Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchOpenAI’s Sora video generator reportedly cost about $1 million a day to operate before the company discontinued its app and API on March 24, 2026. That estimate is striking, but it is not the same as an audited loss attributable to Sora—and it does not explain OpenAI’s much larger company-wide deficit.
The clearest interpretation is narrower: Sora appears to have been a technically impressive but economically unattractive use of scarce computing capacity. Its ongoing costs reportedly outweighed the direct revenue and strategic value OpenAI was receiving from the product.
As an Amazon Associate I earn from qualifying purchases.
The short answer
The Wall Street Journal reportedly estimated that operating Sora cost OpenAI approximately $1 million per day, according to TechCrunch’s account of the reporting. A simple calculation would put that at roughly $30 million per month, $90 million per quarter, or $365 million per year—but those are annualized estimates, not disclosed expenses.
Separately, reporting based on shareholder financial disclosures put OpenAI’s first-half 2025 revenue at about $4.3 billion, research-and-development spending at about $6.7 billion, operating loss at roughly $7.8 billion, and cash burn at approximately $2.5 billion. Those figures cover OpenAI as a whole, not Sora. The Information reported the figures in its coverage of OpenAI’s first-half results; Reuters-based reporting also put OpenAI’s cash and securities at roughly $17.5 billion at the end of that period (summary via Yahoo Finance).
#1 Best Overall
Sora was discontinued on March 24, 2026, amid reports of heavy compute consumption, weak or declining engagement, and pressure to reserve computing resources for higher-priority products and research (Axios; CBS News).
What the reported numbers actually mean
| Measure | Amount | What it describes |
|---|---|---|
| Reported Sora operating cost | About $1 million per day | An estimate attributed to later reporting, not an audited Sora statement |
| Annualized Sora estimate | About $365 million | Simple math assuming the daily estimate stayed constant for a full year |
| OpenAI first-half 2025 revenue | About $4.3 billion | The entire company |
| OpenAI first-half 2025 R&D spending | About $6.7 billion | The entire company |
| OpenAI first-half 2025 operating loss | About $7.8 billion | The entire company |
| OpenAI first-half 2025 cash burn | About $2.5 billion | Cash consumed by the entire company during the period |
These categories should not be collapsed into one headline number:
- Cash burn is the net cash consumed over a period.
- Operating loss is revenue minus operating expenses and may include non-cash accounting items.
- Compute cost covers resources used to train, serve, and host models.
- Research spending is broader than Sora and can include model development, safety, infrastructure, robotics, and unrelated research.
- Revenue includes money from subscriptions, APIs, enterprise contracts, licensing, and other activities.
Therefore, saying that “OpenAI burned $1 million in cash every day on Sora” would overstate what is known. The reported figure is an estimated operating cost, and the public evidence does not show exactly how it was calculated or whether it represented only inference or a broader allocation of expenses.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Why generating video is unusually expensive
A text request can often be answered with a relatively limited sequence of generated tokens. Video generation requires the system to create and maintain a large number of images across time.
Each additional second, frame, and increase in resolution raises the computational workload. The model must also preserve temporal consistency: a person’s face, clothing, motion, lighting, and surroundings should remain coherent from one frame to the next. That is substantially more demanding than producing a single still image.
Rank #2
The cost is not limited to the first attempt. Users commonly generate multiple variations, extend clips, change prompts, or rerun a request when the result is unsatisfactory. A product that gives users generous or free access can therefore incur costs each time a user experiments, even when no finished video is downloaded or monetized.
Other expenses can include:
- GPU inference and model serving;
- training and fine-tuning newer versions;
- safety screening and moderation;
- video storage and delivery;
- app, account, and queue infrastructure; and
- engineering and customer support.
The reported Sora figure appears to concern ongoing operation, while OpenAI’s company-wide financial figures include both research and operating expenses. Those are different scopes.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Was Sora profitable?
There is no publicly available, audited Sora profit-and-loss statement in the cited reporting. That means nobody outside OpenAI can reliably calculate Sora’s gross margin, total revenue, or exact loss.
The basic economic problem is still understandable. A user’s payment to OpenAI cannot automatically be counted as Sora revenue. Subscribers may have joined for ChatGPT, image generation, coding tools, or other services. OpenAI would need to determine how much incremental subscription revenue was attributable to video, then compare it with the incremental compute and infrastructure consumed by video users.
The product may also have served free or subsidized users. In that situation, high engagement can increase expenses without creating a matching transaction. A viral product can be valuable for growth while remaining loss-making if the company cannot convert attention into subscriptions, advertising, licensing, or enterprise contracts.
Some online discussions have repeated a claim that Sora generated only about $2.1 million in lifetime revenue. That figure is not established by the cited primary or high-quality financial reporting, so it should not be used as the basis for calculating Sora’s losses.
Why launch an expensive product at all?
OpenAI may have viewed Sora as an investment rather than a product that had to be profitable immediately. Generative video offered several possible strategic benefits:
- establishing a lead in a fast-moving market;
- attracting creators and developers;
- demonstrating multimodal model capabilities;
- collecting information about prompts and user preferences;
- building a future licensing or enterprise business;
- creating a consumer distribution channel; and
- advancing research into world simulation, robotics, and physical-environment reasoning.
Those benefits can justify temporary losses if they produce enough growth, data, technical progress, or future revenue. But the trade-off changes when computing capacity becomes scarce. Every GPU assigned to Sora may be unavailable for ChatGPT, coding, enterprise workloads, model training, or frontier research.
This is the opportunity cost of compute. A product does not need to consume all of a company’s cash to be cut; it only needs to provide less value per unit of scarce capacity than the alternatives.
Why OpenAI shut down Sora
Contemporaneous reports point to a combination of factors rather than one decisive failure:
- high ongoing compute consumption;
- low or declining usage relative to that cost;
- weak direct monetization;
- competition with higher-priority OpenAI products for GPUs and staff; and
- a strategic shift toward core and enterprise offerings.
Axios reported that compute scarcity was a major consideration. TechCrunch’s summary of Wall Street Journal reporting described Sora as costly and insufficiently used (TechCrunch). CBS reported that OpenAI said its Sora research would continue in connection with world simulation and robotics (CBS News).
That distinction matters. Discontinuing the consumer app and API did not necessarily mean OpenAI abandoned the underlying models, research, or intellectual property. It meant the publicly available product no longer justified its costs and resource demands under the existing strategy.
What would determine whether an AI-video product works financially?
The important metrics are more specific than user counts or viral clips:
- Cost per generation: How much compute is required for each usable clip?
- Generation volume: How many attempts does the average user make?
- Free-to-paid conversion: How many users generate enough revenue to cover their usage?
- Retention: Do users continue creating videos after the initial novelty?
- Attributable revenue: How much subscription or API revenue comes specifically from video?
- Infrastructure efficiency: Can batching, queueing, or improved hardware reduce costs?
- Indirect value: Does the model support licensing, enterprise sales, or other products?
- Research value: Are the outputs advancing capabilities that matter elsewhere?
The central tension is likely virality versus unit economics. More users can be beneficial, but if each user receives heavily subsidized generations, growth can increase losses faster than revenue.
Free tools Windows power users keep installed
One-click scans. No signup required.
Does Sora’s shutdown prove AI video is a failed business?
No. It shows that one high-quality consumer video product was reportedly uneconomic under its particular pricing, usage, and infrastructure model.
Best Value
Other businesses may have different economics. A professional production tool can charge for workflow integration and editing features. An enterprise service may sell predictable API access under contracts. A cloud provider may bundle video generation with broader infrastructure spending. Competitors may also use different architectures, quality targets, pricing systems, or distribution channels.
The shutdown is better understood as a warning about mass-market AI-video economics. A product can be technically impressive, popular, and strategically important while still failing to earn enough incremental revenue to justify its compute consumption.
What this means for consumers and investors
For consumers, Sora’s end is a reminder that access to an AI feature is not the same as a durable standalone business. Subscription prices, usage credits, resolution limits, queues, watermarks, commercial rights, and API terms all affect the provider’s costs and the user’s experience.
Recommended Free Tools
For investors and industry observers, the episode reinforces the importance of separating:
- company-wide revenue from product-level revenue;
- operating loss from cash burn;
- reported estimates from audited figures; and
- technical progress from sustainable unit economics.
OpenAI’s reported first-half 2025 figures show why product-level claims must be handled carefully. The company spent billions on research and reported billions in losses across its business. Sora’s estimated annualized cost—about $365 million if the $1 million daily figure persisted—would be meaningful, but it cannot be treated as the explanation for OpenAI’s entire deficit.
OpenAI also reportedly projected that its losses could reach about $14 billion in 2026, according to The Information. That was a forward-looking internal-document analysis, not an audited result, and it further illustrates the scale difference between Sora’s reported estimate and OpenAI’s broader financial commitments.
What is known, estimated, calculated, and unknown?
| Category | Information |
|---|---|
| Reported | OpenAI’s first-half 2025 revenue, R&D spending, operating loss, and cash burn; the March 24, 2026 discontinuation of Sora. |
| Estimated | Sora’s reported operating cost of about $1 million per day. |
| Calculated | Monthly, quarterly, and annual equivalents obtained by multiplying the daily estimate. |
| Unknown | Sora’s standalone revenue, profit or loss, exact cost allocation, cost per video, and precise number of paying users. |
The evidence supports calling Sora an expensive and apparently poor near-term use of compute. It does not support claiming that Sora alone caused OpenAI’s losses, that OpenAI paid exactly $1 million in cash every day, or that generative video as an entire category is unprofitable.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
