# No chips, no paradise: OpenAI shuts down Sora to stem $14 billion in losses

> OpenAI shuts down Sora due to the unsustainable $15M daily compute costs, prioritizing the Spud model and robotics ahead of its IPO.
> By José Maldonado · Artificial Intelligence · 2026-03-27

The global technology industry is undergoing one of the most drastic and sobering transformations in the recent history of advanced computing. On March 24, 2026, OpenAI, the organization that led the artificial intelligence revolution with ChatGPT, [**announced on X**](https://x.com/soraofficialapp/status/2036546752535470382)  in a completely surprise move, the definitive shutdown of Sora, its ambitious video generation platform.

The decision, made just six months after the mass launch of its standalone app, makes clear the collapse of a business model based on the indiscriminate burning of capital and energy to sustain experimental tools without a clear path to profitability.

The decision entails the immediate dissolution of one of the sector's most high-profile partnerships: the $1 billion agreement with The Walt Disney Company, which was intended to integrate iconic characters into the Sora ecosystem, but has ultimately been canceled without any real capital being transferred between the parties.

## The end of capital burning

The closure of Sora marks the end of what analysts and economists in the technology sector have called the "era of virality at a loss." Over the past two years, the dominant narrative suggested that market dominance and massive model training would justify any level of operating expense. However, current financial and geopolitical reality has imposed a severe corrective.

**[Larry Fink proposes tokenization to curb the inequality brought by AI](https://observatorioblockchain.com/en/tokenization/larry-fink-propone-la-tokenizacion-frente-a-la-desigualdad-de-la-ia/)**

OpenAI is currently facing an unprecedented liquidity and efficiency crisis, with internal forecasts pointing to [**a net loss of $14 billion for the current fiscal year**](https://sybrid.com/resources/blog/openai-torching-14b-2026-crisis-chatgpt-ads/). This is despite having reached a projected revenue run rate of $20 billion for the same year.

This massive deficit is the result of a cost structure that has grown at a much faster pace than the monetization capacity of consumer tools. In fact, since its creation, OpenAI has burned more than $50 billion, and if the 2026 forecast comes true, that figure will reach $64 billion, and could reach as much as $115 billion in 2029 if the pace continues.

Hence, Sam Altman's decision to shut down the project responds to an existential need to optimize the financial balance sheet ahead of the company's long-awaited initial public offering (IPO), expected for late 2026 or early 2027, where fiscal discipline will be the main success indicator for Wall Street investors.

## The compute and inference barrier

From a technical perspective, the closure of Sora reveals the fundamental limitations of the "Diffusion Transformers" (DiT) architecture when attempting to scale for mass use. Although Sora amazed the world in its initial demonstrations for its ability to understand basic physical laws and maintain spatial coherence, the computational intensity required for inference proved to be an insurmountable bottleneck.

Unlike language models (LLMs), which process text tokens sequentially and relatively efficiently, video models must compute thousands of pixels across multiple time layers, ensuring that each frame is coherent with the previous one.

## Relentless demand

This process consumes a massive amount of graphics processing units (GPUs), specifically NVIDIA H100s and the new Blackwell chips, which are currently the scarcest and most expensive resource on the planet. Sam Altman admitted internally that the company "had run out of GPUs" sufficient to sustain both the training of future frontier models and Sora's daily operation.

**[Forget apps: in China you already chat with the AI that works for you on WeChat](https://observatorioblockchain.com/ia/en-china-ya-chateas-con-la-ia-que-trabaja-por-ti-en-wechat/)**

This energy and hardware demand has led the company to prioritize other projects. Every chip dedicated to rendering a viral video on Sora was a chip that was not available to respond to queries from ChatGPT's enterprise users or to advance the training of the model with [**codename "Spud"**](https://www.theneurondaily.com/p/sora-lasted-6-months-disney-s-1b-deal-lasted-3).

. And in a market where competitors like Anthropic have gained ground by offering extremely efficient coding tools with Claude, OpenAI could not afford to waste its "compute moat" on a recreational product.

In fact, Anthropic at this moment is a company with a much greater potential for economic sustainability, and could achieve it this 2026 if it manages to exceed $24 billion in revenue as planned (with planned spending of around $19 billion).

## Decentralized infrastructure (DePIN) in the face of the compute crisis

However, Sora's closure due to a lack of profitability in GPU usage has laid bare the vulnerability of companies that depend exclusively on centralized and costly infrastructure.

In this context, the digital asset ecosystem has begun to offer alternative solutions through Decentralized Physical Infrastructure Networks (DePIN). These platforms allow AI developers to rent computing power from a distributed network of independent providers, often at a fraction of the cost of traditional cloud providers such as Microsoft Azure or Amazon Web Services.

Akash Network and Render Network are two of the most prominent projects that have seen an increase in their relevance following OpenAI's announcement. Akash, for example, has implemented a critical upgrade called **"Burn-Mint Equilibrium" (BME)**, which introduces a deflationary mechanism for its native token, AKT, directly linking the asset's value to actual network usage for AI workloads.

**[Peter Steinberger, the AI luthier, joins OpenAI to close the chat window](https://observatorioblockchain.com/ia/peter-steinberger-llega-a-openai-para-cerrar-la-ventana-del-chat/)**

By decentralizing the GPU supply, these networks eliminate the massive profit margins of hyperscalers, allowing resource-intensive generative video AI projects to be economically viable.

## Toward a sustainable AI economy

In any case, Sora's closure represents a necessary moment of maturity for the global technology ecosystem. The era of burning billions of dollars on "technological vanity" projects is over, giving way to a consolidation phase where efficiency, security and return on investment are the only valid indicators of success.

**[Is Microsoft the real winner of OpenAI's new structure?](https://observatorioblockchain.com/ia/es-microsoft-el-ganador-de-la-nueva-estructura-de-openai/)**

OpenAI, by sacrificing Sora, has shown that it is willing to make painful decisions to ensure its survival and its leadership on the path toward artificial general intelligence (AGI) and its imminent stock market listing. The reallocation of resources toward robotics and the "Spud" model suggests that the future of AI is not on the screen, but in the ability of systems to act autonomously and productively in the physical and digital world.

For the cryptoasset sector, the end of the era of exuberance in centralized AI is the beginning of an era of solid growth for decentralized infrastructures that can offer real solutions to the chronic problem of compute costs. The industry has learned that virality is noise, but infrastructure is value.
