Koyeb

Koyeb operates a serverless cloud platform that lets developers deploy containers, code from Git repositories and machine learning models without managing servers or orchestration. Workloads run in Cloud Hypervisor microVMs on bare metal servers equipped with CPUs, GPUs and accelerators from AMD, Intel and Nvidia, with Tenstorrent n300s accelerators also listed. The platform provides autoscaling, scale-to-zero, autohealing, zero-downtime deployments with health checks, a full service mesh with service discovery, a global edge network with CDN and load balancing, custom domains, native HTTP/2, HTTP/3, WebSocket and gRPC support, NVMe local storage, volumes and snapshots, real-time logs and metrics, and log export.

Beyond stateless services, Koyeb offers a fully managed serverless Postgres with pgvector for embedding storage and search, available in Washington D.C., Frankfurt and Singapore, and Koyeb Sandboxes — isolated environments intended for running AI agents and executing AI-generated code, managed through the CLI and API. Supported service types include web applications, APIs, MCP servers, AI agents, background workers, model serving, and model training and fine-tuning. Developers interact with the platform through a web console, CLI, REST API, and Terraform and Pulumi providers, with quickstarts for languages and frameworks including Node.js, Python, Go, Ruby, PHP, Java, Scala, Rust, Django, FastAPI, Next.js and Streamlit, plus integrations with third-party database services such as Aiven, CockroachDB, Crunchy Bridge, MongoDB Atlas, Neon, PlanetScale and Upstash.

In February 2026 Mistral AI announced its acquisition of Koyeb. Mistral said it would apply Koyeb's technology to its sandboxing feature for code generation models, to Mistral Compute, to more efficient GPU utilization, and to running its models on customer on-premises infrastructure. The Koyeb platform is to continue operating and become a component of Mistral Compute, with the team refocusing on enterprise customers and new sign-ups restricted to paid plans.

Founding story

Koyeb was founded in Paris in 2020 by three former employees of the French cloud provider Scaleway — Yann Léger, Edouard Bonlieu and Bastien Chatelard. The founders described starting the company to address the gap they saw between complex, expert-intensive infrastructure stacks such as Kubernetes and simple platforms that lacked flexibility and scale, aiming instead for a managed serverless experience built on their own technology and bare metal hardware.

Business model

Koyeb sells cloud infrastructure directly to developers and companies on a self-service, consumption basis, operating its own software stack on bare metal servers rather than reselling capacity from incumbent hyperscalers. Compute is billed per second by instance type, with published GPU rates ranging from $0.50/hour for an RTX-4000-SFF-ADA to $24.00/hour for an 8x H200 instance, plus managed serverless Postgres priced by instance size and $0.50 per GB-month of storage, and bandwidth beyond an included 1TB per month. Subscription tiers layer team, support and capacity features on top of compute: Pro at $29/month, Scale at $299/month, and a custom Enterprise tier starting at $1,000/month with private dedicated locations, SSO/RBAC, certifications and a 99.99% uptime SLA. Long-term reservations are offered at up to 50% off on-demand pricing.

Usage-based revenue from per-second billing of serverless compute instances (CPU, GPU and accelerator), managed serverless Postgres and storage, plus bandwidth overages, combined with recurring monthly subscription plans (Pro $29/month, Scale $299/month, Enterprise custom from $1,000/month) that include a compute allowance and additional platform, support and compliance features. Discounted long-term capacity reservations are also offered.

Traction

The website reports 100,000 developers, 10 datacenters and more than 50 deployment locations. Koyeb states customers use its GPUs for training and inference, and that its Sandboxes product can support launching up to thousands of AI agents per day for tasks such as code generation. The company raised $7 million in seed funding in 2023 and $8.6 million in total, and was acquired by Mistral AI in February 2026.

Latest developments

In February 2026 Mistral AI announced the acquisition of Koyeb, with a definitive agreement disclosed in Koyeb's 20 February 2026 changelog and closing subject to customary conditions; terms were not disclosed. The 16-person team (reported elsewhere as 13 people) was to join Mistral in March 2026 under CTO Timothée Lacroix to work on on-premises model deployment, GPU optimization and AI inference scaling, while the platform continues to run and is planned to become a core part of Mistral Compute. Recent product updates include manual scaling of services, Sandboxes access through the Koyeb CLI, Koyeb Skills for coding agents, 50% faster time-to-healthy deployments, improved scale-to-zero deployments, and a one-click deployment for the OpenClaw agent platform.