charan Teja G., and I'm proud of what the team delivered. A special thank you to Samuel for the leadership and guidance throughout the writing process. It made a real difference. Give it a read if you're working with EKS or thinking about how to scale environments without scaling cost and complexity along with it 👇 https://lnkd.in/eGDVPT-X #AmazonEKS #Kubernetes #AWS #CloudArchitecture #DevOps #PlatformEngineering #CostOptimization #InfrastructureAsCode #vCluster #EKSAutoMode #QAEngineering

Tomorrow, May 1 @ 12PM, Main Stage at AI Infra Summit in Sunnyvale and the topic is, "Building a Hyperscaler-Grade Neocloud". Lukas Gentele will be moderating an exceptional panel: Duncan Ng, VP of Solutions Engineering at Vultr, Yujing Qian, Head of Engineering at GMI Cloud, J. Jubran, Global Head of AI Infrastructure at IBM They will go deep on the topics that actually matter to the AI Cloud operators: - Is hyperscaler-grade even the right bar? - Training vs. inference: how has the mix shifted? - SLURM, Kubernetes, and a wave of inference-specific runtimes - SLAs and failure domains - The tokenomics - How billing models are being renegotiated #AIInfra #AICloud #Kubernetes #GPUCloud #InferenceAtScale Bill Barry Serena Yu IgniteGTM

NeoClouds are moving fast. Getting GPUs is no longer the hard part. Running them at hyperscaler grade is. That’s where things break. Provisioning environments in minutes, not days. Isolating tenants without killing performance. Scheduling GPUs across fragmented clusters. Keeping Kubernetes, Slurm, and bare metal from turning into operational drag. That’s the real work behind “AI cloud.” At AI INFRA SUMMIT 5 (May 1), we’re putting the operators doing this on the same stage: Building a Hyperscaler-Grade NeoCloud Moderated by: Lukas Gentele With Yujing Qian, GMI Cloud Duncan Ng, Vultr J. Jubran, @IBM This is a working conversation. What actually works for fast tenant provisioning. How teams are handling GPU scheduling at scale. Where Kubernetes holds up, and where it doesn’t. What breaks when you try to operate like a hyperscaler without hyperscaler margins. If you’re building or evaluating NeoCloud infrastructure, this is one of the sessions to be in. May 1 https://luma.com/aiinfra5 IgniteGTM Alex Moon Andreas Mueller Bob Karr Brad Garsten Brian💥 Sparkes Carmen Li Dawn DeBruyn Novarina Deepak Rana Doug Shannon Edward Preston Faisal Mushtaq Henry Park Igor Krasnykh

🚨 Live Panel Alert: Building a Hyperscaler-Grade Neocloud Join Duncan Ng, VP of Solutions Engineering at Vultr, Yujing Qian, Head of Engineering at GMI Cloud and our very own Lukas Gentele on May 1 as they dive deep into the architectural and operational playbooks powering the next generation of AI clouds. 📍 AI Infra Summit, Plug and Play Tech Center, Sunnyvale, CA 📅 Friday, May 1 ⏰ 12:00 PM PT 🔗 https://www.aiinfra.live/ Bill Barry Serena Yu IgniteGTM

Build or Buy? (Managed Kubernetes for your AI Cloud) Ryan DiRocco, CTO at QumulusAI says: "We asked ourselves this exact question." "We're an AI company. We dog food our own tools. Stitching together open source Kubernetes seemed doable." "Our focus and our genesis is around infrastructure, infrastructure at scale and delivering infrastructure quickly." "The ability to have effectively a trusted advisor, a partner there to help deliver the platform, as opposed to, just a software vendor, just a piece of downloaded software, that allows us to put together the right offering, for our clients." Watch the full livestream here: https://lnkd.in/d_hWx6tt

🚨 GPU/AI Infra Livestream Alert! Join Ryan DiRocco, CTO at QumulusAI and Lukas Gentele, CEO at vCluster as they discuss: - How to dynamically scale clusters based on compute needs and achieve an EKS-like experience on GPUs - How to deliver enterprise-grade Kubernetes environments for AI development - How QumulusAI Deploys AI Infrastructure in Minutes, Not Days 📅 April 14 ⏰ 7:00 AM PT / 10 AM ET / 4:00 PM CEST #GPU #AI #Infrastructure #Pipelines #Kubernetes #vCluster #BareMetal

❗️ How QumulusAI Deploys AI Infrastructure in Minutes, Not Days. Interested in learning more? Register for the livestream on April 14 here >> https://lnkd.in/dVqGs74W #AIInfrastructure #GPU #Kubernetes #vCluster