Cloudflare reaches 500 Tbps provisioned external capacity
Key Points
- 500 Tbps provisioned external interconnection capacity
- Line-rate eBPF (l4drop) DDoS mitigation on every server
- Edge compute (Workers + Containers) deployed in 330+ cities
Summary
Cloudflare announced it has provisioned 500 Tbps of external interconnection capacity across 330+ cities and 125+ countries. This is provisioned port capacity (not peak traffic) and includes headroom used as a DDoS absorption budget. The network couples distributed edge compute (Workers, KV, Durable Objects, Containers) with line-rate, per-server mitigation so attacks are detected and dropped locally without human intervention.
Key Points
- What 500 Tbps means: total provisioned external interconnection capacity across transit, private peering, IXPs, and CNI ports — used for scale and DDoS absorption, not observed peak traffic.
- Distributed DDoS stack (per-packet, per-server): NIC -> xdpd (driver mode) -> l4drop (eBPF rules) -> Unimog L4 load balancer; dosd samples traffic and shares heavy-hitter tables colo-wide; Quicksilver distributes mitigation rules globally in seconds; flowtrackd provides stateful TCP inspection for Magic Transit.
- Real-world scale: a 31.4 Tbps attack in 2025 was mitigated in seconds with no human paging; >5,000 attacks were blocked the same day.
- Edge compute and security converge: Workers (and Containers on Workers) run in every city on the same machines that drop attack traffic, so malicious packets are discarded before application processing.
- Routing security: Cloudflare enforces RPKI (ROA signing and origin validation) and is pushing for ASPA adoption to prevent route leaks; they will drop RPKI-invalid routes.
- AI traffic signal changes: AI agents and crawlers now account for ~4%+ of HTML requests; detection uses verified bot ranges, TLS ClientHello fingerprinting, behavioral analysis, and robots.txt signals to separate legitimate crawlers from abusive traffic.
- Operational asks: network operators are encouraged to peer (PeeringDB) or join the Edge Partner Program; adoption of RPKI/ASPA is recommended for routing hygiene.
Practical takeaways for engineers
- Expect mitigation decisions to be made at the NIC/host level via eBPF; architect services assuming malicious traffic can be dropped before hitting application CPU.
- If you operate a network, enable RPKI ROAs and plan for ASPA to avoid accidental reachability loss and improve global routing hygiene.
- For apps sensitive to crawler behavior, rely on combined signals (TLS fingerprints, behavior, robots.txt) and expose controls so site owners can accept or block crawlers.
- Consider peering or colocating to reduce latency and increase resilience; Cloudflare emphasizes peering and interconnection as key to scale.