Fly Io Pricing & Developer Hub
Everything you need to know about Fly Io pricing in 2026: platform fees, serverless quotas, compute instance rates, and bill optimization strategies.
Architectural Deep Dive: Anycast Routing & Edge Linux MicroVMs
Fly.io transforms Docker containers into lightweight Linux microVMs running on bare-metal hardware in over 35 edge regions worldwide. By deploying compute physically close to end users and connecting every region over an encrypted global WireGuard mesh network, Fly.io eliminates the multi-hundred-millisecond cross-continental latency overhead inherent in centralized single-region cloud architectures.
Fly.io's billing model is strictly utility-metered based on seconds of microVM execution. Containers automatically scale down to zero when HTTP requests cease and wake up in under 300 milliseconds upon the next incoming TCP connection, offering true serverless elasticity with standard Linux operational semantics.
Fly.io MicroVM & GPU Machine Rates (March 2026)
| Machine Configuration | Hardware Specifications | Hourly Price | Monthly Continuous Run |
|---|---|---|---|
| Shared-CPU 1x (256MB) | 1 shared vCPU, 256MB RAM | $0.0026 / hr | $1.94 / mo |
| Performance-CPU 2x (4GB) | 2 dedicated vCPUs, 4GB RAM | $0.0420 / hr | $31.00 / mo |
| GPU A100 (40GB PCIe) | 1x A100, 8 dedicated vCPUs, 32GB RAM | $1.5000 / hr | $1,080.00 / mo |
| GPU L40S (48GB) | 1x L40S, 8 dedicated vCPUs, 32GB RAM | $1.7500 / hr | $1,260.00 / mo |
Key Platform Capabilities
- Anycast IP Networking: Fly.io assigns a single Anycast IPv4/IPv6 address to your application. Border Gateway Protocol (BGP) automatically routes user traffic to the geographically nearest server datacenter.
- LiteFS Distributed SQLite: Built-in support for LiteFS allows read-heavy applications to run SQLite databases replicated across 20 global regions with sub-5ms local read latencies.
Frequently Asked Questions
Explore More Providers
Fly.io Edge Architecture: Firecracker MicroVMs & LiteFS Distributed SQLite
Fly.io transforms container deployment by compiling Docker images into lightweight **Firecracker microVMs** running directly on bare-metal hardware close to users. With Anycast BGP networking and native WireGuard mesh routing across 35+ global regions, applications execute with sub-20ms edge latency. Crucially, Fly.io's **LiteFS** allows teams to run distributed SQLite databases with automatic primary election and asynchronous streaming replication to read replicas worldwide, eliminating the multi-hundred-dollar fees associated with managed cloud PostgreSQL clusters.
Fly.io Production Machine Rate Card (Pay-as-you-go)
| Machine Type | CPU / RAM Spec | Hourly Rate | Monthly Continuous (730 Hrs) | Auto-Stop Capability |
|---|---|---|---|---|
| shared-cpu-1x (256MB) | 1 Shared vCPU / 256MB | $0.0026 / hr | $1.94 / mo | Yes (Wake in 300ms) |
| shared-cpu-1x (1GB) | 1 Shared vCPU / 1GB | $0.0078 / hr | $5.70 / mo | Yes (Zero cost when stopped) |
| performance-2x (Dedicated) | 2 Dedicated vCPU / 8GB | $0.0849 / hr | $62.00 / mo | Dedicated CPU cores |
| NVMe Volume Storage | Fast Local SSD | $0.15 / GB-mo | $1.50 / 10 GB disk | Retained across machine stops |
Fly.io Auto-Stop Production Configuration (fly.toml)
app = "production-api"
primary_region = "iad"
[http_service]
internal_port = 8080
force_https = true
auto_stop_machines = true
auto_start_machines = true
min_machines_running = 1
[[vm]]
size = "shared-cpu-1x"
memory = "1gb"Engineering Q&A: Fly.io Edge Infrastructure
How does LiteFS solve multi-region database latency on Fly.io?
LiteFS intercepts SQLite system calls at the FUSE filesystem level. Read queries execute against local NVMe flash memory in under 1ms in every regional node, while write transactions are automatically forwarded to the primary region via WireGuard with zero custom application code.
What happens when a Fly Machine is auto-stopped?
When incoming traffic ceases, Fly.io stops the microVM, halting CPU and RAM billing completely. The machine's persistent NVMe volume remains attached ($0.15/GB-mo). When a new packet hits the Anycast IP, the machine boots in ~300ms.
WireGuard Private Mesh Peering & Distributed Key-Value Caching
Every application deployed on Fly.io automatically joins a private, encrypted **6PN (IPv6 Private Network)** powered by kernel-level WireGuard. Services communicate across datacenters using private `.internal` domain names without exposing ports to the public internet. This architecture allows teams to deploy distributed cache clusters (like Redis or Dragonfly) in primary regions and replicate data to edge workers across London, Tokyo, and Sydney with sub-millisecond local network hops.
Fly.io Production Reliability Guardrails
- Minimum 2 Replicas per Region: Always provision at least 2 machines per critical region (`fly scale count 2 --region iad`). If an underlying physical host undergoes hardware maintenance, Fly's proxy shifts traffic to the surviving machine with zero dropped connections.
- Volume Snapshots Automation: Fly.io automatically snapshots persistent NVMe volumes every 24 hours and retains them for 5 days. For long-term compliance, export backups to Cloudflare R2 using daily cron tasks.
How does Fly.io route incoming global web traffic?
Fly.io announces a single Anycast IPv4/IPv6 address across 35+ global edge locations. Incoming HTTP requests hit the nearest Fly edge proxy, which terminates TLS and forwards the request over encrypted WireGuard tunnels to the nearest healthy application microVM.
What are the outbound bandwidth charges on Fly.io?
Fly.io includes 100GB of outbound public bandwidth free each month on paid plans. Additional egress is billed at $0.02 per GB in North America and Europe, which is 77% cheaper than AWS S3/EC2 egress rates.
Fly.io Machine REST API & Ephemeral Workload Orchestration
Fly.io exposes a direct Machine REST API that allows applications to programmatically provision, start, stop, and destroy Firecracker microVMs within milliseconds. Rather than keeping a cluster of idle GPU or high-RAM instances running 24/7, SaaS platforms can launch an isolated microVM on-demand to process heavy video encoding or large LLM batch requests, shutting down the hardware the instant the computation finishes, cutting batch processing infrastructure costs by up to 85%.
Edge Observability Tip: Stream real-time Fly Machine metrics directly to Grafana Cloud or Datadog using Fly's native Vector log forwarder. Monitor CPU throttles, memory limits, and Anycast routing latencies across global regions.