OpenAI puts it in reverse: 1-year-old Sora is euthanized

Author: Xiao Jing, Tencent Technology

On March 24, 2026, OpenAI announced the shutdown of Sora.

This was OpenAI’s first standalone application launched after ChatGPT, officially released for only six months. On the same day, Disney announced the termination of its $1 billion investment and character licensing agreement with OpenAI, with neither side completing the fund transfer.

According to foreign media reports, OpenAI CEO Sam Altman told employees in an internal meeting that shutting down Sora was to “free resources for the next-generation AI models.” Altman also announced that he would shift his personal focus to fundraising and infrastructure development for computing power. A new model codenamed “Spud” within OpenAI has completed initial development.

Sora is not a failed product. After launching in September 2025, it surpassed one million downloads in ten days and briefly topped the App Store’s free charts, even faster than ChatGPT. However, according to mobile analytics firm Appfigures, Sora’s downloads peaked at about 3.33 million in November 2025 before rapidly declining, reaching approximately 1.13 million by February 2026. Throughout its lifecycle, in-app purchase revenue was only about $2.1 million.

Shutting down Sora is not the most important issue. The key point is that this marks OpenAI’s official acknowledgment that the “do everything” strategy over the past year has failed. And this admission comes rather late.

“Too many side quests, losing sight of the main storyline”

2025 was OpenAI’s most bloated year for product lines. Video generation model Sora, AI browser Atlas, hardware devices in collaboration with Jony Ive, and ChatGPT’s e-commerce features—Altman once compared this strategy to “betting on a series of startups within OpenAI.”

This logic was somewhat reasonable at the time. ChatGPT boasted 900 million weekly active users, and OpenAI seemed qualified to attack multiple fronts simultaneously, consolidating its position as the “AI era’s definers.” But internally, the reality was quite different. Multiple current and former employees reported that the proliferation of parallel projects caused strategic confusion, with computing resources frequently reallocated among teams. The organizational structure became increasingly chaotic, for example, the Sora team was placed under the research department, but in reality, it was responsible for one of the company’s most high-profile products.

On March 16, 2026, Fidji Simo, CEO of the application business, outlined a new direction during a company-wide meeting. Foreign media reported on the meeting notes, where Simo said: “We cannot miss this moment because we are distracted by side quests.” (“我们不能因为被副本任务分心而错过这个时刻。”) She further stated: “We really have to nail productivity in general and particularly productivity on the business front.” (“我们必须在生产力方面，特别是企业生产力方面做到极致。”)

Simo described the rise of Anthropic as a “wake-up call,” and said the company’s state was “very much acting as if it’s a code red” (“完全在按红色警报运转”).

With the shutdown of Sora, OpenAI’s new strategy boils down to two keywords: code and enterprise.

OpenAI is merging ChatGPT, coding tool Codex, and browser Atlas into a single desktop “super app.” Sora’s video generation features will no longer exist as an independent product, and the team will shift focus to long-term research on robots and world simulation.

Failure in coding, and the rise of Anthropic have created a technological gap

OpenAI’s strategic shift was directly driven by Anthropic’s rapid rise in the programming and enterprise markets.

According to fintech firm Ramp, as of February 2026, Anthropic accounted for about 73% of the spending among companies purchasing AI tools for the first time, while OpenAI only accounted for about 27%. In the overall enterprise AI subscription market, OpenAI still led with 34.4% versus 24.4%, but Anthropic’s month-over-month growth far exceeded OpenAI’s. In February 2026, Anthropic grew by 4.9%, while OpenAI declined by 1.5%, marking the largest single-month drop since Ramp began tracking AI enterprise procurement.

Ramp economist Ara Kharazian pointed out that a year ago, only one in twenty-five companies paid for Anthropic’s services; now, that ratio has increased to one in four.

[Chart: Growth curves of both companies since reaching $1 billion annual revenue—Anthropic about 10x, OpenAI about 3.4x]

In the coding segment, the gap is even more pronounced. According to Menlo Ventures, Anthropic’s Claude Code holds about 54% of the programming market, while OpenAI’s Codex accounts for only about 21%. After Claude Code’s public release in May 2025, it reached $1 billion in annualized revenue in just six months, and by February 2026, exceeded $2.5 billion. About 70%-90% of Anthropic’s engineering team’s code is produced by Claude Code, and Boris Cherny, head of Claude Code, said that he completes 100% of his daily coding tasks using the tool.

[Chart: Monthly enterprise AI procurement rate trend—OpenAI (about 34.4%) vs. Anthropic (about 24.4%)—OpenAI’s curve is trending downward]

This is not just a success in “programming.” In January 2026, Anthropic launched Cowork, a desktop agent tool aimed at non-developers, capable of handling spreadsheets, file management, and report generation. Four engineers completed development in ten days, with most of the code generated by Claude Code itself. After its release, the global SaaS sector saw a valuation evaporation of about $2 trillion.

In contrast, Anthropic’s strategy has always been focused: no audio, no images, no videos—only text, code, and enterprise scenarios.

This approach is very different from OpenAI’s, which has bet all its limited computing power on coding—proven to be the highest return on investment.

Problems extend beyond products—people, architecture, and the North Star may also be at risk

OpenAI’s deeper challenges go far beyond multi-front product strategies.

Since 2024, OpenAI has experienced a rare loss of core talent. Co-founder and chief scientist Ilya Sutskever left in May 2024 to establish Safe Superintelligence (SSI), which was valued at over $30 billion within a year.

Jan Leike, co-lead of the super alignment team, resigned at the same time, publicly criticizing OpenAI’s safety culture as having “given way to shiny products,” and later joined Anthropic. CTO Mira Murati left in September 2024 to found Thinking Machines Lab, while Chief Research Officer Bob McGrew and VP of Research Barret Zoph left on the same day. Co-founder John Schulman initially went to Anthropic, then joined Murati’s company.

According to Seoul Economic Daily, over the past year, more than 50 researchers and engineers have moved to competitors like Anthropic and Meta. By early 2026, only Altman and President Greg Brockman remained among OpenAI’s original 11 co-founders.

All these individuals were pillars of specific directions within OpenAI.

Sutskever was the soul of OpenAI’s technical direction, leading core breakthroughs from the GPT series to reasoning models like o1. His departure could mean the company has lost its most discerning judgment on “which direction models should evolve.” Murati is the key link in transforming research into products; without her, ChatGPT’s launch would be impossible.

When a company loses both its technical soul and product hub, who can make the right decisions? Recent photos show Sam Altman with a head full of white hair.

Capital, IPO, and strategic tug-of-war

OpenAI’s current situation is further complicated by a unique pressure: it might become the largest unlisted company in human history, with a staggering burn rate.

In February 2026, OpenAI completed a $110 billion funding round, with a pre-money valuation of $730 billion and post-money valuation of about $840 billion—making it the largest single-round private funding ever. Amazon invested $50 billion, Nvidia and SoftBank each invested $30 billion. But these figures come with many conditions: Amazon’s $35 billion investment depends on OpenAI achieving AGI or an IPO.

In 2025, OpenAI’s revenue was about $13.1 billion, but it lost roughly $8 billion in the same period. Foreign media reports forecast that 2026 losses could soar to $25 billion, with cumulative losses from now until 2029 possibly reaching $115 billion. OpenAI expects to turn profitable only in the early 2030s. Its cash burn rate in 2026 is about 83.3%, an extremely rare figure in startup history.

This capital structure forms a delicate but fragile closed loop: Nvidia is both an investor and GPU supplier; Amazon and Microsoft are investors and cloud service providers; SoftBank is a partner in the Stargate data center project. A significant portion of the funding is expected to flow back to investors as service fees. Altman acknowledged this concern in a CNBC interview but said, “This only makes sense when new revenue streams feed into the entire AI ecosystem.”

Under this capital structure, an IPO is an inevitable choice.

OpenAI aims to go public by the end of 2026. Before listing, it must tell a clear growth story to public investors. This is also the direct motivation behind the strategic shift to “cut side quests and focus on productivity.”

Meanwhile, Anthropic’s funding is equally impressive. It completed a $30 billion Series G round in February 2026, with a valuation of $38 billion. Its revenue structure appears healthier: about 80% from enterprise clients, with annualized revenue reaching approximately $14 billion, maintaining about tenfold growth for three consecutive years. Research firm Epoch AI predicts that if current growth continues, Anthropic’s annual revenue could surpass OpenAI’s in the second half of 2026.

The giants have not fallen yet

The competition is not over. OpenAI still holds significant structural advantages: over 800 million weekly active users, the world’s highest consumer AI brand recognition, and the recent $110 billion war chest. ChatGPT’s consumer subscription revenue continues to grow, with over 1 million enterprise clients, and the new model Spud is about to launch. Codex, the coding tool, has over 2 million weekly active users.

But the AI industry is at a paradigm shift—value is shifting from “whose model is the strongest” to “who can help enterprises improve productivity,” from general-purpose capabilities to Agentic AI systems capable of autonomously completing complex tasks.

In this paradigm shift, Anthropic has bet on the right direction, while OpenAI has bet everything but ended up not really betting at all.

The shutdown of Sora is a signal that OpenAI’s management has recognized the problem. But recognizing the problem and solving it are still separated by a long distance.

With the founding team nearly dispersed, capital obligations unprecedented, and competitors already pulling ahead, can OpenAI return to its original vision, stay true to its core, and make the right choices?

It took less than four years for OpenAI to go from a research lab to the world’s most valuable unlisted company. The next four quarters will determine whether it can transform from a “big” company into a “right” one. In extreme scenarios, it could even become a company heading toward decline.