Best enterprise AI gateways for European companies (2026)
Last updated 9 November 2026. Rival facts checked on each vendor's own pages that day.
Our pick for most European companies is Standard Compute. It routes each step to the cheapest model that can do it and the cheapest provider fast enough to serve it, across frontier and open models, so the bill falls without anyone choosing models by hand. The controls a European security review asks for are built in: a Europe rule that keeps every request on approved European inference endpoints, zero data retention, SAML SSO with Microsoft Entra ID, a fixed monthly budget per person and app, and an audit log of every admin change.
Standard Compute is our product, so weigh this page accordingly. Every rival is described from its own public pages, with links, and each one gets an honest note on when it is the better choice. A self-hosted gateway such as LiteLLM or Kong fits a platform team that wants to run the infrastructure itself, and a cloud-native option such as AWS Bedrock or Azure API Management fits a company standardised on one cloud.
Ten gateways compared
What each vendor publishes about the controls European buyers ask for first. "Not stated" means we could not find it on the vendor's own pages on 9 November 2026. Sources are listed under each gateway below.
| Gateway | Data location and retention | Access and spend | Models and routing | Pricing model |
|---|---|---|---|---|
| Standard Compute | EU processing option: Europe rule per company, team or key; EU gateway in Finland, account data in Frankfurt Zero data retention: Per company, team or key; inherited and cannot be loosened below | SSO: SAML incl. Microsoft Entra ID; Microsoft sign-in for Claude Desktop and Claude Code Budgets: Fixed monthly budget per person and app; caps per key and company | Open + frontier models: Claude, GPT, Gemini plus GLM, DeepSeek, MiniMax, Kimi Routing to cheapest provider: Each step to the cheapest capable model and the cheapest provider fast enough | Pricing model: Flat monthly plan per person or app |
| EUrouter | EU processing option: EU-hosted endpoints by default; EU-owned provider and country pinning options Zero data retention: Prompts not stored by default; zdr and no-training request flags | SSO: Not stated Budgets: Per-key spend caps with daily, weekly or monthly reset; team budgets not stated | Open + frontier models: Claude, GPT, Llama, Mistral, DeepSeek, Qwen, Kimi Routing to cheapest provider: Price sort or :floor tries the cheapest provider first | Pricing model: Plans from free to EUR 99/month plus a 3% to 15% markup |
| Requesty | EU processing option: EU endpoint hosted in Frankfurt, on every plan Zero data retention: Self-serve logs prompts for 30 days by default, can be disabled per key; org-wide ZDR on request | SSO: Enterprise: Okta, Azure AD, Google Workspace, OIDC Budgets: Spend limits per key; groups and team budgets listed | Open + frontier models: 600+ models incl. Claude, GPT, DeepSeek, Qwen, Llama Routing to cheapest provider: Failover, weighted and latency routing; automatic cheapest provider not stated | Pricing model: Provider cost plus 5% |
| Portkey | EU processing option: Data can be hosted in a region of your choice; VPC deployment on Enterprise. EU region not named Zero data retention: Per-request do-not-track; org-wide metrics-only logging on Enterprise | SSO: Enterprise Budgets: Per key and per workspace (Enterprise, select Pro); per team not stated | Open + frontier models: 1,600+ models incl. OpenAI, Anthropic, DeepSeek, Ollama Routing to cheapest provider: Fallbacks, load balancing, conditional routing; cost-based routing not stated | Pricing model: Free open source; Production from $49/month; Enterprise custom |
| LiteLLM | EU processing option: Self-host in your own EU infrastructure; hosted EU region not stated Zero data retention: Message logging can be turned off globally or per request | SSO: Okta, Google, Microsoft, OIDC, SAML; free up to 5 users, then Enterprise Budgets: Budgets per team, key, user and org | Open + frontier models: About 135 providers incl. OpenAI, Anthropic, DeepSeek, vLLM, Ollama Routing to cheapest provider: Documented cost-based routing strategy, plus latency and usage routing | Pricing model: Free open source; Enterprise license custom |
| OpenRouter | EU processing option: EU in-region routing on Business and Enterprise; account metadata may be in the US Zero data retention: ZDR routing on every plan; prompts not logged by default | SSO: SAML SSO and SCIM on Enterprise Budgets: Per workspace, key and member | Open + frontier models: 600+ models incl. Claude, GPT, DeepSeek Routing to cheapest provider: Price sort or :floor; default balances price, speed and reliability | Pricing model: Fee on credit purchases (5.5% Standard, 8% Business) |
| Kong AI Gateway | EU processing option: Self-host, or Konnect EU geo and dedicated cloud gateways in EU regions Zero data retention: Payload logging off by default; ZDR not stated | SSO: OIDC or SAML for Konnect on Enterprise Budgets: Token quotas and cost budgets per consumer group or team | Open + frontier models: Bring your own providers and keys, incl. self-hosted Ollama and vLLM Routing to cheapest provider: Lowest-usage routing by tokens or cost (AI Proxy Advanced, Enterprise) | Pricing model: Plus from $25/month plus usage; Enterprise custom |
| Cloudflare AI Gateway | EU processing option: Not stated; AI Gateway is listed as not compatible with Regional Services Zero data retention: Logging can be turned off per gateway or per request | SSO: Dashboard SSO (SAML) on all plans Budgets: Spend limits per gateway, scoped by provider, model or metadata such as team | Open + frontier models: Your own keys, or unified billing for OpenAI, Anthropic, Google; open models via Workers AI Routing to cheapest provider: Dynamic routes with fallbacks; price-based routing not stated | Pricing model: Core features free; 5% fee on unified-billing credits |
| Azure API Management | EU processing option: Deploy in EU regions such as Germany West Central and France Central Zero data retention: Not stated as ZDR; prompt logging is optional per API | SSO: Microsoft Entra ID for the developer portal Budgets: Token quotas per subscription or custom key; spend budgets not stated | Open + frontier models: Fronts models you deploy: Azure OpenAI, Foundry, Bedrock, Vertex, self-hosted Routing to cheapest provider: Weighted, priority and round-robin load balancing; cheapest routing not stated | Pricing model: Per gateway unit by tier; Consumption tier per call |
| AWS Bedrock | EU processing option: EU cross-Region inference profiles keep requests in EU Regions Zero data retention: Data retention mode none; not used for training; some models require retention | SSO: AWS IAM Identity Center Budgets: Cost tracking per application inference profile; hard team caps not stated | Open + frontier models: Anthropic, OpenAI, Meta, Mistral, DeepSeek, Amazon, Cohere Routing to cheapest provider: Prompt routing between two models of one family | Pricing model: Per token on demand; batch and provisioned throughput |
When to choose each gateway
Choose Standard Compute when you want lower AI spend and a clean security review
Pick it when people and agents across the company need Claude, GPT, Gemini and open models through one API, and you want routing to bring the cost down for you. Set a Europe rule for your company, a team or a single key, and every request, including retries and fallbacks, is served only by approved European inference endpoints. If no European endpoint can serve it, the request is refused before your prompt is sent. Every admin change is recorded with before and after values, you can export a record of every request for 12 months, and you sign the DPA in the dashboard.
Our pages: Security · Sub-processors · Pricing
Choose EUrouter when you want a European router with euro pricing
EUrouter describes itself as the European AI gateway: one OpenAI-compatible API with data kept in Europe by default, provider routing by price, and options to restrict routing to EEA-headquartered providers or a chosen country. It suits a European team that wants usage-based pricing in euros and per-key spend caps.
Choose Requesty when you want a managed router with a Frankfurt endpoint
Requesty offers an OpenAI-compatible router with an EU endpoint in Frankfurt on every plan, failover chains and usage-based billing at provider cost plus 5%. It fits a team that wants pay-as-you-go access to a large catalog without running its own gateway. Standard Compute vs Requesty for business.
Sources: EU · DPA · Pricing · Latency routing
Choose Portkey when observability and guardrails come first
Portkey positions itself as a production stack for generative AI builders, with logs, guardrails, prompt management and governance in one place. It suits an engineering team that brings its own provider keys and wants deep request observability. Standard Compute vs Portkey for business.
Sources: Pricing · Request logging · Budget limits · AI Gateway · Changelog, July 2025
Choose LiteLLM when you want to run the gateway yourself
LiteLLM is an open-source proxy you deploy in your own infrastructure, so data and keys stay where you put them. It has team budgets, cost-based routing and broad provider support. It fits a platform team ready to operate, patch and scale a gateway and hold its own provider contracts. Standard Compute vs LiteLLM for business.
Sources: Enterprise · Logging · SSO · Team budgets · Routing
Choose OpenRouter when catalog breadth matters most
OpenRouter offers one API to 600+ models with managed provider routing and fallbacks, and EU in-region routing on its Business and Enterprise plans. It suits a team that wants the widest model choice and is comfortable buying credits. Standard Compute as an enterprise alternative to OpenRouter.
Sources: Provider routing · Pricing · Enterprise
Choose Kong AI Gateway when AI traffic should sit in your API platform
Kong adds AI routing, quotas and policies to the Kong API gateway, self-hosted or in Konnect with an EU geo. It fits a company that already runs Kong for its APIs and wants one governance layer for AI traffic, with its own provider contracts. Standard Compute vs Kong AI Gateway.
Sources: AI Gateway · Konnect geos · AI Proxy Advanced · SSO · Pricing
Choose Cloudflare AI Gateway when you already run on Cloudflare
Cloudflare AI Gateway adds caching, rate limits, analytics and spend limits in front of the providers you already use, with the core features free. It fits a team on Cloudflare that wants visibility and control at the edge and holds its own provider keys. Standard Compute vs Cloudflare AI Gateway.
Sources: Data localization compatibility · Logging · Spend limits · Dynamic routing · Pricing
Choose Azure API Management when your company runs on Azure
Azure API Management governs and monitors AI backends inside the API gateway many Microsoft shops already run, with token quotas and load balancing across deployments. It fits a company standardised on Azure that will deploy and contract the models itself. Standard Compute vs Azure OpenAI.
Sources: Regions · AI gateway capabilities · LLM logging · Entra ID · Pricing
Choose AWS Bedrock when your workloads live in AWS
Bedrock gives AWS customers managed access to foundation models from several labs under their existing AWS agreement, IAM and EU Regions. It fits a team building applications inside AWS that is happy to manage quotas, profiles and cost tags in the AWS console. Standard Compute vs AWS Bedrock.
Sources: EU cross-Region inference · Data retention · Inference profiles · Prompt routing · Pricing
Buyer checklist for a European AI gateway
Ask every vendor these questions and ask for the answer in writing. Each one maps to a column above.
Where is each request processed, including retries and fallbacks?
A region option on the main call is common. Ask what happens when the first provider fails, and whether a request is refused when no European endpoint can serve it.
Is zero data retention enforced or only available?
Ask whether retention is off by default, who can switch it on, and whether a team or key can loosen a company setting.
Does SSO work with Microsoft Entra ID, and on which plan?
Many gateways keep SSO on an enterprise plan. Check whether employees can also reach coding tools such as Claude Code with their work account.
Can spend be capped per person, app and team?
Ask whether a budget stops requests at the limit or only reports on them, and at which levels limits can be set.
Does routing lower cost automatically?
Load balancing and fallbacks protect uptime. Ask whether the gateway also picks a cheaper model or provider for each step, and how it protects quality.
Which models are available inside Europe?
A large catalog worldwide can shrink sharply under a Europe restriction. Ask for the list of frontier and open models served by European endpoints.
What will the security review need?
Ask for a signed DPA, a public sub-processor list, an admin audit log and a per-request log export. Our security review pack shows the format we send.
Questions buyers ask
- What is the best AI gateway for European companies in 2026?
- For most European companies we recommend Standard Compute: it routes each step to the cheapest capable model and provider across frontier and open models, and adds a Europe rule, zero data retention, SAML SSO with Microsoft Entra ID, budgets per person and app, and an audit log. Self-hosted LiteLLM or Kong suits platform teams that want to run the gateway themselves, and AWS Bedrock or Azure API Management suits companies standardised on one cloud.
- Which AI gateway keeps data processing in Europe?
- Several publish a European option: Standard Compute has a Europe rule per company, team or key, EUrouter routes to EU-hosted endpoints by default, Requesty has a Frankfurt endpoint, OpenRouter offers EU in-region routing on Business and Enterprise, and Bedrock has EU cross-Region inference profiles. Self-hosted LiteLLM and Kong run wherever you deploy them. With a Standard Compute Europe rule, every request, including retries and fallbacks, is served only by approved European inference endpoints.
- What is a GDPR-ready LLM router?
- A router that lets you keep processing in Europe, turn off prompt retention, sign a DPA, see every sub-processor and prove who changed what. Standard Compute covers each of these: a Europe rule, zero data retention per company, team or key, a DPA you sign in the dashboard, a public sub-processor list with 14 days' notice of changes, and an admin audit log.
- Does Standard Compute support SSO with Microsoft Entra ID?
- Yes. Connect Microsoft Entra ID or any SAML identity provider so your people sign in to Standard Compute with their work accounts. Employees can also sign in to Claude Desktop and Claude Code with their Microsoft work account through our EU gateway, with access following your Entra groups through SCIM.
- How does an AI gateway lower costs?
- By sending simple steps to cheaper models and choosing the cheapest provider that is fast enough for each model. Standard Compute does this on every request, across frontier and open models, and gives every person and app a fixed monthly budget that never overruns.
- Is OpenRouter GDPR compliant?
- OpenRouter publishes EU in-region routing on its Business and Enterprise plans and zero data retention routing on every plan, and notes that account metadata may be processed in the US. Read its enterprise page and DPA for your own assessment, or compare Standard Compute as an enterprise alternative.
Related comparisons
How we compared: each rival fact comes from that vendor's own website or documentation, checked on 9 November 2026 and linked above. Where a vendor does not publish something, the table says not stated. Vendors change quickly, so confirm details with them. Corrections: contact@standardcompute.com.
Create a company account, set a Europe rule and invite your team. It takes a few minutes, and every request after that follows your rules.