Why multi-cloud RAID beats multi-cloud sync

Replication vs. erasure coding

The dominant approach in "multi-cloud storage" today is replication: your file is uploaded to provider A, then synced to provider B, then synced to provider C. Three full copies, three places. Each copy is independently readable. The marketing says "multi-cloud" — what it actually means is "three different lock-ins."

Erasure coding is a different mathematical object. Your file is split into N data chunks and parity-extended to N+M chunks total, where any N of the N+M can reconstruct the original. SkyeConnex uses RS(5,2)">Reed-Solomon RS(5,2) — 5 data shards plus 2 parity shards, distributed across 7 providers, tolerant of losing any 2 and unreadable from any single one.

Why this matters operationally

Storage cost

RS(5,2) carries 7/5 = 1.4× storage overhead. Triple-replication carries 3× overhead. For every petabyte of data, that's a meaningful difference in spend before you even account for egress.

Failure isolation

A replicated multi-cloud setup degrades to single-cloud the moment two providers fail. RS(5,2) degrades the same way — but the failure point is two providers, not three (because RS(5,2) needs 5 of 7 — which means any 2 can fail). With triple-replication you only need 1 of 3, but the cost is the 3× overhead.

The right way to think about it: RS(5,2) buys you the same failure tolerance as triple-replication, at less than half the storage cost — and it gives you the security property that no single provider can read the data, which replication does not give you at all.

The security property

This is the part replication does not get you, at any cost. With triple-replication, any single provider has a full copy of your file. A breach of that one provider exposes everything. A subpoena to that one provider reveals everything. The "multi-cloud" of triple-replication buys you availability — it does not buy you any security against single-provider compromise.

With RS(5,2), no single provider — and no two providers — can read your file. The provider you trust least is no more dangerous than the provider you trust most. The architecture itself enforces a property that no contract can.

Why "multi-cloud" gets conflated

The word "multi-cloud" got pulled in two directions over the last decade. On one side: enterprises spinning up workloads on AWS and Azure and GCP at the same time, for reasons that have nothing to do with storage topology. On the other: the small handful of vendors building genuine cross-cloud primitives that treat the cloud as a substrate, not a destination.

The conflation is unfortunate because it lets every vendor with a sync tool call themselves "multi-cloud." For sovereignty buyers, the only multi-cloud that matters is the kind where no single cloud can betray you. That's erasure coding. Not sync.

What this looks like in SkyeConnex

Every file is framed into ~5 MB chunks. Each chunk is independently Reed-Solomon RS(5,2) encoded into seven shards. Shards are distributed across the customer's chosen seven providers — with idempotent upload keys so scatter is retryable, and per-shard SHA-256 hashes for bit-rot detection.

Reads work in reverse: parallel-fetch all seven shards, return when any five succeed, RS-decode locally, AES-GCM-decrypt on the client. The provider never assembles the file. See the architecture →

---

Published March 8, 2026 · Written by SkyeConnex Inc. · More from the SkyeConnex blog