Key takeaways
- Enterprise-grade platform with VPC/BYOC deployment across AWS, GCP, Azure, and Oracle
- GPU support with H100s at $2.74/hour — 62% cheaper than major cloud providers
- Flexible ephemeral or persistent modes with microVM and gVisor isolation
FAQ
What is Northflank?
Northflank is a developer platform that provides secure microVM sandboxes for running untrusted code at scale, with options to deploy in their cloud or your own VPC.
How much does Northflank cost?
CPU costs $0.01667/vCPU-hour, memory $0.00833/GB-hour. GPUs range from $0.80-$3.14/hour (H100s at $2.74/hour). Free sandbox tier available.
Who competes with Northflank?
E2B, Modal, Daytona, Runloop, and Fly.io Sprites compete in the AI sandbox space. Northflank differentiates with VPC deployment and GPU support.
Executive Summary
Northflank is an enterprise-grade developer platform that provides secure microVM sandboxes for running AI agent workloads, with the unique ability to deploy in their managed cloud or your own VPC across AWS, GCP, Azure, and Oracle. The platform has been running millions of containers monthly since 2021 and is trusted by companies like Sentry.
| Attribute | Value |
|---|---|
| Company | Northflank |
| Founded | 2019 |
| Funding | $24M+ |
| Developers | 50k+ in production |
| Headquarters | London, UK |
Product Overview
Northflank provides secure sandboxed environments for running untrusted AI-generated code at scale using microVMs with hardware-level or gVisor isolation. Unlike pure-play sandbox providers, Northflank is a full-stack developer platform that includes databases, CI/CD, and GPU workloads alongside sandbox capabilities.
The platform differentiates primarily on enterprise deployment flexibility — teams can run sandboxes in Northflank's managed cloud or deploy into their own VPC while maintaining the same APIs and developer experience.
Key Capabilities
| Capability | Description |
|---|---|
| Sub-second Creation | Boot microVMs in under 1 second |
| Dual Isolation | Choose microVMs (hardware-level) or gVisor (kernel-level) |
| VPC Deployment | Deploy in your own AWS, GCP, Azure, or Oracle account |
| GPU Support | H100, A100, and other GPUs with full microVM isolation |
| Persistent Storage | Volumes from 4GB to 64TB, S3-compatible object storage |
| Horizontal Autoscaling | Scale based on CPU, memory, and RPS thresholds |
Product Surfaces / Editions
| Surface | Description | Availability |
|---|---|---|
| Northflank Cloud | Managed multi-tenant cloud | GA |
| BYOC (VPC) | Deploy in your cloud account | GA |
| Enterprise | Custom requirements, SLAs, white-label | Contact |
Market Position
Comparison with Competitors
| Feature | Northflank | E2B | Modal | Daytona |
|---|---|---|---|---|
| Isolation | microVM/gVisor | Firecracker | gVisor | Docker |
| VPC/BYOC | ✅ | — | — | ✅ (self-host) |
| GPU Support | ✅ | — | ✅ | — |
| Ephemeral + Persistent | ✅ | Ephemeral only | Ephemeral | Persistent |
| Open Source | — | ✅ | — | ✅ |
| Computer Use | — | — | — | ✅ |
Differentiation: Northflank is the only platform combining enterprise VPC deployment with GPU support and flexible ephemeral/persistent modes. E2B leads on adoption and open-source community; Modal leads on serverless Python; Daytona leads on Computer Use.
Target Segments
| Segment | Fit | Notes |
|---|---|---|
| Enterprise AI teams | ⭐⭐⭐ | VPC deployment, SOC2, enterprise features |
| GPU-heavy workloads | ⭐⭐⭐ | H100s 62% cheaper than big clouds |
| Startups | ⭐⭐ | May be overkill; E2B simpler for pure sandboxes |
| Open-source preference | ⭐ | Not open-source; E2B or Daytona better fit |
Pricing
Northflank uses usage-based pricing with transparent per-resource costs.
Compute Pricing
| Resource | Price |
|---|---|
| CPU | $0.01667/vCPU-hour |
| Memory | $0.00833/GB-hour |
| Disk (SSD) | $0.15/GB/month |
| Network Egress | $0.06/GB |
GPU Pricing
| GPU | Price/Hour |
|---|---|
| T4 | $0.80 |
| L4 | $1.42 |
| A10 | $1.76 |
| H100 | $2.74 |
| A100 | $3.14 |
GPU pricing is up to 62% cheaper than AWS, GCP, and Azure equivalents.
Example Monthly Costs
| Workload | Specs | Monthly Cost |
|---|---|---|
| Light sandbox | 0.5 vCPU, 1GB RAM, 10hrs/day | ~$4 |
| Standard agent | 2 vCPU, 4GB RAM, 8hrs/day | ~$20 |
| GPU workload | H100, 4hrs/day | ~$330 |
Technical Architecture
Isolation Model
Northflank offers two isolation technologies:
microVMs (Hardware-level):
- Each workload runs in its own lightweight VM
- Complete kernel isolation from host and other tenants
- Best for untrusted code execution and multi-tenant platforms
gVisor (Kernel-level):
- Application kernel intercepts syscalls
- Lower overhead than full VMs
- Good balance of security and performance
Infrastructure
| Component | Implementation |
|---|---|
| Orchestration | Kubernetes (EKS, GKE, AKS, or self-managed) |
| Networking | Service mesh with mTLS |
| Storage | Persistent volumes, S3-compatible object storage |
| CI/CD | Built-in pipelines with GitHub, GitLab, Bitbucket |
Viability Assessment
| Factor | Rating | Notes |
|---|---|---|
| Market Position | ⭐⭐⭐ | Strong in enterprise/VPC segment |
| Funding | ⭐⭐⭐ | $24M+ raised, sustainable |
| Customer Base | ⭐⭐⭐ | Sentry, 50k+ developers, Fortune 500 |
| Technical Moat | ⭐⭐ | BYOC is differentiator but replicable |
| Product Velocity | ⭐⭐⭐ | Active development since 2019 |
| Ecosystem | ⭐⭐ | Less dev community than E2B |
Overall Viability: Strong — Northflank has carved out a defensible position in enterprise deployments. The combination of VPC support, GPU access, and full-stack platform capabilities makes it sticky for teams that need more than just sandboxes.
Use Cases
1. Enterprise AI Agent Execution
Deploy secure sandboxes in your own VPC to meet compliance requirements while running untrusted AI-generated code.
2. GPU-Intensive Agent Workloads
Run ML inference, model fine-tuning, or compute-heavy agent tasks with H100s at competitive pricing.
3. Multi-Cloud Agent Infrastructure
Deploy the same agent infrastructure across AWS, GCP, and Azure using consistent APIs.
Bottom Line
Northflank is the right choice for enterprise teams that need VPC deployment, GPU support, or a full-stack platform beyond just sandboxes. It's overkill for simple stateless code execution (use E2B) or Computer Use workflows (use Daytona).
Best for:
- Enterprise teams requiring VPC/BYOC deployment
- GPU workloads where cost matters (62% savings vs big clouds)
- Teams wanting sandboxes + databases + CI/CD in one platform
Consider alternatives if:
- You want open-source (E2B, Daytona)
- You need Computer Use / browser automation (Daytona)
- You want the simplest possible SDK for ephemeral execution (E2B)
Related Research
- AI Agent Sandboxes Compared — Full market comparison
- E2B — Market leader for ephemeral sandboxes
- Modal — Serverless Python with GPUs
- Daytona — Fastest creation, Computer Use support