The cost of keeping the work

Last month I spent $3,464.97 on three hard drives.

They were 28-terabyte Seagate IronWolf Pro drives for our studio NAS. Eighty-four terabytes of raw capacity. A serious amount of storage, but not an unusual amount of data for a working video production company.

High-resolution cameras produce enormous files. A documentary project can involve days of interviews, multiple cameras, location sound, drone footage, stills, graphics, project files, proxies, renders, and final masters. Then there are the copies, because a file that exists in only one place is a file you should assume you have already lost.

Storage has always been part of the cost of this work. What changed is the assumption underneath it.

For most of my career, storage got cheaper. Cameras produced more data every year, but the cost per terabyte kept falling fast enough to make the problem manageable. Buy what you need now. Wait a year. The next drive will be bigger and probably cheaper.

That assumption broke in 2026.

The drive that more than doubled

In January 2024, I bought a 4TB Crucial X10 Pro portable SSD for $352.34 before tax. It’s the kind of drive that moves constantly through a production workflow: footage comes off a camera card, travels into the edit, and follows a project between systems.

The same 4TB model is now listed by CDW Canada for $812.99.

That’s a 131 percent increase.

It’s one product at one retailer, not a consumer price index. Retail inventory, promotions, sellers, and exchange rates all complicate a direct comparison. But it reflects something much larger happening upstream.

TrendForce projected NAND flash contract prices to rise 55 to 60 percent in the first quarter of 2026. For the second quarter, it projected another 70 to 75 percent increase.

Those aren’t retail price increases. They’re the prices manufacturers negotiate for the flash memory inside SSDs, NVMe drives, and camera cards. The increase takes time to move through the supply chain, and it doesn’t land evenly.

But it lands.

What AI has to do with your camera card

The short version is that AI infrastructure needs memory and storage at a scale the consumer market wasn’t built to compete with.

The longer version is more interesting.

AI is usually described as a compute problem. GPUs. Accelerators. Power. Cooling. All of that is real. But AI systems also need somewhere to put the data they train on, the data they produce, and the expanding context they use when they run.

That has made enterprise SSDs one of the most valuable parts of the storage market. Cloud providers are placing enormous orders. Manufacturers are responding rationally by putting limited production capacity toward the customers and products with the highest margins.

In the first quarter of 2026, enterprise SSD revenue reached a record US$18.46 billion, up 86 percent from the previous quarter. Contract prices rose by roughly 80 percent. TrendForce says demand from cloud service providers nearly doubled during the quarter.

At the same time, the major NAND manufacturers are expected to add almost no new production capacity this year. Micron says data-centre DRAM and NAND shipments in 2026 will be more than double what they were two years ago.

A CFexpress card and a data-centre SSD aren’t the same product. But they draw from the same broad pool of NAND flash production. When the most powerful buyers in the world are locking up supply, a memory card for a camera sits very far down the priority list.

Sony made that visible in March when it suspended orders in Japan for most of its CFexpress Type A, Type B, and SD memory-card lines, citing the global semiconductor shortage.

This is how an AI data centre reaches into a camera bag.

Not directly. Not because an AI company bought the card you were going to buy. Because it changed what the companies making memory can earn, what they choose to manufacture, and who gets served first.

Hard drives are a different squeeze

It would be easy to describe every storage increase as part of the same chip shortage. That would also be wrong.

NAS hard drives don’t store data in NAND flash. They use magnetic platters. The three 28TB drives I bought in June came through a different supply chain than the SSD in my editing kit.

But they arrive at the same destination.

AI workloads create enormous amounts of data that must be stored persistently and economically. Flash is fast. Hard drives are still the cheapest practical place to keep huge volumes of data online. Western Digital now says 90 percent of its revenue is driven by AI and cloud, and its leadership has been explicit: virtually every AI workload creates data that eventually needs to be kept on hard drives.

The result is two storage squeezes happening at once.

AI and data-centre demand are pulling NAND production toward enterprise SSDs. The same buildout is absorbing high-capacity hard-drive output for long-term storage. One affects the fast media creators use to capture and edit. The other affects the slower media we rely on to keep the work.

Different mechanisms. Same invoice.

The archive nobody priced

The uncomfortable part isn’t that drives cost more. It’s that most creative businesses have never priced the promise they make when they keep everything.

Clients tend to assume their footage still exists.

Sometimes they come back six months later for a new cut. Sometimes three years later. A staff member has changed. A campaign has returned. Someone remembers an interview that didn’t make the original piece and now matters for a different reason.

When the footage is still there, it feels like good service. And it is.

But keeping it was never free.

There is the working copy. The NAS. The backup. The offsite copy. The replacement drives. The electricity. The network. The time required to maintain it. The risk carried by the production company when a client assumes an informal archive is a permanent one.

For years, falling storage prices made that cost easy to hide inside the business. Rising prices expose it.

Creators now have to answer questions many of us have avoided. How long do we keep camera originals? What does the project fee include? When does active media move into an archive? Who pays to restore it? Are we keeping every take forever, or only the material with a plausible future use?

These aren’t IT questions anymore. They’re business-policy questions.

What a more deliberate workflow looks like

The answer isn’t to delete everything. It’s to stop treating every file as if it needs the same kind of storage forever.

Fast flash should be used where speed earns its keep: camera capture, active projects, cache, and high-bitrate finishing. Completed work can move to less expensive storage. A local NAS can hold the accessible archive. A second copy can live offsite, in cloud storage, or on tape when the volume justifies the equipment.

Puget Systems recommends separating applications, project media, and cache, and points out that many editing workflows don’t need NVMe speed for every file. Buying the fastest possible storage for the entire chain is expensive and often unnecessary.

Proxy workflows matter more for the same reason. DaVinci Resolve can automatically generate small proxy files, let editors work from them, and relink the camera originals when it’s time to finish. The original media still needs to exist somewhere, but it doesn’t need to occupy the most expensive tier through the whole edit.

The backup rule doesn’t change. Three copies across two kinds of media, with one copy offsite, remains a sensible baseline. RAID is useful. RAID is not a backup.

What changes is the intention behind it.

Keep active work fast. Keep the archive economical. Keep the policy clear.

And stop promising indefinite storage by accident.

Deciding what deserves to remain

There is a strange contradiction in creative work right now.

AI is making it easier to generate more images, more video, more versions, and more content than anyone can meaningfully watch. At the same time, the infrastructure behind that generation is making it more expensive for working creators to preserve the material we actually captured.

The world is producing more and making the act of keeping more expensive.

That pressure will force some practical changes. Storage will appear as a line item. Retention periods will become part of contracts. Restoring old projects will carry a fee. Creators will get better at separating active media from archive media, originals from proxies, and valuable footage from files kept only because deleting them felt uncomfortable.

Some of that is overdue.

An archive isn’t just a pile of drives. It’s a set of decisions about what might matter later.

For a documentary filmmaker, that isn’t a trivial responsibility. Unused footage can become historically important. An interview that didn’t fit one story can become the centre of another. The value of an image isn’t always visible on the day it is captured.

So I’m not interested in a ruthless deletion policy. I’m interested in an honest one.

The cost of storage is forcing creators to decide what the promise to keep something is actually worth, who that promise is for, and how long it can reasonably last.

The drives in the NAS aren’t the archive.

The decision to keep the work is.