Setting Up a Fully Hosted Cloud Studio

Why Go Fully Cloud

The traditional model for building a VFX studio starts with a large capital outlay: workstations, render nodes, high-speed storage, networking gear, a server room with cooling, and the lease on a physical space big enough to house it all. For a 20-artist studio, that initial hardware investment alone can run $400,000 to $800,000 before a single frame is rendered. A fully cloud-hosted studio replaces all of that with monthly operational expenditure, fundamentally changing the economics of starting and running a VFX business.

But the financial argument is only part of the story. A cloud-native studio gains four structural advantages that are difficult or impossible to replicate with on-prem infrastructure.

Zero Capital Expenditure

Every piece of hardware in a traditional studio is a depreciating asset. A $7,000 workstation purchased today will be worth $2,000 in three years and will need replacing in four to five. A cloud studio converts this capital expenditure into a predictable monthly operating expense. There is no depreciation schedule, no hardware disposal costs, and no risk of being stuck with outdated equipment when a new GPU generation launches. If NVIDIA releases a new architecture tomorrow, your cloud provider upgrades their fleet and you benefit immediately -- no forklift upgrade required.

Global Talent Access

A cloud studio is not tied to a geographic location. An artist in Vancouver, a compositor in London, and a lighting TD in Mumbai can all work on the same project, accessing the same storage and the same tools, with identical performance. This is not remote access bolted onto an on-prem studio -- it is a fundamentally location-independent operation. You hire the best artists regardless of where they live, and you never pay relocation costs or compete for talent in overheated local markets like Los Angeles or Vancouver.

Instant Scaling

When a new project arrives with a tight deadline, a cloud studio can add 10 more workstations in an afternoon. When the project wraps, those workstations disappear from the bill. On-prem studios face an agonizing choice during growth periods: buy more hardware (expensive, slow to procure, may sit idle later) or turn down work. Cloud infrastructure eliminates this constraint entirely. Your studio's capacity is limited only by your budget, not by the number of machines in your server room.

Disaster Resilience

A cloud studio has no single physical point of failure. If your office floods, burns, or loses power for a week, every artist simply continues working from home or a coffee shop. Your data lives in multiple availability zones with automatic replication. There is no backup tape to retrieve, no disaster recovery site to fail over to -- the production environment is the disaster recovery environment. For studios working on high-value content with hard delivery dates, this resilience is not a luxury. It is insurance against the catastrophic scenario of losing a server room mid-production.

Virtual Workstations

The virtual workstation is the foundation of a cloud studio. Each artist gets a cloud-hosted machine with a dedicated GPU, and a streaming protocol sends the display output to their local screen in real time. When configured correctly, the experience is nearly indistinguishable from sitting in front of a physical workstation. When configured poorly, it is a laggy, frustrating mess that tanks productivity. The difference comes down to three factors: the streaming protocol, the GPU instance type, and the network path between the artist and the data center.

Streaming Protocol Comparison

The streaming protocol encodes the workstation's display output into a video stream, sends it to the artist's local device, and relays keyboard and mouse input back to the cloud machine. The three major options each have distinct strengths.

HP Anyware (formerly Teradici PCoIP) is the industry standard for VFX virtual workstations. Its hardware-accelerated encoding delivers the lowest latency and the most accurate color reproduction, which matters for color grading and compositing where artists need to trust what they see on screen. The 4:4:4 chroma subsampling at 10-bit depth means no color information is lost in the streaming process -- the artist sees exactly what the GPU rendered. The downside is cost: the per-seat license plus the requirement for specific GPU instance types makes it the most expensive option.

Parsec has gained significant traction in the VFX industry because it is dead simple to set up and performs surprisingly well for most workflows. It runs on commodity hardware, does not require specialized GPU instance types, and the software-based H.265 encoding is efficient enough that an artist with a 50Mbps internet connection gets a smooth experience. The trade-off is that Parsec does not match HP Anyware's color accuracy for critical finishing work, and its USB device passthrough is more limited -- a potential issue for artists who rely on Wacom tablets with pressure sensitivity and custom button mappings.

NICE DCV is Amazon's own remote display protocol, included free with any EC2 instance. For studios already committed to AWS, it eliminates the streaming protocol license cost entirely. DCV's QUIC-based transport adapts well to variable network conditions, and recent versions have added 4:4:4 color support. The limitation is that it only runs on AWS -- if you want multi-cloud flexibility or plan to use GCP or Azure, DCV is not an option.

GPU Instance Types

The GPU attached to your cloud workstation determines what software you can run and how responsive it feels. For VFX work, you need instances with professional-grade GPUs that support certified drivers for DCC applications.

• AWS G5 instances (NVIDIA A10G). The workhorse for most VFX cloud workstations. 24GB VRAM, ray tracing cores, hardware encoding for streaming protocols. A g5.4xlarge (16 vCPUs, 64GB RAM, 1x A10G) runs approximately $1.01/hour on-demand or $0.40/hour with a 1-year reserved instance. Suitable for Nuke compositing, Maya animation, and Houdini modeling.
• AWS G6 instances (NVIDIA L4). The newer generation with better power efficiency and improved ray tracing performance. 24GB VRAM with Ada Lovelace architecture. Pricing is similar to G5 but with roughly 20-30% better performance per dollar for GPU-accelerated workloads.
• AWS P4d/P5 instances (NVIDIA A100/H100). Overkill for interactive workstation use, but necessary for GPU rendering or machine learning workloads. 40-80GB VRAM per GPU. These are render farm instances, not artist workstations -- a single p4d.24xlarge costs $32.77/hour.
• GCP G2 instances (NVIDIA L4). Google's equivalent to AWS G6. Competitive pricing and good integration with Google Cloud Storage. A g2-standard-16 with 1x L4 GPU runs approximately $0.95/hour on-demand.
• Azure NVads A10 v5. Microsoft's VFX workstation instance. Supports fractional GPU allocation, allowing studios to share a single A10 across multiple lighter workloads (editorial, review) while giving compositors a full dedicated GPU.

Latency Requirements by Workflow

Not all VFX work has the same latency sensitivity. Understanding which tasks tolerate higher latency helps you make smarter decisions about when cloud workstations are viable and when artists need to be closer to the data center.

Cloud Storage Architecture

Storage is the most complex layer of a cloud VFX studio because VFX data has unusual characteristics: files are enormous (multi-layer EXRs can be 200MB each), access patterns are highly sequential (reading frame sequences), data volumes grow rapidly (a single project can generate 50TB+), and older data needs to be retained for years but is rarely accessed. A well-designed cloud storage architecture uses multiple tiers to balance performance and cost.

Hot / Warm / Cold: The Three-Tier Model

Every cloud-native VFX studio should organize storage into three tiers based on access frequency and performance requirements.

Hot Tier -- Active Project Storage. This is where artists work. It holds the current project's shot files, textures, caches, and render outputs that artists are actively reading and writing. In AWS, this means EBS (Elastic Block Store) volumes attached directly to workstation instances for local-speed access, backed by EFS (Elastic File System) or FSx for Lustre for shared storage that multiple workstations and render nodes access simultaneously. Hot storage is expensive -- $0.08-0.16/GB/month for EBS gp3 volumes -- but the performance is non-negotiable for interactive work. A typical 20-artist studio needs 20-40TB of hot storage.

Warm Tier -- Recent Project Archive. Projects completed in the last 6-12 months live here. The data is not actively worked on but needs to be accessible within minutes for revisions, client change requests, or reference by artists working on sequels or related projects. S3 Standard or S3 Infrequent Access is the right choice: $0.023/TB/month for Standard, $0.0125/TB/month for Infrequent Access. The data can be accessed immediately but at lower throughput than hot storage. A studio with 3-5 recently completed projects might have 100-200TB in the warm tier.

Cold Tier -- Long-Term Archive. Everything older than 12 months goes here. Studio contracts typically require retaining project data for 3-7 years, and some studios keep data indefinitely for potential sequels or derivative works. S3 Glacier Instant Retrieval ($0.004/TB/month) provides millisecond access when needed, while S3 Glacier Deep Archive ($0.00099/TB/month) is for data that will almost never be accessed but must be retained. A studio that has been operating for several years might have 500TB-2PB in cold storage, costing as little as $500-2,000/month.

Data Flow Across Tiers

Understanding how data moves through the tiers is critical for designing an efficient architecture. The flow follows a predictable lifecycle.

When a new project begins, plate scans and editorial media are ingested from the client (typically via Aspera or Signiant transfer) directly into hot storage. As artists work, they generate scene files, caches, and intermediate renders that also live in hot storage. Completed renders and approved versions are published to the production tracking system and review media is generated and pushed to the review platform.

When a project wraps, a pipeline script migrates the project from hot to warm storage. The migration preserves the full directory structure but moves the data from high-performance EFS/FSx to cost-effective S3 Standard. If a revision request comes in, the specific shots are promoted back to hot storage, worked on, then demoted again.

After 12 months with no access, lifecycle policies automatically transition the data from warm to cold storage (Glacier). The transition is logged in the studio's asset management system so the data can be located and retrieved if needed years later.

Cloud Rendering

Rendering is where a cloud studio's elastic capacity truly shines. Instead of owning a fixed render farm that is either underutilized or overwhelmed, you spin up exactly the compute you need, run the job, and shut it down. A render that would take 200 hours on a 10-node on-prem farm can be completed in 2 hours on 1,000 cloud instances -- and you only pay for those 2 hours.

Because the render farm and the workstations live in the same cloud region, data transfer between artists and the farm is essentially instantaneous -- no staging, no syncing, no waiting for uploads. This is a major advantage of fully-cloud over hybrid architectures, where syncing render inputs to the cloud can add hours of lead time to every job.

Render Management Platforms

AWS Deadline Cloud is the most mature option for AWS-native studios. It is the cloud evolution of Thinkbox Deadline, the render manager used by the majority of VFX studios worldwide. Deadline Cloud handles fleet management automatically: you define your compute requirements (instance type, max count, budget cap), submit jobs through the same interface artists already know, and the service provisions spot instances, distributes tasks, re-queues interrupted frames, and terminates instances when the queue drains. It supports Maya, Houdini, Nuke, Blender, and most other DCC renderers out of the box. The managed service costs $0.016 per worker-hour on top of EC2 instance costs.

Conductor takes a different approach -- it is a cloud-native render service, not a self-managed farm. You submit jobs to Conductor's platform, and they handle all the infrastructure. You never manage instances, AMIs, or scaling policies. Conductor supports Google Cloud and AWS, and pricing is per-core-hour with no separate infrastructure management overhead. This simplicity comes at a premium -- Conductor's effective cost per core-hour is higher than self-managed Deadline -- but for studios without dedicated cloud infrastructure engineers, the time savings are substantial.

Spot Instances: 60-90% Savings

Spot instances (called Preemptible VMs on GCP) are excess cloud capacity sold at a steep discount -- typically 60-90% off on-demand pricing. The catch is that the cloud provider can reclaim them with as little as 2 minutes notice. For rendering, this is almost always an acceptable trade-off.

When a spot instance is reclaimed, the render manager detects the lost worker and re-queues the interrupted frame. The frame renders again on a different instance, adding perhaps 10-20 minutes of redundant compute. Across a farm of hundreds of instances running thousands of frames, spot interruptions add roughly 2-5% to total render time -- a trivial cost compared to the 60-90% price reduction.

Key configuration decisions for spot rendering:

• Diversify instance types. Do not request only c6i.8xlarge. Request a pool of equivalent types (c6i.8xlarge, c6a.8xlarge, c5.9xlarge, m6i.8xlarge) so the spot allocator has more capacity to draw from, reducing interruption rates.
• Set a max price. Configure your spot requests with a maximum price of 50-60% of on-demand. If spot prices spike above that threshold, the system falls back to on-demand for critical jobs or waits for prices to drop for non-urgent work.
• Use checkpointing for long frames. If individual frames take more than 30 minutes to render, enable render checkpointing so that an interrupted frame does not restart from scratch. Most modern renderers (Arnold, V-Ray, RenderMan) support this.

GPU vs CPU Rendering in the Cloud

The GPU vs CPU rendering decision takes on a different dimension in the cloud because pricing scales differently than on-prem hardware.

For most cloud studios, CPU rendering on spot instances remains the most cost-effective approach for production rendering. GPU rendering makes sense for look development (where interactive feedback is valuable) and for specific renderers like Redshift that are GPU-only. The key advantage of CPU rendering in the cloud is availability: there are far more CPU spot instances available than GPU instances, which means less risk of being unable to get capacity during crunch periods.

Collaboration Tools

A cloud studio without strong collaboration tools is just a collection of remote artists working in isolation. The collaboration layer is what makes a distributed team function as a studio -- it provides the shared context, feedback loops, and production visibility that replace the hallway conversations and over-the-shoulder reviews of a physical office.

Production Tracking

ShotGrid (Autodesk) remains the dominant production tracking platform in VFX. It manages shot status, artist assignments, version history, and review workflows. For a cloud studio, ShotGrid's cloud-hosted option is essential -- it provides a single source of truth accessible to every artist regardless of location. ShotGrid's API enables deep integration with your pipeline, allowing automated status updates when renders complete, versions are published, or reviews are approved. Pricing is approximately $33/user/month.

ftrack is the main alternative to ShotGrid, with a cleaner interface and a more modern API. Its cloud-hosted version includes built-in review tools, reducing the need for a separate review platform. ftrack's custom attribute system is more flexible than ShotGrid's for studios with non-standard workflows. Pricing starts at $25/user/month for the Studio tier.

Review Platforms

Daily reviews and client approvals are the heartbeat of VFX production. In a cloud studio, review tools must handle high-resolution playback, frame-accurate annotation, and secure client access without requiring viewers to download files.

• Frame.io (Adobe). The leading cloud review platform. Supports 4K playback, frame-accurate comments, drawing annotations, and comparison tools. Its camera-to-cloud workflow is increasingly relevant for on-set VFX. The Teams plan ($15/user/month) includes 250GB of storage per user and supports up to 4K ProRes review media.
• Moxion. Purpose-built for feature film dailies and secure studio review. Moxion differentiates on security -- it supports forensic watermarking, DRM-protected playback, and TPN-compliant delivery. It is the platform of choice for studios working on major theatrical releases where content leaks are a material risk. Pricing is per-project and typically negotiated directly.
• ShotGrid Review / ftrack Review. Both production tracking platforms include built-in review tools. While not as feature-rich as dedicated platforms like Frame.io, they have the advantage of being integrated directly into the production database, so review notes automatically link to the correct shot, version, and artist. For studios that want to minimize the number of tools artists need to learn, the built-in review in your tracking platform is a pragmatic choice.

Communication and Real-Time Collaboration

Beyond production tracking and review, a cloud studio needs real-time communication tools that replicate the spontaneous collaboration of a shared office. Slack or Microsoft Teams provide text channels organized by project, department, and topic. Zoom or Google Meet handle scheduled meetings and dailies.

The most impactful pattern for cloud VFX studios is leaning into asynchronous workflows wherever possible. When your team spans multiple time zones, requiring everyone to be online simultaneously for reviews and handoffs is impractical. Use review tools that allow supervisors to leave frame-specific notes that artists pick up when they start their day. Reserve synchronous sessions for creative direction discussions, kickoffs, and complex problem-solving that benefits from live interaction.

Security & TPN

Security is the area where cloud studios face the most skepticism from major studios and distributors. The concern is understandable: when your artists work from home on personal internet connections, and your data lives on shared cloud infrastructure, how do you demonstrate the same level of control as a locked-down facility with badge readers and camera surveillance? The answer is that cloud security is different from physical security, but when implemented correctly, it can be equally rigorous -- and in many ways superior.

VPC Isolation

The network architecture of a cloud studio is built on VPCs -- isolated virtual networks that provide the same logical separation as physically separate network segments. A well-designed VPC architecture for a VFX studio includes:

• Production VPC. Contains artist workstations, shared storage (EFS/FSx), and render farm instances. All production data lives here. Ingress is limited to streaming protocol ports (TCP 4172 for PCoIP, or TCP 8443 for DCV) and VPN connections for pipeline tools. No direct internet access for workstation instances.
• Management VPC. Contains pipeline services, license servers, production tracking integrations, and monitoring tools. Peered to the Production VPC with restricted access rules.
• DMZ VPC. Handles external-facing services: client review platform access, file transfer endpoints (Aspera/Signiant), and VPN termination points. This VPC has internet access but is strictly isolated from production data through security groups and NACLs.
• Per-project isolation. Within the Production VPC, use security groups and IAM policies to isolate projects from each other. An artist assigned to Project A cannot access Project B's storage or workstations. For the most sensitive content, use separate AWS accounts per project for absolute isolation.

Encryption

A cloud studio should encrypt everything, everywhere. Data encrypted at rest using AES-256 on all EBS volumes, S3 buckets, and EFS file systems. Data encrypted in transit using TLS 1.3 for all connections. The streaming protocol between the cloud workstation and the artist's local display is also encrypted, so even if someone intercepted the network stream, they would see encrypted data, not video frames. Use AWS KMS or your provider's key management service to manage encryption keys, and rotate keys on a regular schedule.

IAM and Access Control

Every artist authenticates through a centralized identity provider (Okta, Azure AD, or AWS SSO) with mandatory multi-factor authentication. Access to specific projects, storage buckets, and workstations is controlled through IAM policies that follow the principle of least privilege -- artists can access only the projects they are assigned to. When an artist finishes a project or leaves the studio, their access is revoked immediately through the identity provider, and their cloud workstation is terminated.

Audit Trails

Every action is logged: who accessed which workstation, when they logged in and out, which files they opened, which S3 objects they read or wrote. AWS CloudTrail, VPC Flow Logs, and workstation session logs provide a complete audit trail that can be presented to TPN assessors or client security teams on demand. Store audit logs in a separate, tamper-proof account that production users cannot modify or delete.

TPN Compliance in the Cloud

The Trusted Partner Network (TPN) assessment evaluates vendors on content security practices. TPN does not prohibit cloud infrastructure, but it requires equivalent controls to what a physical facility would provide. For a cloud studio, the key controls are:

• No content on local devices. Virtual workstations stream pixels only. Artists cannot download, copy, or screenshot content. HP Anyware and NICE DCV both support disabling local clipboard, file transfer, and print functionality. This is actually stronger security than a physical office, where an artist could photograph a monitor.
• Physical security delegation. Cloud providers like AWS operate SOC 2 Type II certified data centers with 24/7 physical security, biometric access, and CCTV. Document your provider's certifications in your TPN assessment -- they typically exceed what a small studio's own server room provides.
• Network segmentation. Demonstrate per-project VPC isolation equivalent to separate physical networks.
• Comprehensive logging. Show that every access is logged and auditable. Cloud logging is often more complete than on-prem logging because cloud APIs capture every action by default.

Cost Analysis

The ultimate question for any studio considering a fully cloud-hosted operation: what does it actually cost? The following breakdown models a 20-artist cloud studio running a typical VFX project load -- 2-3 concurrent projects, moderate rendering volume, and standard collaboration tools.

Monthly Cost Breakdown: 20-Artist Cloud Studio

Costs based on AWS US-East pricing as of March 2026. DCC software licenses (Nuke, Maya, Houdini, etc.) excluded as they are identical regardless of infrastructure model. Render compute is highly variable -- crunch months can be 3-5x the baseline shown here.

Cloud vs On-Prem: 20-Artist Comparison

How does the cloud studio stack up against the traditional model? The following table compares the fully cloud studio above to an equivalent 20-artist on-prem studio.

* On-prem monthly cost includes hardware amortization over 4 years, facility costs (power, cooling, lease), and 1 IT FTE. Excludes upfront investment which is shown separately.

The three-year total cost of ownership is remarkably similar between the two models. The cloud studio costs slightly less when you factor in the on-prem capital expenditure, but the on-prem studio has lower monthly operating costs once the hardware is paid for. The deciding factors are rarely financial -- they are operational. Cloud wins on flexibility, speed, and remote work support. On-prem wins on latency for interactive work and cost predictability.

Getting Started

Launching a fully cloud-hosted studio is a project that takes 2-4 weeks for a basic setup and 2-3 months for a production-hardened deployment with full pipeline integration. Here is the step-by-step sequence.

Phase 1: Foundation (Week 1-2)

1. Choose your primary cloud provider. AWS is the most common choice for VFX due to Deadline Cloud, the broadest GPU instance selection, and the most VFX-specific documentation. GCP is competitive on pricing and has excellent GPU availability. Azure is best if your studio is Microsoft-centric. Pick one and commit -- multi-cloud adds complexity that is rarely justified for studios under 50 artists.
2. Set up your account structure. Create separate accounts or projects for Production, Management, and Billing. Enable CloudTrail (AWS) or Cloud Audit Logs (GCP) from day one. You will need these logs for TPN assessment and incident response.
3. Design your VPC and network architecture. Create production, management, and DMZ VPCs with peering connections. Define security groups for workstations, render nodes, storage, and management services. Document everything -- your TPN assessor will want to see network diagrams.
4. Deploy identity management. Set up AWS SSO or integrate with your existing identity provider (Okta, Azure AD). Configure MFA for all users. Define IAM roles for artists, supervisors, pipeline TDs, and administrators with appropriate permission boundaries.
5. Provision initial storage. Create your FSx for Lustre or EFS file system for hot storage. Set up S3 buckets for warm and cold tiers with lifecycle policies. Test throughput by writing and reading a representative EXR sequence.

Phase 2: Workstations and Pipeline (Week 2-3)

1. Build your workstation AMI. Install your DCC applications (Nuke, Maya, Houdini, etc.), configure license server connections, install the streaming protocol agent (HP Anyware, Parsec, or DCV), and configure storage mounts. Test the image thoroughly, then snapshot it as your base AMI. Every artist workstation will be launched from this image.
2. Launch pilot workstations. Spin up 3-5 workstations and have artists test real workflows. Measure latency, verify color accuracy, test USB device passthrough (Wacom tablets), and confirm that all DCC tools perform acceptably. Iterate on the AMI based on feedback.
3. Set up render infrastructure. Configure Deadline Cloud with worker fleets, job queues, and auto-scaling policies. Build render worker AMIs with all required renderers and plugins. Test with a representative render job.
4. Deploy production tracking. Set up ShotGrid or ftrack in the cloud. Configure pipeline integrations: publish tools, version creation, status automation, review media generation.
5. Configure review platform. Set up Frame.io or your chosen review tool. Integrate with production tracking so that published versions automatically generate review media. Test the full loop: artist publishes, review media appears, supervisor leaves notes, notes appear in the artist's tracking dashboard.

Phase 3: Hardening (Week 3-4)

1. Security audit. Review all security groups, IAM policies, and encryption settings. Verify that workstations cannot access the internet directly. Confirm that clipboard, file transfer, and USB storage are disabled on virtual workstations. Run a vulnerability scan on all exposed endpoints.
2. Backup and disaster recovery. Configure automated snapshots of EBS volumes and EFS file systems. Set up cross-region replication for critical S3 buckets. Document recovery procedures and test them by simulating a failure scenario.
3. Monitoring and alerts. Deploy CloudWatch dashboards for workstation utilization, storage capacity, render queue depth, and cost tracking. Set up alerts for anomalies: unexpected cost spikes, storage nearing capacity, workstation sessions running outside business hours.
4. Cost controls. Configure AWS Budgets with monthly caps and forecasted overage warnings. Set up auto-shutdown policies for workstations outside business hours (accounting for time zones). Review Reserved Instance recommendations and commit to 1-year reservations for predictable workloads.
5. Documentation and onboarding. Write artist onboarding guides (how to connect, where to find files, how to submit renders), IT runbooks (how to add a new artist, how to respond to a security incident, how to scale render capacity), and architecture documents for TPN assessment.