# Fantasti

> Fantasti is a GPU cloud for AI teams: NVIDIA GPU instances, clusters, CPU machines and storage behind one API and one bill.

Private Preview. We are not taking access requests yet, and no account is open.

The platform opens to its first cohort by invitation. Until your account opens, nothing is provisioned and nothing is billed.

The Fantasti Cloud Platform is built by Fantasti Technologies, the company behind Fantasti. Designed to be operated end to end by AI agents, under written rules and hard limits. The desks open with the first cohort. Until then Fantasti Technologies reads every request and answers it.

- Ask a question: https://fantasti.ai/contact
- If requests open, an approved request is invited by a sign-up link that is tied to the email address, valid for 72 hours and works once.
- API (Preview API): Not reachable yet. It answers with the first cohort. One client for the whole platform: Python, REST and a CLI over the same objects, ids and limits. The SDK and the CLI are published when the first cohort opens.
- MCP server (Early access): Not reachable yet. It answers with the first cohort. A remote MCP server that gives AI agents the Fantasti API as tools, inside the limits your admins set.
- Preview list prices. Nothing is billed before your account opens. In USD, last reviewed 2026-10-10. Metered by the second, billed hourly.
- Each page listed here is also served as text at its address plus ".md" (Home: https://fantasti.ai/index.md). All of them in one file: https://fantasti.ai/llms-full.txt

Status words used on this site:

- Private Preview: Opens with the first cohort. By request, if requests open.
- Early access: Being built for the first cohort. By request, if requests open. Details may change.
- By request: Quoted on reserved terms, if requests open. Not a release stage.
- Planned: On the roadmap, described as direction. No date.

## Products

- [GPU Instances](https://fantasti.ai/compute#instances): Virtual machines with one or eight NVIDIA GPUs, on demand, on spot or reserved. Private Preview.
- [CPU Instances](https://fantasti.ai/compute#cpu-instances): General-purpose virtual machines from 2 to 256 vCPU with 4 GiB of memory per vCPU. Private Preview.
- [GPU Clusters](https://fantasti.ai/clusters): Groups of 8-GPU nodes joined by InfiniBand on a single fabric, delivered as an isolated Kubernetes cluster or as plain machines. Private Preview.
- [Rack-scale](https://fantasti.ai/compute#rack-scale): GB300 NVL72 and GB200 NVL72 racks, 72 GPUs in one NVLink domain, on reserved terms. By request.
- [Spot](https://fantasti.ai/spot): GPU capacity at a market price, with an optional max price per GPU-hour. Private Preview.
- [Workspaces](https://fantasti.ai/workspaces): GPU development environments that start on one GPU and scale the same Ray code to an autoscaling GPU cluster. Early access.
- [Sandboxes](https://fantasti.ai/sandboxes): Isolated, short-lived CPU and GPU environments that AI agents create and end through the API. Early access.
- [Flows](https://fantasti.ai/flows): Python workflows that run step by step on Fantasti GPUs, versioned and scheduled, from an experiment to production. Planned.
- [Storage](https://fantasti.ai/storage): Volumes, shared filesystems and S3-compatible object storage in the same region as your compute. Private Preview.
- [Networking](https://fantasti.ai/networking): Private networks, public addresses, security groups, load balancers and private DNS, plus the InfiniBand fabric for clusters. Private Preview.
- [Batch Inference](https://fantasti.ai/batch-inference): Offline inference over a dataset on spot or on-demand capacity, with checkpoints, retries and a max price. Private Preview.
- [Serverless](https://fantasti.ai/serverless): Run a container on a GPU as a job that finishes or an endpoint that serves requests, with no machines to manage. Early access.
- [Integrations](https://fantasti.ai/integrations): Run Kubernetes tools, Ray, SkyPilot or NVIDIA OSMO on your Fantasti cluster the way their docs describe. Early access.
- [API](https://fantasti.ai/api): One client for the whole platform: Python, REST and a CLI over the same objects, ids and limits. Private Preview.
- [MCP server](https://fantasti.ai/mcp): A remote MCP server that gives AI agents the Fantasti API as tools, inside the limits your admins set. Early access.
- [Console and one bill](https://fantasti.ai/pricing#one-bill): One console and one statement across every product, split by cost center. Private Preview.
- [Orchestrator](https://fantasti.ai/orchestrator): The control plane that checks policy, finds capacity and quota, places each workload on one fabric, works the spot market inside your max price, recovers work after a stop and meters it onto one statement. Private Preview.
- [Observability](https://fantasti.ai/observability): Metrics, logs and traces for everything on Fantasti, with prebuilt dashboards. Early access.
- [MLflow](https://fantasti.ai/integrations#mlflow): Managed MLflow tracking and model registry, wired into Workspaces, Flows, Batch Inference, Serverless jobs and Clusters. Early access.
- [Applications](https://fantasti.ai/integrations#applications): One-click applications installed on your cluster: notebooks, inference servers, vector databases, UIs. Planned.

## GPUs and preview list prices

Price list 1 · preview · the list in force when you use it applies. Reviewed 2026-10-10.

- [NVIDIA B300](https://fantasti.ai/gpus/b300): $10.45 per GPU-hour on demand. Spot: From $1.09. Private Preview.
- [NVIDIA B200](https://fantasti.ai/gpus/b200): $9.35 per GPU-hour on demand. Spot: From $1.09. Private Preview.
- [NVIDIA H200 SXM](https://fantasti.ai/gpus/h200-sxm): $5.94 per GPU-hour on demand. Spot: From $0.87. Private Preview.
- [NVIDIA H100 SXM5](https://fantasti.ai/gpus/h100-sxm5): $5.40 per GPU-hour on demand. Spot: From $0.87. Private Preview.
- [NVIDIA RTX PRO 6000 Blackwell Server Edition](https://fantasti.ai/gpus/rtx-pro-6000): $2.16 per GPU-hour on demand. Spot: From $0.87. Private Preview.
- [NVIDIA L40S](https://fantasti.ai/gpus/l40s): $1.86 per GPU-hour on demand. List price for the base shape: 1 GPU, 8 vCPU, 32 GiB. Larger shapes are quoted. No spot. Private Preview.
- [NVIDIA GB300 NVL72](https://fantasti.ai/gpus/gb300-nvl72): Reserved, by quote. By request.
- [NVIDIA GB200 NVL72](https://fantasti.ai/gpus/gb200-nvl72): Reserved, by quote. By request.
- [CPU Instances](https://fantasti.ai/compute#cpu-instances): $0.0396 per vCPU-hour. Shapes from 2 to 256 vCPU.

"From" is the lowest spot price for that GPU. The spot price moves with supply and demand.
Reserved rates are quoted per term.

## Pages

- [Home](https://fantasti.ai/)
- [Platform](https://fantasti.ai/platform): Compute, development environments, storage, networking and operations on one API, one set of policies and one statement.
- [Compute](https://fantasti.ai/compute): Virtual machines with one or eight NVIDIA GPUs, CPU machines from 2 to 256 vCPU and NVL72 racks on reserved terms.
- [Agents](https://fantasti.ai/agents): Fantasti is designed to be operated by AI agents under written rules. High-stakes decisions pass a second, senior review.
- [GPU catalog](https://fantasti.ai/gpus): Specifications and preview list prices for NVIDIA B300, B200, H200 SXM, H100 SXM5, RTX PRO 6000 and L40S. Reviewed 2026-10-10.
- [Pricing](https://fantasti.ai/pricing): Preview list prices per GPU-hour, spot with a max price, reserved terms and three ways to pay. Reviewed 2026-10-10.
- [Support](https://fantasti.ai/support): Five support plans. Set up so that AI agents take every case and say they are AI. Escalations go to senior review.
- [Community](https://fantasti.ai/community): Planned.
- [Feedback](https://fantasti.ai/feedback): Planned.
- [Previews](https://fantasti.ai/previews): Previews give a team new Fantasti features while they are being built. Each one states its stage, billing, support and what may change.
- [Developer Program](https://fantasti.ai/developers): Planned.
- [Changelog](https://fantasti.ai/changelog): Dated entries for every change to the Fantasti Cloud Platform, by product and release stage. RSS and JSON feeds.
- [Research](https://fantasti.ai/research): Research in progress: network-based GPU virtualization, delivery protocols for GPU workloads, placement and interruption prediction.
- [Trust and security](https://fantasti.ai/trust): What you secure, what Fantasti secures and what the infrastructure layer secures, with a status on every control.
- [Company](https://fantasti.ai/company): Fantasti Technologies is the company behind Fantasti, a GPU cloud for AI teams designed to be operated by AI agents under written rules.
- [Contact](https://fantasti.ai/contact): Ask a question about the Fantasti Cloud Platform, or send a partnership proposal.

## Legal

- [Legal](https://fantasti.ai/legal)
- [Terms of Service](https://fantasti.ai/legal/terms)
- [Privacy Policy](https://fantasti.ai/legal/privacy)
- [Acceptable Use Policy](https://fantasti.ai/legal/aup)
- [Service Level Agreement](https://fantasti.ai/legal/sla)
- [Cluster requirements v1](https://fantasti.ai/reserved/requirements/v1)

## Feeds

- https://fantasti.ai/changelog/index.json
- https://fantasti.ai/changelog/rss.xml
- https://fantasti.ai/sitemap.xml