Mistral is giving enterprise customers more control over where AI requests are processed, but the new controls stop short of full platform-wide data residency. The Decoder reports that Mistral has made regional inference generally available, letting customers route requests through either a European endpoint, api.eu.mistral.ai, or a US endpoint, api.us.mistral.ai. According to The Decoder, requests sent through those endpoints are processed in the selected region. That is the core change for customers in regulated sectors that need tighter control over where model inputs are handled. The outlet says Mistral is presenting the move as part of a broader European AI sovereignty strategy. The control comes with a price and with scope limits. The Decoder reports that regional routing costs 10% more than standard pricing. Customers using Mistral’s default endpoint receive no guarantee about where a request is processed, according to the report. The feature coverage is narrower than the regional label may suggest. Among Mistral’s add-on tools, The Decoder says only function calling works with the regional endpoints. Agents, batch processing, and file management are not available through the regional addresses. Model availability also varies by region. The Decoder reports that Mistral does not publish a fixed list of supported models per regional endpoint; customers have to query each endpoint to see what is available. That makes implementation a technical validation task rather than a simple account-level switch. The data boundary is also not total. The Decoder, citing Mistral documentation, reports that account settings, API keys, billing, and usage statistics can still be processed outside the chosen region. The report also says Mistral’s blog post mentions limited, secured transfers to subcontractors outside the selected region. In practice, the regional commitment applies to the compute step, not necessarily to every platform operation around it. There is a separate setting for whether requests are stored or logged after processing. The Decoder identifies that setting as Zero Data Retention. That distinction matters: regional inference governs where a request is processed, while retention settings govern what happens to request data afterward. Mistral is also selling a Priority Tier, currently in open beta, according to The Decoder. The tier gives paying customers priority in the shared infrastructure queue when traffic is heavy. The outlet says Mistral is targeting latency-sensitive systems, including customer service chatbots and production systems on factory floors. The Priority Tier includes a 99.5% uptime service-level agreement, The Decoder reports, equivalent to roughly three and a half hours of downtime per month. The broader message for enterprise buyers is that Mistral is moving beyond model access into operational guarantees — but those guarantees are specific, priced, and incomplete rather than blanket sovereignty or dedicated infrastructure. Who benefits: Mistral benefits by turning enterprise governance and latency requirements into paid platform features. Customers with straightforward model-call workloads may benefit most, because regional endpoints appear to fit simple inference better than agents or file-based workflows. Who's exposed: Customers that assumed “EU processing” meant full platform-level data residency are exposed to implementation risk. Teams using agents, batch processing, or file management will need to verify whether their workflows qualify for the regional guarantees described by The Decoder.