Sora is dead: what OpenAI’s video generator shutdown reveals about AI

On March 9, 2026, OpenAI officially shut down Sora, its AI video generator launched with great fanfare six months earlier.

This decision, both brutal and predictable, marks far more than a product discontinuation: it exposes the economic tensions fracturing the entire generative AI video sector.

Here is the complete autopsy of a foreseeable strategic failure, and what it means for creators who work with these tools.

Sora’s complete timeline

From demo to launch

In February 2024, OpenAI unveiled Sora and shattered social media with stunning demonstrations: a golden retriever surfing, the streets of Tokyo filmed at night, mammoths crossing a snow-covered field.

The quality of the videos clearly surpassed everything the competition was producing at the time, when the best tools were still generating clips of 4 to 8 seconds with visible temporal inconsistencies.

Sora promised up to 60 seconds of video with unprecedented temporal coherence: the technical gap seemed insurmountable for competitors.

Between February 2024 and the public launch on September 30, 2025, a stretch of 17 months of silence, Sora remained locked behind restricted access while its competitors closed the gap at an accelerating pace.

OpenAI had turned its initial advantage into a strategic temporal disadvantage: the window of opportunity had closed before the public launch even happened.

As we anticipated in our analysis of Sora at its 2024 unveiling, the length of the private testing period represented a major strategic risk in a sector that evolves in a matter of weeks.

Six months of ups and downs

On launch day, Sora reached 100,000 installs despite requiring an invitation, before surpassing ChatGPT in US App Store rankings.

The one million download milestone was crossed in under 5 days, a record even for OpenAI.

The peak came in November 2025 with 3.3 million monthly downloads, confirming genuine viral interest.

By December 2025, downloads had dropped 32%, during the holiday season, which is typically favorable for mobile apps.

In January 2026, the decline accelerated by a further 45%, bringing installs down to 1.2 million per month.

Over its entire lifespan, Sora generated only 9.6 million cumulative downloads, compared to 900 million weekly active users for ChatGPT at the same time.

Why OpenAI killed Sora: 4 real reasons

The financial abyss

The main reason is arithmetic: brutal and beyond appeal.

According to estimates by Deepak Mathivanan, analyst at Cantor Fitzgerald, generating a 10-second clip cost OpenAI approximately $1.30 in GPU fees.

Each video generation required around 40 minutes of total GPU time, running simultaneously across multiple cards at approximately $2 per hour.

Scaled to actual usage, the total daily cost exceeded $15 million, or more than $5.4 billion annually to keep the service running.

Against these expenses, the revenue the app generated over its entire six months of operation amounted to just $2.1 million in in-app purchases.

In a single day of operation, OpenAI was burning through seven times the total amount collected from users over six full months.

To reach break-even, Sora would have needed to generate $450 million per month, or 200 times more than its best result.

Bill Peebles, who led the project, had himself stated back in October 2025: “The economics are currently completely unsustainable.”

The IPO in sight

OpenAI is preparing its IPO for 2026, with a target valuation of $730 billion.

In this context, every loss-making budget line becomes a negative signal for institutional investors who will scrutinize the accounts.

On March 16, 2026, a week before Sora’s official shutdown, Fidji Simo, OpenAI’s CEO of Applications, outlined an unambiguous change of direction at an internal meeting.

Her statement: “We can’t lose this moment because we’re distracted by side quests.“

The era of consumer product proliferation at OpenAI was declared over: the focus shifts to profitability and the enterprise market.

Competition caught up

In February 2024, Sora’s technological lead seemed unassailable.

During the 17 months of restricted access, Runway launched its Gen-3 Alpha suite with rapid iteration, Kling (by Kuaishou) deployed 2-minute full videos with exceptional scene coherence, and Google announced Veo 2, repositioned as an enterprise solution with a robust API.

By the time Sora launched publicly in September 2025, the technical gap that had seemed insurmountable in 2024 had narrowed considerably.

Worse: competitors had spent months refining their models, collecting user data, and building loyal creator communities.

Sora arrived late to a market it had created.

The deepfakes problem

Within the first weeks of public deployment, Sora became a machine for producing disputed content.

Users generated deepfakes of protected characters without consent: unauthorized versions of Disney, Marvel, and Star Wars characters, and videos of public figures placed in fabricated contexts.

OpenAI faced growing legal liability on the grounds of intellectual property and the spread of misleading content.

Video content moderation is structurally more expensive than text or static image moderation: every second of video generates dozens of frames to be analyzed individually.

The strategic signal: OpenAI chooses enterprise

Sora’s shutdown is not an accident: it is the marker of a deliberate strategic pivot.

OpenAI is now concentrating its resources on two priority areas: enterprise solutions and robotics.

On the robotics front, AMI Labs (Autonomous Machine Intelligence) is working on World Models: internal representations of the physical world used to teach robots to anticipate the consequences of their actions.

High-quality video generation serves as an ideal training ground for these World Models: understanding how objects move, how light changes, how physical interactions unfold.

Sora’s closure as a consumer product does not mean the abandonment of the underlying technology.

Sora as a tool for creators is dead.

The technology lives on, redirected toward less glamorous and far more profitable uses.

It is the Google+ of AI video: a technically impressive product, unable to find its business model, abandoned in favor of more strategic priorities for the parent company.

AI video facing the profitability wall

Sora’s shutdown raises a structural question that no one in the industry dares ask: can a consumer AI video generator be economically viable today?

Video generation costs are fundamentally different from those of text or images.

Generating an image with Flux or Midjourney costs fractions of a cent: a few seconds of compute, a few cents of infrastructure.

Generating 10 seconds of 720p video costs between $0.50 and $2 depending on the model, with 8 to 40 minutes of GPU time required.

At that ratio, even with a $20/month subscription, an active user generating 20 videos a month costs the platform more than they pay.

Look at Meta Movie Gen: despite Meta’s impressive technical capabilities, the project remains confined to internal research, with no consumer launch announced.

The quiet dominance of Chinese players is partly explained by lower GPU infrastructure costs and government subsidies that reshape the economics for companies like Kuaishou (Kling) or ByteDance (Seedance).

Viable business models in AI video are concentrated in B2B: enterprise pricing passed on to clients with production budgets, not individuals on $20-a-month subscriptions.

Sora alternatives in 2026

For creators who were using Sora, the market offers solid alternatives, each with its own distinct strengths.

Kling 3.0

Kling, developed by Chinese giant Kuaishou, stands out as the most direct alternative to Sora in terms of visual quality.

Kling generates videos up to 2 minutes long with scene coherence and motion realism that often match or exceed Western competitors’ standards.

The free plan is generous, and paid plans start at lower prices than their American equivalents.

One caveat: Kling is a Chinese solution, which raises the usual questions about data protection and dependency on a foreign provider.

Google Veo 3.1

Veo stands apart through its enterprise integration and API access via Vertex AI, designed for companies looking to embed video generation into their production workflows.

Cinematic quality is excellent, with meticulous handling of physics and lighting.

Access is progressive, with priority given to existing Google Cloud customers: a deliberately enterprise-first strategy.

Runway Gen-4

Runway Gen-4 remains the reference tool for video post-production professionals and creative studios.

Its suite of tools built around video generation (inpainting, outpainting, motion brush, lip sync) has no direct equivalent among competitors.

Pricing is higher than Kling, but the level of creative control and output quality justify the premium for professional use cases.

Seedance, Pika and Hailuo: three agile tools

Seedance (ByteDance) delivers fast video generation with text-to-video prompts calibrated for social media content creators.

Pika targets individual creators with an accessible interface and motion control features for specific zones.

Hailuo AI (MiniMax) rounds out the picture with long-form video capabilities and remarkable character fidelity across consecutive shots.

These three tools share something in common: they are cheaper than Sora was, faster to generate, and available without a waitlist.

What creators should take away from this

Sora’s death is a practical lesson for any professional who integrates AI tools into their workflow.

First lesson: never build a critical dependency on a single AI tool, especially when the company behind it is structurally operating at a loss on that product.

Second lesson: early adoption numbers do not validate the long-term viability of a service.

Sora had 3.3 million monthly downloads at its peak: that was not enough to justify its survival against exponential infrastructure costs.

Third lesson, and the most important: quality generative AI video remains a sector in search of a viable business model.

The tools that survive will be those that have found either a solvent enterprise niche, or a cost structure low enough to reach profitability at scale.

AI video is not dead: it is changing hands, migrating toward enterprise, and settling durably with players whose business models differ from OpenAI’s.

Explore our analyses of the AI tools that are genuinely reshaping creative professions in our Artificial intelligence section.

FAQ

Why did OpenAI shut down Sora in March 2026?

OpenAI shut down Sora due to an insurmountable economic imbalance: GPU costs exceeded $15 million per day against only $2.1 million in revenue generated over the app’s entire lifespan.

The strategic pivot toward enterprise and IPO preparation accelerated the decision.

How much did Sora earn OpenAI in total?

Over six months of operation (October 2025 to March 2026), Sora generated only $2.1 million in in-app purchases, including $1.1 million in the United States.

That is roughly seven times less than what a single day of GPU operation cost.

Was the $1 billion Disney deal with OpenAI real?

An agreement was announced in December 2025, but it was never formally signed or executed.

Disney never paid the promised billion, and OpenAI never paid royalties for Disney characters: the deal was conditional on corporate approvals that never materialized.

What are the best alternatives to Sora in 2026?

The main alternatives are Kling (Kuaishou) for quality at the best price, Google Veo for enterprise use cases, Runway Gen-4 for post-production professionals, and Seedance, Pika, and Hailuo AI for content creators.

Is Kling technically better than Sora?

On several criteria, yes: Kling generates videos up to 2 minutes long with scene coherence and motion realism comparable or superior to Sora.

It is also cheaper and available without a waitlist.

Can generative AI video be profitable?

The consumer model is structurally unprofitable at current GPU costs.

Viable models target enterprise (costs passed on to B2B clients with production budgets), or rely on cheaper infrastructure like that of Chinese players benefiting from government subsidies.

What does OpenAI’s robotics pivot actually mean?

OpenAI is investing in AMI Labs and World Models: internal representations of the physical world used for robotic AI.

Sora’s video generation technology serves as a training foundation for teaching robots to simulate the consequences of their actions in real physical space.

Did Fidji Simo play a role in Sora’s shutdown?

As OpenAI’s CEO of Applications since August 2025, Fidji Simo announced on March 16, 2026 the end of the consumer product proliferation era, one week before Sora’s official shutdown.

Her mandate is explicitly oriented toward profitability and the enterprise market.

What should creators who were using Sora do now?

They should migrate to an alternative suited to their needs: Kling for quality at the best price, Runway for professional workflows, Veo for enterprise projects.

The main adaptation involves prompts: each tool has its own optimizations for achieving quality results.

Could OpenAI relaunch Sora in the future?

The technology remains among OpenAI’s assets and could be reintegrated into enterprise offerings or APIs for studios and businesses.

A consumer comeback is possible if GPU costs drop sufficiently: it is clearly not a strategic priority for OpenAI in 2026.