CloudThrill uses Nebius for inference demonstrations and infrastructure deployments. Its vLLM-Omni lab served Z-Image, Wan2.2, Qwen3-TTS and NVIDIA Cosmos 3 on a single H100, swapping models on the same GPU. Models ran on the Nebius VM while requests came from a laptop through an SSH tunnel. CloudThrill published the demonstration scripts and announced a conference session featuring the Nebius H100 demo.
CloudThrill also documented vLLM Production Stack and NVIDIA Dynamo deployments on Nebius AI Cloud using Managed Kubernetes and Terraform. Its Dynamo deployment served Qwen3-8B and TinyLlama on one L40S through a shared endpoint, with TLS, Prometheus, Grafana and persistent model caches. In a routing test, it reported that all 20 requests sharing a prompt prefix reached the cache-warm replica. The multimodal article separately described a planned inference gateway on Nebius using vLLM Production Stack, LiteLLM, LLM Guard and token-based rate limiting.
CloudThrill reported end-to-end deployment times of approximately 20 minutes 41 seconds for vLLM Production Stack and 35–40 minutes for Dynamo. The Dynamo stack used private subnets, VPC-Cilium networking with Hubble observability and NGINX exposed through a Nebius load balancer. Its multimodal request menu dispatched and timed calls from a laptop, while model servers were started separately and swapped manually on the GPU.
CloudThrill's Nebius H100 lab uses a laptop menu to dispatch and time requests while model servers are started separately and swapped manually. Wan video generation uses asynchronous submission and polling, while Cosmos returns video synchronously.
vLLM-Omni on Nebius H1007 Jul 2026 · CloudThrill · Web pageCloudThrill's Nebius H100 lab uses a laptop menu to dispatch and time requests while model servers are started separately and swapped manually. Wan video generation uses asynchronous submission and polling, while Cosmos returns video synchronously.
vLLM Production Stack on Nebius Kubernetes
20 Jan 2026 · CloudThrill · Web page
CloudThrill reports an approximately 20-minute, 41-second vLLM Production Stack deployment on Nebius AI Cloud. The deployment serves TinyLlama-1.1B with a round-robin router across replicas and persistent model caching.
vLLM Production Stack on Nebius Kubernetes20 Jan 2026 · CloudThrill · Web pageCloudThrill reports an approximately 20-minute, 41-second vLLM Production Stack deployment on Nebius AI Cloud. The deployment serves TinyLlama-1.1B with a round-robin router across replicas and persistent model caching.
CloudThrill reports a verified NVIDIA Dynamo deployment on Nebius AI Cloud Managed Kubernetes using Dynamo 1.3.0, an L40S GPU and Kubernetes 1.34. Its Terraform blueprint includes persistent model caches and monitoring.
GitHub - CloudThrill/nvidia-dynamo-stack-terraform · GitHub19 Oct 2026 · github.com · Web pageCloudThrill reports a verified NVIDIA Dynamo deployment on Nebius AI Cloud Managed Kubernetes using Dynamo 1.3.0, an L40S GPU and Kubernetes 1.34. Its Terraform blueprint includes persistent model caches and monitoring.
NVIDIA Dynamo on Nebius Managed Kubernetes
8 Sept 2026 · CloudThrill · Web page
CloudThrill reports a 35–40-minute NVIDIA Dynamo deployment on Nebius Managed Kubernetes. The stack uses private subnets, VPC-Cilium with Hubble observability and NGINX exposed through a Nebius load balancer. Initial runtime-image and model downloads account for much of the deployment time.
NVIDIA Dynamo on Nebius Managed Kubernetes8 Sept 2026 · CloudThrill · Web pageCloudThrill reports a 35–40-minute NVIDIA Dynamo deployment on Nebius Managed Kubernetes. The stack uses private subnets, VPC-Cilium with Hubble observability and NGINX exposed through a Nebius load balancer. Initial runtime-image and model downloads account for much of the deployment time.
How vLLM-Omni Unifies Multimodal Inference
19 Oct 2026 · YouTube · Video
Kosseila describes using an H100 VM for recorded multimodal inference tests. CloudThrill's accompanying announcement identifies the demonstration platform as Nebius. The session includes reference-image video generation and reported diffusion-caching acceleration. Evidence at 40:27.
How vLLM-Omni Unifies Multimodal Inference19 Oct 2026 · YouTube · Video · Starts at 40:27Kosseila describes using an H100 VM for recorded multimodal inference tests. CloudThrill's accompanying announcement identifies the demonstration platform as Nebius. The session includes reference-image video generation and reported diffusion-caching acceleration. Evidence at 40:27.
CloudThrill provides cloud, AI infrastructure and DevOps consulting for businesses managing cloud and hybrid environments. Its services include infrastructure migration, automation, cloud cost optimization, security and compliance, and Oracle database migration. Its Private AI Inference Service helps organizations deploy open-source language models within their own infrastructure, using tools such as vLLM and Ollama. CloudThrill also offers technical advisory services and corporate training for engineering teams adopting cloud-native and AI infrastructure technologies.
CloudThrill provides cloud, AI and DevOps consulting, infrastructure migration, management and private inference services. · Business services
How they use Nebius AI Cloud
CloudThrill uses Nebius for inference demonstrations and infrastructure deployments. Its vLLM-Omni lab served Z-Image, Wan2.2, Qwen3-TTS and NVIDIA Cosmos 3 on a single H100, swapping models on the same GPU. Models ran on the Nebius VM while requests came from a laptop through an SSH tunnel. CloudThrill published the demonstration scripts and announced a conference session featuring the Nebius H100 demo.
CloudThrill also documented vLLM Production Stack and NVIDIA Dynamo deployments on Nebius AI Cloud using Managed Kubernetes and Terraform. Its Dynamo deployment served Qwen3-8B and TinyLlama on one L40S through a shared endpoint, with TLS, Prometheus, Grafana and persistent model caches. In a routing test, it reported that all 20 requests sharing a prompt prefix reached the cache-warm replica. The multimodal article separately described a planned inference gateway on Nebius using vLLM Production Stack, LiteLLM, LLM Guard and token-based rate limiting.
CloudThrill reported end-to-end deployment times of approximately 20 minutes 41 seconds for vLLM Production Stack and 35–40 minutes for Dynamo. The Dynamo stack used private subnets, VPC-Cilium networking with Hubble observability and NGINX exposed through a Nebius load balancer. Its multimodal request menu dispatched and timed calls from a laptop, while model servers were started separately and swapped manually on the GPU.
CloudThrill's Nebius H100 lab uses a laptop menu to dispatch and time requests while model servers are started separately and swapped manually. Wan video generation uses asynchronous submission and polling, while Cosmos returns video synchronously.
vLLM-Omni on Nebius H1007 Jul 2026 · CloudThrill · Web pageCloudThrill's Nebius H100 lab uses a laptop menu to dispatch and time requests while model servers are started separately and swapped manually. Wan video generation uses asynchronous submission and polling, while Cosmos returns video synchronously.
vLLM Production Stack on Nebius Kubernetes
20 Jan 2026 · CloudThrill · Web page
CloudThrill reports an approximately 20-minute, 41-second vLLM Production Stack deployment on Nebius AI Cloud. The deployment serves TinyLlama-1.1B with a round-robin router across replicas and persistent model caching.
vLLM Production Stack on Nebius Kubernetes20 Jan 2026 · CloudThrill · Web pageCloudThrill reports an approximately 20-minute, 41-second vLLM Production Stack deployment on Nebius AI Cloud. The deployment serves TinyLlama-1.1B with a round-robin router across replicas and persistent model caching.
CloudThrill reports a verified NVIDIA Dynamo deployment on Nebius AI Cloud Managed Kubernetes using Dynamo 1.3.0, an L40S GPU and Kubernetes 1.34. Its Terraform blueprint includes persistent model caches and monitoring.
GitHub - CloudThrill/nvidia-dynamo-stack-terraform · GitHub19 Oct 2026 · github.com · Web pageCloudThrill reports a verified NVIDIA Dynamo deployment on Nebius AI Cloud Managed Kubernetes using Dynamo 1.3.0, an L40S GPU and Kubernetes 1.34. Its Terraform blueprint includes persistent model caches and monitoring.
NVIDIA Dynamo on Nebius Managed Kubernetes
8 Sept 2026 · CloudThrill · Web page
CloudThrill reports a 35–40-minute NVIDIA Dynamo deployment on Nebius Managed Kubernetes. The stack uses private subnets, VPC-Cilium with Hubble observability and NGINX exposed through a Nebius load balancer. Initial runtime-image and model downloads account for much of the deployment time.
NVIDIA Dynamo on Nebius Managed Kubernetes8 Sept 2026 · CloudThrill · Web pageCloudThrill reports a 35–40-minute NVIDIA Dynamo deployment on Nebius Managed Kubernetes. The stack uses private subnets, VPC-Cilium with Hubble observability and NGINX exposed through a Nebius load balancer. Initial runtime-image and model downloads account for much of the deployment time.
How vLLM-Omni Unifies Multimodal Inference
19 Oct 2026 · YouTube · Video
Kosseila describes using an H100 VM for recorded multimodal inference tests. CloudThrill's accompanying announcement identifies the demonstration platform as Nebius. The session includes reference-image video generation and reported diffusion-caching acceleration. Evidence at 40:27.
How vLLM-Omni Unifies Multimodal Inference19 Oct 2026 · YouTube · Video · Starts at 40:27Kosseila describes using an H100 VM for recorded multimodal inference tests. CloudThrill's accompanying announcement identifies the demonstration platform as Nebius. The session includes reference-image video generation and reported diffusion-caching acceleration. Evidence at 40:27.
CloudThrill provides cloud, AI infrastructure and DevOps consulting for businesses managing cloud and hybrid environments. Its services include infrastructure migration, automation, cloud cost optimization, security and compliance, and Oracle database migration. Its Private AI Inference Service helps organizations deploy open-source language models within their own infrastructure, using tools such as vLLM and Ollama. CloudThrill also offers technical advisory services and corporate training for engineering teams adopting cloud-native and AI infrastructure technologies.
