From SAN to cloud and back: a short history of shared storage for post

Shared storage for post production has run a full lap: from fibre-channel SANs in a locked machine room, to Ethernet that quietly killed the fibre premium, to the cloud that promised to make the machine room disappear entirely, and now back to on-prem boxes that editors plug into the wall. The technology under each era changed, but the editor's wish never did. People want to open a 4K clip the instant they click it, and they want everyone on the team to see the same media without copying it around. This is the short version of how the industry has tried to grant that wish for twenty-five years, and why the pendulum is swinging home again.

The SAN era: fibre channel and the locked machine room #

Shared storage for editors effectively starts in 1999, when Avid shipped Unity MediaNetwork: fibre-channel storage and networking hardware bolted to a file system written expressly for video, so two editors could touch the same media without overwriting each other's work (Avid's own history page dates this to 1999, checked Jun 2026). Before Unity, "sharing" meant sneakernet. You finished a cut, copied the media to a drive, and walked it down the hall. Unity replaced the hallway with a Storage Area Network, where each workstation saw the central storage as if it were a local disk, talking to it over fibre at the block level rather than asking a file server for whole files.

The analogy that helps here: a SAN is less like a shared Dropbox folder and more like everyone in the building having a private cable running straight into one enormous hard drive. It was fast and deterministic, which is exactly what real-time playback of uncompressed video demands. It was also expensive and fragile. Fibre-channel switches, host bus adapters, and the cabling carried a real premium, and Avid was explicit that for 2K and up, raw, or DPX work you wanted fibre rather than Ethernet because 10 GbE would not provide the bandwidth (Avid SAN guidance, checked Jun 2026). For a decade, if you ran a serious Avid or broadcast facility, the cost of admission was a fibre SAN and someone who knew how to keep it alive.

Ethernet quietly erodes the fibre premium #

The first swing was not cloud. It was Ethernet eating fibre channel's lunch. In 2004 EditShare pioneered "in-place editing" over standard Gigabit Ethernet, demonstrating that Avid and Final Cut workstations did not need fibre channel to play multiple streams of uncompressed SD or compressed formats like DNxHD 220 (EditShare company history, checked Jun 2026). Avid itself conceded the point: in 2006 it launched Unity ISIS, which one trade summary described as "part Ethernet-connected NAS and part robust, powerful SAN," built on industry-standard GigE rather than fibre (Studio Daily and TV Tech coverage, checked Jun 2026).

This mattered because Ethernet was the network everyone already had. You did not need a separate fabric and a specialist to run it; you needed a switch and a NIC. The tradeoff was honest: early Gigabit Ethernet was bandwidth-constrained compared to fibre, which is why high-end DI and uncompressed work stayed on SANs for years. But for the broad middle of post, the price of collaboration fell sharply. That same family of storage is still shipping. Avid NEXIS, the descendant of ISIS, was still being updated in 2026, with NEXIS 2025.5.7 released in February 2026 (Avid knowledge base, checked Jun 2026). The old-guard enterprise vendors followed a parallel arc, which I leave to the dedicated Facilis and OpenDrives old-guard piece and the history of Avid ISIS and Nexis.

The cloud promise: make the machine room disappear #

The second swing was the big one. LucidLink, founded in January 2016 by storage engineers George Dochev and Peter Thompson, asked a simple, radical question: what if you could open a file in the cloud the same way you open one on your desktop, with no download, no sync, and no progress bar (LucidLink ten-year anniversary post, checked Jun 2026)? The answer was file streaming. A bucket of objects sitting in someone else's data center, presented to the editor as an ordinary mounted volume, with only the bytes you actually scrub pulled down on demand. It is the SAN idea, the central pool everyone shares, stretched across the public internet.

Early on it was a hard sell. LucidLink took 33 straight investor rejections, and post production was famously slow to trust the cloud because the files are enormous and the tools are specialized (LucidLink anniversary post, checked Jun 2026). Then March 2020 happened. Facilities sent editors home overnight, and a central volume that anyone could mount from a kitchen table went from curiosity to lifeline. A later Caretta Research survey cited by the DPP put cloud and remote-editing adoption among video professionals at roughly 90 percent (DPP cloud post-production research, checked Jun 2026). For a few years it genuinely looked like the machine room was going to disappear.

Then the bill arrives #

The catch with renting the machine room is that the rent never stops, and getting your data back out costs extra. Cloud storage hides two meters that on-prem never had: a monthly fee that scales with every terabyte you keep, and an egress fee every time bytes leave the building. On AWS S3, transfer out to the internet runs about 0.09 dollars per GB for the first 10 TB each month, tapering to 0.05 dollars per GB once you are past 150 TB (AWS S3 pricing page, checked Jun 2026). For a post house cycling tens of terabytes of dailies, proxies, and renders, that is a line item that grows with success rather than shrinking. I keep the specific platform teardown out of this history and point you to the LucidLink 2026 real-bill breakdown and the piece on what leaving actually costs.

Egress fees were such an obvious lock-in mechanism that regulators forced the issue. Under pressure from the European Data Act, AWS and Google both announced free data-transfer-out for customers leaving entirely. AWS's program, announced in March 2024, waives DTO charges for eligible accounts in good standing with more than 100 GB stored that are moving all their data off, with a 90-day window to complete the exit (AWS news blog and InfoQ coverage, checked Jun 2026). Note the shape of that concession: leaving is free only if you leave completely, and only once. The everyday meter on bytes that move while you work stays on.

The swing back to on-prem #

The third swing is happening now, and it is driven by arithmetic, not nostalgia. The Barclays CIO Survey found 86 percent of CIOs planning to move some workloads back from public cloud to private or on-prem infrastructure, the highest figure recorded (Barclays survey via repatriation coverage, checked Jun 2026). The poster child is 37signals: after spending about 700,000 dollars on Dell hardware to run workloads it had hosted in AWS, its cloud bills fell by roughly 2 million dollars a year, and when it moved its data off S3 to on-prem Pure Storage arrays, AWS waived 250,000 dollars in egress fees to let it leave (The Register, May 2025, checked Jun 2026).

What makes the return practical for editors specifically is that the hardware caught up. The thing that justified the cloud, fast random access to huge files from anywhere, is now achievable on a NAS in a closet because local NVMe is cheap and 10 and 25 GbE switching is no longer exotic. The on-prem of 2026 is not the fragile fibre SAN of 1999; it is commodity Ethernet and flash with a modern mount layer in front. I dig into why this moment arrived in the cloud-cost backlash piece, and into the appliances that make it real in the 2026 storage-appliance roundup.

This is the lane where I will be honest about JuiceMount, because it is genuinely native to this part of the story and nowhere else in it. JuiceMount is an attempt to keep the one good idea from the cloud era, an object store presented to the editor as a real mounted volume with block-level streaming and a local SSD cache, while moving the bucket back onto hardware you own so the egress meter and the per-seat rent both go to zero. Where it does not fit: if your team is genuinely distributed across cities with no shared site, a hosted cloud mount is still the simpler answer, and you should not contort an on-prem setup to pretend otherwise. The history here is not "cloud bad, local good." It is a pendulum, and the smart move is knowing which side of the swing your particular team is standing on.