blog/Articles/OpenAI Shut Down Sora. $22M/Day Losses, Disney Pullout, Plummeting Market Share

News

kkm

Backend Engineer / AWS / Django

2026.03.2910 min1 views

Share X Hatena LINE LinkedIn RSS

Key takeaways

OpenAI announced the shutdown of video generator Sora. Disney pulled its $1B partnership. Losing share to Gemini and Claude across coding and enterprise markets.

OpenAI Has Shut Down "Sora"

On March 24, 2026, OpenAI announced the termination of its AI video generation service "Sora." The standalone app, developer API, and video generation within ChatGPT are all being discontinued.

Translation

We're saying goodbye to Sora. We'll share timelines for the app and API shutdown, along with details on saving your creations.

Sora first stunned the world with demo videos in February 2024 and launched as an app in September 2025. Marketed as a tool that could generate realistic videos from text prompts alone, it hit 1 million downloads in just 5 days. Yet only 15 months after its debut, its story has come to an end.

Why is it shutting down? And what's changing in the landscape surrounding ChatGPT? Let's lay out the facts.

Is Sora 2 Also Shutting Down? A Breakdown of All Versions

You might be thinking, "Only the old Sora is shutting down, and the latest Sora 2 will survive, right?" The answer is no: all Sora-related services, including Sora 2, are being terminated.

Sora went through multiple versions.

Version	Release Date	Features	Status
Sora 1	Dec 2024	Original model	Shut down early on Mar 13
Sora Turbo	Dec 2024	Faster version	Shut down early on Mar 13
Sora 2	Sep 2025	Major improvements in physics accuracy & audio sync	Shutdown announced on Mar 24

On March 13, the Sora 1 lineup was discontinued, and OpenAI announced it had "consolidated everything into Sora 2." Just 11 days later, Sora 2 and everything else was shut down entirely.

The shutdown covers all of the following:

✕Sora standalone app (iOS)
✕sora.com (web app)
✕Developer API
✕Video generation within ChatGPT

OpenAI says the Sora research team will be redirected to "world simulation research for robotics." While the underlying technology will remain within the company, Sora as a user-facing product is finished.

Why Did Sora Fail?

In short, it was burning through too much money.

Running Sora cost approximately $15 million per day in compute costs. That's roughly $5.4 billion annually. Generating just one 10-second video cost about $1.30, and Sora lead Bill Peebles himself admitted the economics were "completely unsustainable."

User numbers were also in freefall.

Period	Monthly Downloads	MoM Change
Sep 2025 (Launch)	1M DLs in 5 days
Nov 2025 (Peak)	~3.3M
Dec 2025	~2.24M	-32%
Jan 2026	~1.2M	-45%
Feb 2026	~1.1M	-8%

Source: Business of Apps

Cumulative downloads reached roughly 9.6 million. Meanwhile, total in-app purchase revenue was just about $2.1 million. Burning $15 million a day for $2.1 million in total revenue. The numbers simply don't add up.

OpenAI is planning an IPO (initial public offering) by the end of 2026. They couldn't afford to show investors a business hemorrhaging $15 million every single day.

Disney Pulled Its $1 Billion Partnership

Disney was hit the hardest by Sora's shutdown.

In December 2025, Disney announced a $1 billion investment in OpenAI. The grand plan was to license Mickey Mouse, Cinderella, Marvel heroes, and Star Wars characters on Sora, letting anyone create Disney videos with AI.

However, with Sora's termination, Disney pulled out of the partnership entirely. No money ever changed hands. The deal collapsed before it could close.

Disney stated, "We respect OpenAI's decision to shift to other priorities in this rapidly advancing AI landscape." Diplomatic language, but it's not hard to imagine the disappointment of having a $1 billion partnership wiped off the table.

Why People Are Saying "Google Has Already Won"

Sora's shutdown is also a symptom of a bigger problem for OpenAI: competitors are catching up fast.

Google's AI model "Gemini 3.1 Pro" was released in February 2026. This model outperforms ChatGPT's latest model, GPT-5.4, across multiple benchmarks (tests that measure AI performance).

Benchmark	Description	Gemini 3.1 Pro	GPT-5.4
ARC-AGI-2	Abstract reasoning	77.1%	73.3%
GPQA Diamond	Graduate-level science questions	94.3%	92.8%
MMMU-Pro	Multimodal complex problems	80.5%	79.5%
SWE-Bench Verified	Software bug fixing	80.6%	—

Source: Google DeepMind Model Card

It's not just benchmarks. Apple has switched Siri's AI backend from OpenAI to Google. This means Gemini will be built into over 2 billion Apple devices. As reported by Fortune, this represents a massive loss of distribution for ChatGPT.

Looking at app market share, ChatGPT's lead has shrunk dramatically over the past year.

App	Jan 2025	Jan 2026	Change
ChatGPT	69.1%	45.3%	-23.8pt
Gemini	14.7%	25.2%	+10.5pt
Grok	1.6%	15.2%	+13.6pt

Source: Fortune

Google's advantage isn't just Gemini's performance. Gmail, Google Search, Android phones, YouTube -- they can embed Gemini into products that billions of people already use every day. That "distribution power" is a weapon OpenAI can't replicate. ChatGPT requires users to install an app; Gemini is "already there from the start."

Coding Has Become "Claude's Domain"

While Google threatens ChatGPT's position as the go-to consumer AI, an even more dramatic shift has occurred in the programming world. Anthropic's AI "Claude" has become the de facto leader in coding.

Anthropic was founded in 2021 by former members of OpenAI's safety research team. They broke away to prioritize "AI safety" -- essentially a splinter group from OpenAI.

What's drawn the most attention is "Claude Code," a programming assistant tool. Since its launch in May 2025, it has grown rapidly, surpassing $2.5 billion in annual revenue. That's a faster pace than ChatGPT, making it the fastest-growing enterprise software product in history.

In a survey of 15,000 programmers, Claude Code captured 46% of votes as "the most loved coding tool", far ahead of second-place Cursor (19%) and third-place GitHub Copilot (9%).

The migration from ChatGPT to Claude is accelerating. According to data from AI measurement platform Larridin, sessions from users switching from ChatGPT to Claude have surged 1,487%. TechCrunch has also reported that "users are ditching ChatGPT for Claude."

The reasons for switching aren't just about performance. When OpenAI signed a contract with the U.S. Department of Defense, Anthropic maintained its stance of "not allowing Claude to be used for mass surveillance or fully autonomous weapons." This resonated with developers, sparking the "#QuitGPT" movement, and Claude gained support on ethical grounds as well.

Revenue growth is overwhelming. Anthropic's annualized revenue run rate reached $19 billion as of March 2026, growing at 7-10x year over year. OpenAI's growth rate is 3.4x. If this pace continues, Epoch AI's analysis suggests Anthropic could surpass OpenAI's revenue around August 2026.

ChatGPT by the Numbers: Where It Stands Now

It's been a lot of tough news so far, but ChatGPT isn't "dead." By the numbers alone, it's still an overwhelming giant.

Metric	Figure	Notes
Weekly Active Users	910 million	As of Feb 2026
Annual Revenue	$25 billion	Annualized run rate
Total Funding Raised	$120+ billion	Amazon, Nvidia, SoftBank, etc.
Valuation	$730 billion	~2.5x Toyota's valuation
Paid Business Users	9+ million	Surged from 5M in Aug 2025

Sources: TechCrunch, Yahoo Finance, CNBC

It took Salesforce 18 years and Google 17 years to reach $25 billion in annual revenue. OpenAI did it in about 39 months. In both user count and revenue, it's growing at a pace that's in a different league from past tech companies.

Still, some numbers are concerning. Enterprise market share has fallen from 50% to 27%. There's also data showing that 70% of companies newly adopting AI are choosing Anthropic's Claude. A picture is emerging: "individuals are still using it, but enterprises are starting to leave."

OpenAI's Next Move Is Called "Potato"

OpenAI isn't sitting idle under pressure. On the same day Sora's shutdown was announced, CEO Sam Altman told staff that development of the next-generation AI model was complete.

Its codename: "Spud" -- slang for potato. For reference, GPT-5.3's codename was "Garlic." OpenAI's development team apparently has a tradition of naming models after vegetables.

Spud's official name and specs remain entirely under wraps. All we know is that "initial development is complete" and it will be "released within weeks." Altman says it "can truly accelerate the economy," and according to The Information, it reportedly possesses "extreme reasoning capabilities."

It's not just about models. OpenAI is also developing a "super app" that combines ChatGPT, the coding tool "Codex," and the AI browser "Atlas" into one. As reported by CNBC, the vision is to evolve ChatGPT into "one app that can do everything."

Even more symbolically, OpenAI renamed its product organization to "AGI Deployment." AGI stands for Artificial General Intelligence -- AI with human-level or greater intelligence. They've essentially declared: "We are the organization that deploys AGI."

Redirecting the massive compute resources that went into Sora toward Spud and the super app. That's OpenAI's bet.

Altman Stepped Away from the Safety Team

Alongside Sora's shutdown and the Spud announcement, another important change was revealed: Sam Altman is stepping away from direct oversight of both the AI safety team and the security team.

Safety oversight is being transferred to Chief Research Officer Mark Chen, and security to President Greg Brockman. Altman himself says he will focus on fundraising and data center construction.

This decision has context. OpenAI's safety team has been through significant turbulence over the years.

← Swipe to navigate

Over the past two years, nearly every key figure on OpenAI's safety team has departed. Furthermore, according to TIME, more than 30 current and former OpenAI employees have publicly supported California's AI safety bill, putting them at odds with their own CEO's lobbying efforts -- an extraordinary situation.

To be fair, a CEO stepping back from safety oversight isn't inherently bad. Delegating to experts is a rational decision. But given this history, the risk of it being perceived as "prioritizing fundraising over safety" is not trivial.

Is ChatGPT Going to Be OK?

Let's sum up the facts so far.

OpenAI killed Sora, which was burning $15 million a day, and lost its $1 billion partnership with Disney. Google's Gemini has overtaken it in benchmarks and the Apple partnership, and app market share dropped from 69% to 45% in one year. In the programming world, Anthropic's Claude has become the dominant player, and enterprise market share has shrunk from 50% to 27%. The CEO has stepped away from safety oversight.

And yet, this is a company with 900 million weekly users, $25 billion in annual revenue, and $120 billion in funding. It's not going under anytime soon.

However, "being the first company to popularize AI" and "continuing to be the leader" are two different things. Just as Yahoo! was overtaken by Google and Bing in search. Just as Netscape was replaced by Internet Explorer and Chrome in browsers. In the world of technology, there's no guarantee that the first mover keeps winning.

OpenAI has placed all its bets on the next-generation model "Spud." Squeezed between Gemini and Claude, whether this model -- set to launch in the coming weeks -- can live up to expectations will determine the outcome. The AI industry's power map for 2026 may look very different depending on the answer.

Update: Not a Single Dollar Had Moved on Disney's $1 Billion Investment

New details have emerged about the contract termination between Disney and OpenAI following Sora's shutdown. According to Variety, the three-year deal signed in December 2025 would have made over 200 characters from Disney, Marvel, Pixar, and Star Wars available on Sora.

Disney was set to make a $1 billion equity investment alongside the contract, with warrants to purchase additional shares. However, according to Deadline, no actual funds were ever transferred. The contract effectively lapsed automatically with Sora's termination.

Disney had also been planning to use ChatGPT Images and Sora to create "fan-inspired" videos and distribute Sora-generated content on Disney+. With Sora's estimated inference cost of approximately $15 million per day making it unsustainable, all of these plans have been scrapped along with it.