My SSD Filled Up — Rethinking Where Data Goes, by Price and by Speed

2026年8月28日2026年8月30日

18 August 2026

Data kept piling up, and the SSD inside my PC is filling up.

Local AI models alone come to 249GB. On top of that, video and photos keep accumulating.

The trouble is that SSD prices in 2026 have gone up — three to six times off the bottom. Buying more is not appealing.

So I decided to think about where data goes, from scratch. Does all of it actually need to sit on an SSD?

This article keeps what I measured here separate from what I looked up. Measurements were taken on 18 August 2026. Prices are Japanese retail; ¥1,000 is roughly $6.70.

Where can data actually live?

Before rethinking anything, I listed what places exist. There are four.

The one filling up here is the first: the SSD inside the PC.

The cloud is a good option — until the capacity grows

To be clear up front, the cloud is a good option.

I looked up what two of the common ones actually cost.

Checked on the providers’ own pages on 19 August 2026 — Japanese consumer pricing.

So what does owning it cost? Comparing against a two-bay NAS

To compare against the cloud I need the purchase price. I looked at the smallest configuration.

These are sold two ways: complete, with drives already fitted, or the enclosure alone, drives bought separately. The first is cheaper and works out of the box.

Street prices checked on Amazon on 20 August 2026. They move with the retailer and the date.

Complete, with drives fitted. This is the form the table above compares. Two 2TB drives, 2TB usable once the reserve is taken.

Enclosure only, drives bought separately. You pick the drives, which makes growing capacity later easier.

The more capacity you need, the sooner they meet

Here the difference in how you pay starts to matter. Cloud pricing scales with capacity; on a NAS the enclosure price is fixed and only the drives cost more.

The next size up is 8TB fitted (two 4TB drives; 4TB usable after the reserve).

Prices checked 19–20 August 2026. The cloud jumps from 2TB straight to 5TB; there is no tier in between.

How many drive bays to choose: four, if you have room

Everything above assumes two bays. The number of bays changes what proportion of the capacity you get to use.

With only two bays, keeping a reserve always costs you half. With four, the reserve is one drive’s worth out of four. The more drives, the smaller the share the reserve takes.

But do not take that as settled safety.

An example of a four-bay unit. More expensive than two bays, but holding back one drive’s worth still leaves three drives of capacity.

What you cannot lose goes on both of two NAS units

This is what I do here. Important data is written to both of two NAS units.

Writing the same data twice costs capacity, so only the irreplaceable goes on both. Everything else sits on one or the other.

How the drives are grouped (how many, in what arrangement, and why the rebuild window is dangerous) is the subject of the next article.

For reference, putting my 59TB in the cloud would take twelve 5TB plans — ¥34,800 a month, ¥417,600 a year. The more you store, the wider that gets.

Which also means a few hundred GB is well served by the cloud. This article is about what happens past that.

What is the difference between an SSD and an HDD, and why do HDDs survive?

Before changing what data lives on, the difference. These two record data in completely different ways.

An HDD spins a platter and reads it with a head — close to a record player. An SSD has no moving parts and stores everything electrically.

Line them up and the SSD wins on speed, silence, durability and weight. PCs moving to SSDs was inevitable.

HDDs survive on price and capacity

Price per terabyte is exactly why I still use HDDs. Everything else about SSDs is better, but assembling 59TB makes the difference count.

HDDs fail — you make up for it with numbers

With a platter spinning, a single HDD is a fragile thing. Which raises the obvious worry: can you trust important data to cheap drives?

So you group several HDDs so the data survives one of them failing. Here that is six 16TB drives with one drive’s worth held in reserve. One dies, you swap it, and it rebuilds.

The details of that arrangement — how many drives, in what layout, and where the danger is — are in the next article.

This state of “it keeps growing and I cannot delete it" is what pushed me to try a machine whose only job is storage. To find out how fast it reads, I measured it.

There are three places an HDD can go

Of the four places listed earlier, three can take an HDD. The cloud drops out because you do not choose what it runs on.

What I have is the third. A machine on the network like this is called a NAS, and mine holds 59TB. What is a speed of one gigabyte in nine seconds good for, and what is it not? I measured it again to find out.

How I measured the NAS read speed

A number means nothing without the conditions. Here is only what you need to reproduce it.

The 117MB/s here is what came out reading a 1GB file that happened to be lying around. It is not an average over repeated runs. Read it as “this is roughly what it did here".

How the drives are grouped does not affect this read speed

The arrangement above (six 16TB drives) has no bearing on this result.

As shown below, 117MB/s is close to everything a 1000Mbps link can deliver. The link runs out first, so two drives or six, the number does not change. The arrangement starts to matter only after the link is upgraded to 10Gbps.

The NAS reads 1GB in nine seconds — is that slow?

My first reaction was that it was slow. The M.2 drive in the PC, measured the same day, does 3,500MB/s. Thirty times faster.

Then I did the arithmetic and changed my mind.

So however many drives you group inside the NAS, or whatever SSD cache you add, as long as the path crosses that 1000Mbps point, 117MB/s is where it stays.

The NAS had two links ten times faster than mine

The unit is a UGREEN NASync DXP6800 Pro. Checking the specification:

The NAS had two links ten times faster, and I was receiving it on one narrow one.

My first thought was to replace the switch and the network port. Then I stopped and thought about it again.

Before speeding it up, I wrote down what I use the NAS for

Writing it out made it obvious. The NAS is a place to keep things, not a place to work.

There are 202GB of models on it, but I do not run them from there — I copy them across first. I am not asking the storage to be fast, which is why 117MB/s has never been a problem.

Even replacing both ends, a 2.5Gbps PC port only doubles it. I could not find a reason to spend that on a place things merely sit.

If not speed, what is a NAS actually good for?

Three things.

1. No monthly payment

A NAS costs money once, then only electricity. It catches up in a year or two, and the gap widens from there.

2. When the service ends, the disks are still yours

This one made me think.

There is a NETGEAR NAS here as well, and the ReadyNAS line is discontinued. From what I could find, manufacturing ended in 2021, support ended during 2022, and firmware updates have stopped for most models.

Cloud services end. So do manufacturers. But there is a difference — and in fact, I am still using that discontinued NAS.

I demoted it. The new unit handles everyday use; the discontinued one holds long-term storage — old data I no longer touch.

Used that way, frozen firmware barely matters. It needs no new features and no speed; it needs to sit there.

If you have an out-of-support unit sitting idle, this arrangement is a practical one. It still works as storage, so move the interface to the newer machine.

While I am here: a NAS is often assumed to mean wrestling with an admin panel, and that has not been my experience. Set it up once and it becomes putting photos and video in, and taking them out.

It is not all upside. An unpatched machine is still sitting on the home network. Moving the interface to the newer unit does not fix that.

Keeping it unreachable from outside is still necessary. On security, a machine that still gets updates is plainly the safer one.

Even so, “what you bought does not become worthless" is a property a monthly subscription does not have.

3. You can use it without opening a PC

This one I only learned by using it.

With the UGREEN, the basic operations are done on a phone. Neither the QNAP I used before nor the NETGEAR I still have gave me that.

The other thing: with the unit within reach, media goes in over USB directly. Cards from a shoot go straight onto the NAS without starting a PC.

“SD card reader" in the specification table above is what that line actually buys you.

Across three units, ingest speed differed the most

Having used three, this is where the difference was largest.

The QNAP and the NETGEAR both read media perfectly well. It is not that they could not.

But the UGREEN uses SSDs as a cache, and ingest is markedly quicker. Push a fast card at it and it takes it without making you wait. For dumping a shoot straight in, that lands every time.

I said above that storage does not need to be fast. Putting things in is the exception. If writing to the NAS is slow, you stop wanting to put anything on it.

Only the NAS speed was measured: what this article cannot tell you

Of everything above, what I measured here is the NAS read speed and the M.2 read speed, and nothing else. Cloud pricing and the NETGEAR end-of-life dates were looked up. The ingest difference is an impression from use; I did not time it.

From a buyer’s point of view it may not matter which. Think of it as “choose a model that can use SSDs as a landing pad" rather than “choose that brand" and it works as advice either way.

Hardware used in this article

The NAS I measured. A six-bay unit running six Toshiba 16TB drives. You do not normally need anything this large; I chose it expecting to keep using it for a long time.

The drives inside it. Built for NAS use, designed to spin continuously.

The external SSD I use for carrying models around. It read at about 1,000MB/s here — roughly a third of the 3,500MB/s from the M.2 inside the PC. The same SSD changes a great deal depending on where and how it is connected.

In summary: find the narrowest point first

This is the thing I most wanted to write down.

Then, once you know where it is, the next question is whether widening it is worth it. Here it was not. A place things merely sit does not need to be fast.

The simplest way to apply this to yourself

If you are worried that HDDs will be too slow, ask yourself whether your current cloud storage bothers you.

What I found here

• The NAS measured 117MB/s — 93.6% of theoretical for a 1000Mbps link. It is using the link up
• What is slow is the cabling, not the HDDs. The NAS had two links ten times faster
• But storage does not need to be fast. What runs, runs off the local SSD; what is kept, sits on the NAS
• 59TB in the cloud works out at ¥417,600 a year. A NAS costs once, then electricity
• What survives the ending differs. The cloud has to be evacuated; with a NAS the disks stay
• Two units, split by role. The new one for everyday use, the out-of-support one for long-term storage, reached through the new one’s app
• Ingest speed seems to come down to whether the model can use SSDs as a landing pad. Whether that is a vendor difference or a tier difference, this article cannot say

I started because SSDs got expensive, and ended up at a question about where to put what. Not everything needs to live somewhere fast.

Next comes what goes inside: RAID, the scheme for grouping drives, and how an SSD’s internals change both its speed and its life. Buy without knowing that difference and the cheap purchase turns out to be the one that dies early.