Skip to content

    For enterprises, system integrators and cloud providers.Design partners

    Your AI Factory. On your hardware

    FlexAI's AI services on your hardware: Token Factory, AgentOS, Dedicated GPUs and an audit log.

    What your users can do

    Request
    curl -s https://eu.api.flex.ai/v1/chat/completions \
      -H "Authorization: Bearer $FLEXAI_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{
        "model": "DeepSeek-V4-Flash-0731",
        "messages": [{
          "role": "user",
          "content": "Say good morning to Europe in French, German, Dutch and Italian. One line each, nothing else."
        }]
      }'
    Response

    Bonjour l'Europe !

    Guten Morgen, Europa!

    Goedemorgen, Europa!

    Buongiorno, Europa!

    Open models behind one OpenAI-compatible API, metered per token. The call was captured from an EU workspace: EU endpoint, billed in euros.

    How your team operates it

    Your operators run the fleet beneath those services with Fleet Manager.

    1. 1
      Provision capacity

      Clusters across locations, on NVIDIA, AMD and Tenstorrent.

    2. 2
      Deploy models

      Onto the hardware you have available.

    3. 3
      Monitor and control

      Fleet capacity and service health.

    Capacity across clusters
    Fleet Manager clusters: cluster-a (NVIDIA A100, EU), cluster-b (AMD MI300X, US), cluster-c (NVIDIA H200, US) and cluster-d (Tenstorrent Galaxy), each with its GPUs, model cache and running models.
    One deployment configuration
    Fleet Manager deploy form: Hugging Face model zai-org/GLM-5.2-FP8, cluster cluster-c (NVIDIA), and its serving runtime.
    Model running
    Fleet Manager pod view: glm-5-2 running and in rotation on cluster-c, one hot replica on GPUs 0 to 7, ready.

    In productionGLM-5.2 running on FlexAI's US H200 cluster: one replica on 8 GPUs.

    75%
    faster time to launch

    For GPUs (bare metal or VM) and managed AI services.

    40%
    higher utilization

    Through tenancy and burst tiers that reduce stranded capacity.

    2×
    revenue streams

    Higher-margin managed AI services on top of GPUaaS revenue.

    Same stack that powers Token Factory

    OpenAI-compatible API, AgentOS, audit log, sensitivity-aware retrieval. Customers feel one platform.

    Deployment and hardware

    Scenario A: Customer has GPU infrastructure running

    • Deploy the AI Services layer on top
    • Fastest path
    • Weeks, not months

    Scenario B: Customer starting from bare metal

    • Set up Cloud Foundation first
    • Then the AI Services layer on top
    • Professional services included
    Model APIs and inference
    • NVIDIA
    • AMD
    • Tenstorrent
    GPU VMs and fine-tuning
    • NVIDIA
    • AMD

    Platform stack

    Products

    What your customers build and ship on.

    AI Services layer

    The managed runtime beneath the products.

    Cloud Foundation

    The control plane and compute pool underneath.

    Top layers deploy on your hardware

    Products

    What your customers build and ship on.

    • Token Factory: OpenAI-compatible inference
    • AgentOS + MCP
    • Dedicated GPUs: VMs, fine-tuning and production inference
    • Blueprint-based templates

    AI Services layer

    The managed runtime beneath the products.

    • Managed AI services (inference, fine-tuning, training, RAG)
    • Token and usage-based monetization
    • Blueprint-based templates
    • Governance, reliability, observability
    • HPC workloads via SLURM on Kubernetes

    Cloud Foundation

    The control plane and compute pool underneath.

    • Intent-driven control plane
    • GPUaaS and VM infrastructure
    • Mixed accelerator fleets: NVIDIA, AMD and Tenstorrent
    • Unified cloud pool with burst-to-cloud
    • S3-compatible object or local storage

    The compute is yours. AI Factory deploys these layers on it.

    SOC 2 Type II certifiedGDPR compliantPlus customer's own compliance frameworks (HIPAA, FedRAMP, IRAP, SAMA, RBI) where deployment topology supports it

    Talk to us

    Loading meeting scheduler…

    Who it's for

    Enterprises

    Private and hybrid AI clouds in your VPC, on-prem or air-gapped.

    See Enterprises →

    System integrators

    White-label AI infrastructure for your customers, with partner economics.

    See System integrators →

    Cloud providers

    Sell Token Factory APIs and Dedicated GPUs on your own GPU fleet.

    See Cloud providers →

    Frequently Asked Questions

    FlexAI's managed cloud runs these services on FlexAI's own GPUs. AI Factory deploys the same AI services on your own hardware: the compute is yours, the platform is FlexAI's.

    For AI Factory, yes. It deploys onto your hardware (Scenario A) or onto bare metal you stand up first with Cloud Foundation (Scenario B, professional services included). If you'd rather not own compute, use FlexAI's managed cloud (Token Factory and Dedicated GPUs) instead.

    Yes. AI Factory runs on Kubernetes and Slurm, supports fractional and time-sliced GPUs, GPU and network isolation, and integrates with your VPC and IaC (Pulumi).

    Any cloud, or on-prem. AI Factory runs mixed accelerator fleets (NVIDIA, AMD and Tenstorrent) with InfiniBand or RoCE-V2 networking and NVLink scale-up, multi-region pools, and burst-to-cloud for extra capacity. If you have a specific accelerator, ask us.

    AI Factory is available to design partners today. Talk to us about a deployment and we'll walk you through scope, timeline, and fit.

    Scope your first deployment

    Tell us about your hardware and the services you want to run on it.