OpenAI Kills Its Video Star to Save the Enterprise
OpenAI is shutting down Sora after rising costs and falling usage, and is shifting its focus to enterprise AI amid intensifying competition from Anthropic.
In less than 15 months, Sora went from being one of OpenAI’s most talked-about consumer experiments to its most visible retreat.
On March 24, 2026, OpenAI announced it was shutting down Sora. The company offered no formal reason. But the signals were already clear with falling usage, unsustainable costs, and a shift in strategy that now places enterprise AI above consumer experimentation.
Sora’s trajectory was steep and short-lived. The app peaked in November 2025 with 3.3 million downloads across app stores, according to data from mobile intelligence firm Appfigures accessed by Wired. By February 2026, downloads had dropped to 1.1 million. Lifetime revenue from in-app purchases reached an estimated $2.1 million, a modest amount for a product that demanded massive computational resources.
At its peak, Sora’s infrastructure costs were staggering. Reports from Forbes that the app was burning an estimated $15 million per day in inference costs. Bill Peebles, who leads the Sora team at OpenAI, had already acknowledged the strain last year, writing that "the economics are completely unsustainable.”
The shutdown is less a surprise than a signal. OpenAI is making a choice about where AI will matter most and where it won’t.
The Economics of Sora
Sora’s core problem was viability. AI video generation remains one of the most compute-intensive applications in the current AI stack.
Unlike text or image generation, video requires sustained processing across multiple frames, increasing both latency and cost. At consumer scale, that quickly becomes untenable unless offset by strong monetisation or massive engagement.
While millions downloaded the app, usage did not translate into sustained revenue. This mismatch, high compute, low monetisation, is becoming a familiar pattern across consumer AI products. It is also forcing companies to rethink where to deploy their most expensive models.
For OpenAI, the answer is becoming increasingly evident: it is not in social apps.
A high-profile partnership that didn’t survive
The shutdown has already disrupted broader ambitions.
In December 2025, The Walt Disney Company reached a three-year agreement with OpenAI to bring its characters to Sora’s AI video generator, with plans for a $1 billion investment.
The deal was positioned as a major step toward AI-driven entertainment experiences.
The deal is now off the table.
A Disney spokesperson told The Hollywood Reporter the agreement would not proceed following OpenAI’s shift in direction, adding that the company “respects OpenAI’s decision to exit the video generation business.”
The collapse of this partnership underscores how quickly priorities are shifting, not just for OpenAI but across the AI ecosystem.
Behind the shutdown is a deliberate move by the company to reposition itself as an enterprise AI company.
OpenAI has begun prioritising long-term, high-value contracts over experimental consumer platforms. It has signed multiyear partnerships with Accenture, Boston Consulting Group, Capgemini, and McKinsey & Company. These partnerships are stated to focus on integrating AI agents into core business workflows.
This is not just about software. OpenAI is deploying forward-deployed engineers alongside consulting partners to embed AI systems into operations such as software development, sales, and customer support. Early enterprise users include Intuit, State Farm, Thermo Fisher Scientific, and Uber.
The shift is part of a larger industry reality: enterprise AI revenue remains unpredictable, while consumer app revenue is not.
Anthropic’s Growing Lead
The pivot is also defensive. According to Ramp’s March 2026 AI Index, Anthropic now captures 73% of first-time enterprise AI spending, up from 50% since January. Businesses are reportedly 70% more likely to choose Anthropic over OpenAI for new deployments.
The advantage extends to developers. The 2025 Stack Overflow Developer Survey shows that professional developers use Anthropic’s Claude Sonnet models at 45%, compared to lower adoption of competing tools.
Anthropic reports that engineers are integrating these models into 59% of their daily workflows, resulting in productivity gains of up to 50%.
Even as AI tools improve, friction remains. Around 66% of developers say they spend more time fixing “almost-right” AI-generated code, highlighting a persistent gap between capability and reliability.
Refocusing the Product Strategy
Internally, the company has already begun tightening its priorities.
In December 2025, CEO Sam Altman initiated what has been described as a “code red” effort to improve ChatGPT, citing competitive pressure from Google and Anthropic. As part of this shift, OpenAI has reduced investments in experimental areas such as advertising and shopping agents.
OpenAI is developing a ChatGPT desktop application designed to function as an autonomous agent, capable of tasks such as codebase refactoring and test execution. The company plans to expand this effort alongside workforce growth.
At the same time, it plans to double its workforce to 8,000 employees by the end of 2026, according to the Financial Times. Many of these roles are dedicated to enterprise deployment and integration.
The Cost of Scaling AI
Financially, the company is still under pressure. OpenAI is operating at a loss as it scales compute-intensive systems. According to The Information, its annualised revenue run rate is estimated at between $25 billion and $28 billion in early 2026, compared to Anthropic’s roughly $19 billion.
Internal projections suggest profitability may not arrive before 2029, when revenues could reach $125 billion, though these figures remain unconfirmed.
To sustain expansion, OpenAI is exploring partnerships with private equity firms such as TPG and Bain Capital, while also preparing for a potential IPO window later in 2026.
What Sora’s Shutdown Really Means
Sora’s closure is a recalibration.
Consumer AI, especially in high-cost formats like video, is proving harder to monetise than expected. Enterprise AI, by contrast, offers clearer returns, tighter use cases, and customers willing to pay.
For OpenAI, the message is pragmatic: scale where the economics work.
The bigger question is what gets left behind. If even a high-profile, well-funded experiment like Sora cannot survive, it suggests that the next phase of AI is increasingly about invisible systems embedded deep inside businesses.