A billion-dollar bet that lasted six months
Sora launched in September 2025 with enormous fanfare. OpenAI's AI video generator could turn text prompts into surprisingly polished short videos — and the internet went wild with it immediately. People were generating Dragon Ball Z clips, Studio Ghibli-style animations of their favourite characters, SpongeBob in unexpected situations — all from a text box. The viral potential was obvious. The business case, it turns out, was not.
Six months later, Sora is shutting down. The number that explains everything: it was reportedly losing $1 million a day to operate. Not a week. Not a month. A day.
The numbers that killed it
At peak, Sora had about 1 million daily active users. That's a genuinely impressive number for a new AI product — most apps never see that kind of adoption. But AI video generation is extraordinarily compute-intensive. Generating a single video requires far more processing power than generating text or even images. You need massive GPU clusters running continuously. The electricity bill alone is staggering, let alone the hardware costs.
As the novelty wore off, users drifted away. The active user count fell below 500,000 — a 50% drop. Meanwhile, the cost structure didn't change. $1 million a day going out, diminishing revenue coming in. The math doesn't work regardless of how much VC money you have sitting in the bank.
OpenAI isn't exactly hurting — the company raised $40 billion earlier this year at a $300 billion valuation. But even at that scale, burning $365 million a year on a product with falling engagement and no clear path to profitability is a problem you have to fix.
The Disney disaster
The most chaotic part of this story isn't the shutdown itself — it's how Disney found out about it.
Disney had signed a reported $1 billion deal with OpenAI to use Sora for content creation. The plan was genuinely ambitious: Disney+ would use Sora-generated short-form videos as a new content category, and Disney executives would get licensed access to Sora tools to storyboard live-action remakes and other productions. Former CEO Bob Iger was telling investors as recently as February that this partnership was moving forward.
When OpenAI decided to pull the plug on Sora, Disney was apparently notified approximately one hour before the public announcement. One hour. For a $1 billion deal. That's not a wind-down conversation or a strategic realignment meeting — that's a text message equivalent of "hey, we're ending this, it's going public in 60 minutes."
Disney is now left scrambling to find alternative AI video partners. Given how publicly Iger had talked up the Sora integration, this is an embarrassing position to be in.
What OpenAI is pivoting to instead
The internal logic at OpenAI seems to be: consumer-facing flashy demos are expensive and unpredictable, enterprise tools and robotics are where the real money is.
The company is reportedly working on a "superapp" that combines ChatGPT's desktop app, Codex (its coding tool), and an AI browser called Atlas — an agentic AI platform that can autonomously execute tasks like writing code, analysing data, and booking travel. That's a much more defensible business than an AI video generator where the main use case appears to be copyright-adjacent fan content.
Robotics is the other direction. Sam Altman has been publicly bullish on physical AI — robots that can perform real-world tasks. It's a longer-term bet, but it has a clearer enterprise customer base than "generate anime clips."
Elon Musk jumps in immediately
You knew this was coming. Within hours of the Sora shutdown news, Elon Musk was on X pointing out that his competing product, Grok Imagine (xAI's image and video generation tool), is "positive gross margin" — meaning it makes more than it costs to run, unlike Sora.
Musk's argument is interesting though: he believes video generation and understanding is essential for achieving AGI (Artificial General Intelligence — the theoretical point where AI reaches human-level capability across all tasks). His reasoning: visual data carries vastly more information than text. Photons are "the highest bandwidth form of communication," as he put it. So investing in video AI isn't just a consumer product play — it's infrastructure for the next phase of AI development.
Whether that's correct is genuinely uncertain. But the contrast is sharp: Musk is doubling down on video AI while OpenAI is abandoning it. One of them is going to look very right or very wrong in a few years.
The copyright problem nobody wants to talk about
Buried in the Sora story is something that deserves more attention. The most viral Sora content was almost universally copyright-adjacent — Studio Ghibli style, Dragon Ball Z characters, SpongeBob scenarios. None of this was licensed. OpenAI trained Sora on enormous amounts of video data without clear consent from the creators of that content, and users then used the tool to generate derivative works from those same creators' styles and characters.
The fact that the $1 billion Disney deal existed at all while users were actively generating unauthorised Disney character videos on the same platform is one of the stranger moments in recent tech history. Disney was paying OpenAI for access to a tool that was simultaneously being used to generate unlicensed Mickey Mouse content. That tension was always going to resolve badly.
What this means for AI video going forward
Sora shutting down doesn't mean AI video generation is dead — far from it. Google has Veo 2, which is genuinely impressive. Meta has movie Gen. Runway, Pika, and Kling are all active and improving. The difference is most of these are either enterprise-focused (priced for studios and businesses, not consumers) or have found more sustainable cost structures.
The consumer AI video dream — where anyone can generate Hollywood-quality clips from a phone — isn't dead, but it's clearly more expensive and harder to monetise than the 2024 hype cycle suggested. Someone will crack it eventually. Just probably not the way OpenAI tried to.
TamilTech's take
The Sora story is a useful reality check on AI hype in general. A product can be technically impressive, genuinely viral, and backed by the world's most-funded AI company — and still fail as a business if the unit economics don't work. $1 million a day for a product that peaked at a million users means each user was costing OpenAI roughly $1/day — and most of them weren't paying that much. The math was never going to close. What's wild is that it took six months and a $1 billion Disney humiliation to force the shutdown rather than a quiet wind-down. That's the thing about AI companies right now — moving fast and breaking things includes occasionally breaking your own billion-dollar partnerships.
Comments (0)
Be the first to comment!