Decentralized CDNs have been attempted for a decade. Most got stuck before they shipped. The thing that's different now isn't ambition — it's that the four prerequisite layers all became production-grade in the last 24–36 months. Until they did, every serious attempt was structurally blocked by problems outside its own protocol: the transport wasn't fast enough, the settlement layer was too expensive, the unit of account was too volatile, or the legal posture made compliance a single point of failure.

decdn is the first attempt that gets to design only the protocol, because the rails it runs on are finally ready. This is what changed.

Transport: QUIC stopped being a research project

QUIC was an IETF draft until 2021. By 2024 it carried roughly a third of public web traffic and shipped natively in every major OS. The properties that matter for a delivery network are now load-bearing parts of the public internet, not a thing we have to build.

0-RTT connection establishment. A repeat connection sends application data in the first packet, no handshake round-trip. For a CDN where probe fan-out is the critical-path latency — a cache-miss probes ~30 peers in parallel — this collapses inter-continental discovery from ~250–500ms down to one-way RTT. We use 0-RTT on cdn/probe/v1 and cdn/dht/v1. Five years ago we'd have had to build it on top of TCP+TLS ourselves and accept 2× the discovery latency.
NAT traversal as a library function. iroh ships hole-punching, relay fallback, and direct-path upgrade as a single Endpoint API. The fraction of node operators behind residential NAT we can usefully include just went from ~10% to ~95% without us writing networking code.
Connection migration. A node operator's IP changing mid-stream — laptop sleep, mobile handoff, ISP renegotiation — doesn't kill the delivery session. This is the difference between "you need a static IP and a real datacenter" and "your spare 1U at home counts."

QUIC plus iroh isn't incremental. It's the difference between "CDN-grade transport requires CDN-grade infrastructure" and "CDN-grade transport runs on a $5 VPS."

Addressing: BLAKE3 made verification free

Earlier networks (BitTorrent, IPFS, Filecoin retrieval) used SHA-256 or SHA-1 piece hashes. Verification was either expensive enough to skip on commodity hardware (SHA-256 at multi-gigabit) or insecure enough to need a separate proof oracle. BLAKE3 changed that.

~3–10× faster than SHA-256 on a single core, and trivially parallelizable across cores via its tree structure.
The hash is a Merkle tree by construction, so subrange proofs come for free — a node serving bytes 5MB–6MB of a 50GB file can prove integrity of just that range without re-streaming the rest.
iroh treats BLAKE3 as the native blob ID, so the wire protocol and the storage layer share one hash. No translation layer, no second proof system.

Result: every byte we deliver is verified by the receiving client at line rate. We do not need a separate proof-of-delivery oracle. The retrieval problem Filecoin had to design a whole sub-protocol around — proving you delivered the right bytes — is just not a problem we have to solve.

Settlement: Stablecoins are now load-bearing financial infrastructure

USDC and USDT cleared north of $10T of on-chain volume in 2024 and are settled by Circle / Tether against actual reserves with monthly attestations. They are no longer "crypto experiments" — they are how dollars move on public chains. For a network of node operators with real infrastructure costs (VPS, bandwidth, storage), this is the difference between a viable business and a casino.

A node operator who books $300/month in earnings keeps $300/month of buying power for their server bill, regardless of token-market conditions. USDC payment channels can be opened, topped up, and settled using the same primitives as any other ERC-20 — battle-tested, audited, with mature tooling.

Five years ago the only way to pay node operators on-chain was a volatile native token. Every prior decentralized delivery network priced bandwidth in something whose value could halve overnight. We don't.

L2s: Payment channels are finally cheap enough to bother with

Per-MB delivery payments are on the order of $0.00001 — six orders of magnitude below any L1 transaction fee. Without an off-chain channel mechanism the economics never close. But payment channels need to open and settle on-chain, which on Ethereum mainnet costs $5–50 per transaction. That kills the channel before it starts.

The current generation of low-cost EVM L2s brought execution costs to ~$0.05–0.10 per transaction while keeping Ethereum-grade security via fraud proofs. A full channel lifecycle on our target L2 — open + close + settle — costs ~$0.23, a 2.3% overhead on a recommended 10 USDC default deposit, amortizing toward zero across long-lived channels. That's the number that makes the unit economics work. It did not exist in production before 2023.

The same maturity wave gave us:

ERC-4337 account abstraction, so node operators and clients can use Safe smart wallets with batched approvals and gas paid in USDC instead of ETH.
OpenZeppelin SignatureChecker supporting both EOA and ERC-1271 smart-account signatures uniformly across all our contracts.
Circle's CCTP for native USDC bridging across L2s, removing the trust assumption of wrapped-token bridges.

Demand caught up at the same time

The supply-side primitives matured in parallel with a demand shock that didn't exist in 2020. Payload sizes grew an order of magnitude — modern open-source ML weights run 10–700GB; high-bitrate media, container images, and dataset distributions all push hundreds of GB; package archives are a hundred-fold larger than they were a decade ago. Traditional CDN list pricing didn't move: Cloudflare R2, AWS CloudFront, and Akamai are still in the $0.04–$0.20/GB band, the same as ten years ago. The teams shipping these payloads cannot run that math.

Centralized risk got worse, not better. A handful of hosts carry the load for industries whose entire pitch is openness. Every major release throttles incumbent CDNs within hours.

The demand-side market existed at this scale for the first time around 2023. The supply-side primitives became viable around the same time. The intersection is the wedge.

The window

The supply-side primitives are ready. The demand-side pain is acute. The category has no incumbent. Three things have to be true at once for a decentralized delivery network to work, and for the first time, all three are.

The transport stack is fast enough that a residential or commodity-VPS node can serve at CDN-comparable latency. (QUIC + iroh + BLAKE3.)
The settlement stack is cheap enough that per-MB delivery economics close. (USDC + L2 payment channels.)
The demand exists at scale so the supply side has something worth showing up for.

Each leg of that tripod arrived in its current form within the last 24–36 months. The team that ships now has a window before either a hyperscaler retrofits delivery onto a storage DePIN or a well-funded entrant starts from scratch. After that, network effects in the operator fleet and content cache state lock in the first competent shipper.

This is what "why now" actually means. Not "the market is hot," but "the prerequisite layers are done, the demand is visible, and nobody has composed them yet."