I had Sora because it came bundled with my ChatGPT subscription, and I used it to make short clips for internal PowerPoints when a slide needed something more than a stock image. Not for anything serious, never built into any work I was doing. When the shutdown announcement landed on 24 March I read it the way I’d read a Microsoft deprecation notice, which is to say I assumed the date would move once enough customers shouted.
Somewhere in the internal PowerPoint archive, this was probably carrying more dramatic weight than the slide deserved.
The cleanest version of the Microsoft pattern is App-V, which I’ve watched play out enough times that I plan around it. Microsoft set an April 2026 end-of-life for the client, customers pushed back, and in November 2024 Microsoft reversed the client and sequencer decision while keeping the server EOL on the original date. The client is now in fixed extended support, which is a long way from gone. The question I ask when a Microsoft EOL gets announced is not “what’s the date” but “what’s the actual date once the customer outcry has finished doing its work.”
The Sora announcement on 24 March said the app and web experience would stop on 26 April and the API would stop on 24 September, with no softening and no second statement walking it back in the weeks since. Disney had agreed a three-year IP deal and a $1B investment, and according to reporting they were told less than an hour before the public announcement that the deal was off. OpenAI’s enterprise marketing was still live during all of this, including a case study on the San Antonio Spurs using Sora for fan interactions.
Google has been killing SaaS products long enough that there’s a website tracking the graveyard, with over 300 entries from Reader and Inbox through to Stadia. Reader got announced in March 2013 and shut down fifteen weeks later. Stadia got announced in September 2022 and shut down sixteen weeks later, and the gaming press treated that as abrupt because the service had only been running for three years. Sora was announced on 24 March and the consumer app stopped five weeks after that, with the API running for another five months. The pattern I’d seen on the SaaS side was already faster than the Microsoft pattern, and Sora was sharper than that again.
Microsoft moves deprecation dates because there’s a contract structure around the relationship and the customer has a hold on the timeline. The App-V client date moved because enough enterprises that depend on App-V told Microsoft they weren’t going to be ready, and Microsoft cares about that because those enterprises are paying for a lot of things that aren’t App-V. That structure is the thing I’d been quietly assuming was there for Sora, and it wasn’t.
The economics work differently for Sora, which was reportedly burning around $1M a day with user numbers down from a peak of a million to under half a million, while OpenAI is heading toward an IPO and reallocating compute to enterprise tools and coding products. The cost of running it decided when it would be shut down, and the Disney deal and the Spurs case study weren’t enough weight on the other side of the decision to change that. A product that costs more to run than it earns gets cut, and being a flagship doesn’t change the maths.
I don’t think this generalises to all SaaS. Some products are profitable, some have switching costs that lock the vendor in as much as the customer. But the reflex I’d built up over years of cleaning up after Microsoft deprecations, which is that announced dates are negotiating positions, doesn’t transfer cleanly to a product whose continued existence depends on the vendor wanting to keep paying for it.
I’m not sure what I’d do differently next time, exactly. I wasn’t going to build anything serious on Sora and I still wouldn’t. But there are SaaS products I treat with the App-V reflex, where I half-assume the vendor will blink if the customers shout loud enough, and Sora is a reminder that the assumption hadn’t been tested in a while.