I ran the division the morning the numbers went up, and the three Game Pass tiers landed within three cents of each other. Ultimate, twenty-three dollars for fifteen hours of streaming, is a dollar fifty-three an hour. Premium, fifteen dollars for ten hours, is a dollar fifty. Essential works out at two dollars, which is the surcharge for buying the small one.
Those first two tiers were built as different products. Different catalogue, different day-one access, years of marketing spent on the gap between them. Price them by the hour and they are the same product, sold twice.
Fifteen hours
Then on Saturday the reason turned up. A report put the fifteen-hour ceiling exactly at the point where Microsoft stops making money on a streaming subscriber. The cap is the break-even line, lifted off a spreadsheet and printed on a product page.
Microsoft’s stated reason was the honest one, and I will give it that. The service costs more the more you use it, and the company said so plainly instead of dressing the limit up as a feature for your own good. It also says around four percent of subscribers currently play enough to hit the new ceiling. One in twenty-five.
Sit with that four percent for a moment. It means the cap changes almost nothing this November. That is what a ceiling installed for later looks like.
Because once a price is pinned to what a thing costs to serve, the ceiling moves whenever the cost moves, and the cost of a datacentre GPU hour is not a number trending downwards in a year when every one of those GPUs has a queue of AI tenants bidding for it. Six years of unlimited streaming, from the Project xCloud days, ended in a paragraph. Nothing you streamed in those six years is on a drive you own.
The correction
Sony spent the same Saturday correcting a different story. The wave of reports saying PlayStation disc manufacturing was about to be gutted by ninety percent in 2028 came from a misread of comments by a Sony DADC executive. The real figure is a ten percent decline over that year.
The correction is fair and it should be said loudly, because getting it wrong in that direction is exactly the failure this blog is supposed to avoid.
And yet. For two days, nobody who works in or around this industry found the ninety percent version implausible. Not one of us read it and thought, no, that can’t be right. We read it and thought, well, there it is. The rumour was wrong about the number and correct about the mood, and a business earns that kind of credulity over years, one quiet withdrawal at a time.
Thirty-six degrees
Here is where it stops being about consumers, because we do this to ourselves and we do it worse.
LTO-10 drives read LTO-10 media and nothing else. No LTO-9, no LTO-8, no firmware update, no slower speed setting that coaxes the old cartridge into tracking. Nine generations of a format whose entire pitch was that the new drive reads what the last drive wrote, and the bridge is gone.
The reason is physical, which I think matters. LTO-10 writes its servo bands — the stripes the head steers against while the tape flies past — at thirty-six degrees, to get finer positioning and less settling delay. A head built for one geometry cannot track the other. There is no policy to appeal, no executive to be irritated at. It is geometry, and geometry does not negotiate.
Which makes it the purest version of the thing. Every studio that has been running more than a few years has the crate. Under a desk, in a cupboard, in the room with the dead monitors: show wrapped, project delivered, source scans and caches and the one version of the file that actually rendered, written to tape and put away with the quiet confidence of people who did the responsible thing. Then the drive dies, and you replace it with the drive that is on sale now, and the crate becomes decoration.
You already know the software version of this. The scene that needs a renderer two majors back. The plugin whose licence server was switched off. The project that opens, technically, and comes up with every material slot empty. We are the people who make the thing, we hold the only masters that exist, and our own retention plan is a shelf and some optimism.
Seventeen and a half million
So look at what people did this year, while all this was going on.
Seventeen and a half million CDs sold in the United States in the first six months of 2026, up from twelve million in the same stretch of 2025. Forty-six percent more units, fifty-nine percent more money — a hundred and seventy-one million dollars, on a format that had shrunk twenty-two percent the year before. Vinyl up eighteen. Physical revenue as a whole up twenty-six percent while streaming, still four times the size, grew under five.
Streaming did not collapse and I am not going to pretend it did. Everything grew. But a lot of people spent real money this year on an object that cannot be renegotiated, delisted, region-locked, or capped at fifteen hours, and they did it in a year when they were reminded roughly weekly that the other kind of access can be. From here — and I watch this industry from a country that mostly buys its outcomes rather than setting them — that is the most rational purchase on the table.
Priced per flat
Which brings me to the funniest object of the week. Kaleidescape announced a media server holding two hundred and forty-six terabytes on a pair of enterprise SSDs, in a box the size of a disc player, that will store somewhere around four thousand films you own outright and play twenty-five of them at once. No price announced. Based on what the company charges for everything else, comfortably north of a hundred and fifty thousand dollars, and you will want the four-thousand-dollar player to go with it.
Two SSDs and a chassis, for the price of an apartment in most of Italy.
Strip the walnut veneer off and look at what is actually being sold there. It is permanence. The guarantee that the film is in the room, on hardware you control, and stays there whatever a licensing negotiation does in 2029. That used to cost fifteen euros and come in a plastic box. It is now a luxury SKU with a concierge, and the mass-market version of the same promise is being retired by ten percent a year on one side and metered by the hour on the other.
Nobody is keeping a copy for you. Not the platform, which has a margin to defend, and not the format, which has physics to obey. The copy you keep is the only ownership that has ever actually worked, and we are watching it get priced like a yacht while we tell ourselves the shelf under the desk is somebody else’s problem.
Go and check whether your drive still reads your crate. I will wait.