SGLang Hosting in Germany
WZ-IT designs SGLang as an inference layer for local AI, benchmarks models and sizes GPUs, concurrency, context and availability for the real workload.
About the technology
Size inference with real load rather than a data sheet
Response time and throughput depend on model, quantisation, context length, batch size, concurrency and GPU architecture. A single token figure does not describe business operations.
We test target models with representative inputs and design routing, model cache, metrics, API protection and scaling as part of the platform.
SGLang for production model serving
Serve models efficiently
Serve supported language and multimodal models with optimised scheduling and caching.Measure throughput
Benchmark latency, time to first token, token throughput and concurrency with realistic load profiles.Provide APIs
Expose OpenAI-compatible endpoints to approved applications and internal platforms.Manage models
Document versions, quantisations, launch parameters and rollout paths transparently.Secure access
Protect API authentication, network paths, tenant boundaries and administrative access.Integrate platforms
Connect Open WebUI, LiteLLM, agents, RAG services and custom applications to the inference layer.Migration and takeover
Assess existing data and configuration, migrate them in a test run and move to managed operations through a controlled cutover.Identity and secure access
Secure SSO, roles, administrative paths and external access for the application and existing infrastructure.Backup and recovery
Back up all stateful components consistently and document the recovery path for the agreed scope.Updates, monitoring and operations
Monitor and update the application and its technical dependencies and operate them under the agreed service level.
Operations beyond installation
Operate SGLang for high-performance LLM inference
Model, quantisation, runtime, GPU, context and scheduling form one system that must be benchmarked under realistic load.
Inference runtime
Deploy SGLang, drivers, kernels and parameters reproducibly.Models and endpoints
Version weights, quantisation, context and APIs.Cache and scheduling
Optimise KV cache, batching, concurrency and memory limits.Benchmark and operations
Monitor latency, throughput, failures and GPU use.
Area
Responsibility for inference and model use
WZ-IT operates the runtime; model quality, licensing and permitted use are shared or customer-owned.
| Area | Responsibility | Scope and boundaries |
|---|---|---|
| SGLang runtime | WZ-IT | Deployment, drivers, containers, monitoring and updates. |
| Model artefacts | WZ-IT | Controlled download, storage and launch configuration. |
| Load profile | Shared | We measure while the customer supplies representative prompts. |
| API integration | Shared | Authentication, limits and clients are tested together. |
| Model approval | Customer | Suitability, licences and outputs remain customer-owned. |
| Routing and scaling | Optional WZ-IT service | Multiple models, GPUs and gateways are scoped separately. |
Hosting & operations
How much does SGLang hosting cost?
We combine the right compute size with ongoing operations, backups, monitoring and a service level appropriate for the criticality of SGLang. High availability and recovery targets are designed separately where needed.
Hosting for SGLang in Germany: GDPR-compliant operations
- One managed standard application included
- .de domain included or your own domain or subdomain via CNAME
- German data centres with ISO 27001 and BSI C5, no US clouds
- 99.9 % standard availability per month
- Regular updates, CVE monitoring and prioritised patches for critical vulnerabilities
- Central logs with immutable retention
- Daily backup with 7-day retention
- 5 TB outbound traffic per workload and month
- 24/7 proactive monitoring
- Five service levels, 24/7 P1 response with Production or Critical
- Personal technical contact
Instance sizes and monthly prices
- XS · 2 vCPU, 2 GB RAM, 40 GB NVMe SSD
- €99.90 in total
- S · 2 vCPU, 4 GB RAM, 80 GB NVMe SSD
- €129.90 in total
- M · 4 vCPU, 8 GB RAM, 80 GB NVMe SSD
- €159.80 in total
- L · 8 vCPU, 16 GB RAM, 80 GB NVMe SSD
- €219.80 in total
- XL · 16 vCPU, 32 GB RAM, 80 GB NVMe SSD
- €329.80 in total
- Additional storage per 50 GB
- +€4.99
Straightforward provisioning within a few business days
- Bank transfer
All prices are net and exclude statutory VAT. The offers are addressed to businesses. All on-premises options
Other options
Try it free for 7 daysRequest a test instance
We set up a test instance for you, usually on the next business day. No payment details required. After seven days it is deleted unless you continue.
Request a test instance →Bring Your Own InfrastructureInstallation on Your Infrastructure
Installation on your own infrastructure · On-premise or in your cloud · Control over data location and access · Custom configuration · Complete documentation · Initial setup & configuration · Optional support and maintenance contract
Request a proposal →On-premises optionRun SGLang on hardware at your organisation
As an alternative to managed hosting in the data centre, WZ-IT provides the hardware, configures SGLang, and handles hardening, monitoring, updates, backup and technical support. Access can be limited to the internal network or enabled through VPN and existing identities.
View WZ-IT App Node →Sizing and deployment
Size SGLang from real model and workload targets
Weights, KV cache, context, batch size, GPU type and TTFT targets must be considered together.
| Usage scenario | Technical starting point | Key factors |
|---|---|---|
| One model and few chats | Model and GPU benchmark | Quantisation, context and interactivity are measured. |
| Several teams or API clients | Concurrency test | Throughput, queueing and response time are measured. |
| Long context or agents | Memory and prefill assessment | KV cache and input load determine capacity. |
| Multiple GPUs or models | Cluster and routing design | Parallelism and failover are planned. |
- Pricing follows model, hardware, context, concurrency and target metrics.
Sizing and deployment
SGLang deployment models
The runtime runs on dedicated GPU hardware close to calling applications.
Dedicated GPU infrastructure
European inference service after benchmarking.Your GPU cloud
Deployment in your account.On-premises
Operation on suitable local GPU servers.Hybrid
Combine local and external inference endpoints.
Target architecture
Expose LLM inference as a bounded platform service
Clients use an authenticated, measured API path rather than direct GPU access.
Chats, agents and applications
OpenAI-compatible or project-specific clients.
Inference gateway
TLS, authentication, quotas and request limits.
Tenants and API access
Separate keys, models and usage boundaries.
SGLang runtime
Scheduling, batching, cache and model serving.
GPU and drivers
Accelerators and kernels for the selected model.
Model repository
Versioned weights and quantisations.
Metrics and logs
TTFT, token rate, queue, memory and failures.
Benchmark results only apply to the tested model, runtime, hardware, context and load profile.
Role in the overall system
Related services for SGLang
This page covers installation and the agreed platform operations for SGLang. Custom code, the access layer and special confidentiality requirements remain clearly separated responsibilities that can be added when needed.
Support custom code and integrations
Keep custom extensions, interfaces and dependencies controlled, updated and supportable.
View service Secure accessConnect users, sites and systems
Add identity-based access and private network paths as a separate operational layer.
View service Operate confidentiallyAssess Section 203 requirements separately
For professional secrecy holders, define contractual, access and infrastructure requirements in a dedicated operating model.
View serviceThree relevant adjacent paths instead of a long list of further products.
View the overall systemFrequently asked questions about SGLang hosting
Answers about models, GPUs, benchmarks and operations.
- SGLang Hosting with WZ-IT starts at €99.90 net per month. It includes a dedicated size XS instance (2 vCPU, 2 GB RAM, 40 GB NVMe SSD), updates, daily backups with 7 days of retention, 24/7 monitoring, 5 TB of outbound traffic and a first response to critical incidents by the end of the next business day. Larger instances and higher service levels cost extra; the configurator shows the total price.
- The entry level is a dedicated size XS instance with 2 vCPU, 2 GB RAM, 40 GB NVMe SSD for €99.90; it suits lightweight applications as well as test and development environments. Size S with 2 vCPU, 4 GB RAM, 80 GB NVMe SSD costs €129.90. We check which size SGLang needs in production before provisioning. Size M (4 vCPU, 8 GB RAM, 80 GB NVMe SSD) costs an extra €29.90 excl. VAT, L (8 vCPU, 16 GB RAM, 80 GB NVMe SSD) an extra €89.90 excl. VAT and XL (16 vCPU, 32 GB RAM, 80 GB NVMe SSD) an extra €199.90 excl. VAT, each per month. Additional storage costs €4.99 excl. VAT per 50 GB and month.
- No. WZ-IT does not limit the number of users. How many users can work smoothly at the same time depends on usage and the chosen instance size XS to XL.
- SGLang Hosting can be cancelled monthly, with no minimum term and 30 days' notice to the end of the month.
- Germany and Switzerland are available at the same prices, in both cases without US clouds. The data centres are ISO 27001 certified; in Germany we additionally operate with a German provider whose data centres are BSI C5 attested. The Section 203 Managed Cloud for professional-secrecy holders is available in Germany only.
- We typically provide a production instance within a few business days after signing. For a trial, we set up a test instance, usually on the next business day; it runs free of charge for 7 days and without payment details.
- Yes. One .de domain is included with every hosting plan. Alternatively, connect your own domain or subdomain with a CNAME record.
- It is relevant for performance-focused serving, especially where caching and concurrency matter.
- Only after measuring context, input, output and concurrency against target latency.
- No. GPU type and quantity are quoted after assessment.
- Yes. We deploy it on suitable local GPU hardware.
- Yes. We benchmark both with identical models, prompts and concurrency.
Knowledge
More about SGLang
Tutorials & guides
- Self-Hosted ChatGPT: vLLM with Qwen3.5-122B on Two RTX PRO 6000 Blackwell GPUsYour own ChatGPT, running entirely in-house: a familiar chat interface, a large language model on your own hardware behind it, without a single prompt going...
- GPT-OSS 120B on the AI Cube: Run OpenAI's Open-Weight Model LocallyOpenAI released GPT-OSS 120B as an open-weight reasoning model on 5 August 2025. Its native MXFP4 quantisation allows OpenAI to position the model for a...
- Local AI Inference with the AI Cube: Run AI Inside Your Own NetworkLocal AI inference means running a language or multimodal model on owned hardware rather than through a public API. Organisations gain control over data paths,...
Matching solutions for your project
- Open WebUIUser-friendly web interface for LLMs with RAG pipeline, document chat, and Ollama integration
- LibreChatMulti-provider AI chat with local and external models, agents, MCP and granular configuration.
- LangfuseSelf-hosted LLM observability for tracing, prompt management, evaluations, costs and quality assurance
- vLLMHigh-performance inference and serving engine for large language models with high concurrency.
- OllamaLocal LLM inference engine for sovereign AI stacks with model management and OpenAI-compatible workflows
- LiteLLMMulti-LLM gateway with OpenAI-compatible API, provider routing, fallbacks, budgets and virtual keys
Reviews & projects
Client feedback and projects worldwide
WZ-IT moved our studio infrastructure from decentralised individual devices to a central platform: every site is securely connected via VPN, new devices are onboarded automatically and an entire site is provisioned from a template, without manual steps on location. What impressed me most is the breadth and depth of their knowledge: Timo and Robin are not a typical IT provider who sets up a server and leaves. The two of them think their way into highly complex infrastructure and software topics, work through every requirement we put in front of them, and build networking, provisioning and operations so that everything fits together in the end. WZ-IT is an excellent partner for complex software, network and architecture projects.
Steve KirchnerManaging Director, nextGYM GmbH
International
Built in Germany's Ruhr Valley. Running worldwide.
WZ-IT designs, develops and operates infrastructure and software for clients in Germany and internationally. We deliver projects remotely and continue supporting them in ongoing operations after go-live.
Selected projects
- LiveRemote access platform for distributed water treatment plantsABCO Water Systems · Australia · Secure Access & OT
- 81%AWS migration with substantially lower operating costsEVA Real Estate · UAE · Cloud exit & infrastructure
- 98%Provision new sites in minutes instead of hoursnextGYM GmbH · Germany · Automation & IoT
Read client reviews
- Secure your Proxmox & backup setup
- Modernize your infrastructure - sovereign
- Plan a sovereign open-source stack
- Integrate a local AI solution
- Modernize your legacy software
- Cut cloud cost - up to −81%
- Build a high-availability Proxmox cluster
- Virtualize with Managed Proxmox
- Design an open-source AI architecture
- Get collaboration fully managed
- Ship your prototype to production
- Connect sites and clusters securely
Hosting & operations
Enquire about hosting and operations for SGLang
Briefly describe the current state and objective for SGLang. We assess infrastructure, integration, and ongoing operations.
- Straight with Timo and Robin - no sales team, no pitch
- A professional assessment of your situation
- Concrete next steps for infrastructure, software or AI
The conversation is free and non-binding.
WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.
Jakob ÖschlbergerInno7 GmbH
WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.