OpenAI Is Getting Utterly Desperate

LLMs were supposed to change the world. We were promised they would cure cancer, augment scientists, make office jobs a thing of the past, and reach superhuman levels of intelligence. But now we can see that was all total bullshit. AI tools are too inaccurate to boost productivity through augmentation, let alone automate jobs or even resemble a thinking thing. To make matters worse, it is becoming increasingly clear, day by day, that these models are not going to improve, given that their returns are diminishing at lightning speed (read more here). On top of that, no AI company is even remotely profitable, and the market leader, OpenAI, is actually getting further and further away from profitability (read more here) as its costs skyrocket and its revenue growth stalls due to people realising these tools are legitimately crap (read more here). Despite these unavoidable facts, well over a trillion dollars is being poured into this industry, with the vast majority of those funds being funnelled into OpenAI, pushing the valuations of associated companies, like Nvidia, to unprecedented heights. If you headed an AI company, how would you square this hole? How would you rapidly expand revenue to at least pay for this investment and avoid a valuation collapse? Well, for the CEO of OpenAI, Sam Altman, the answer is AI browsers, erotica and AI social media sites! Unfortunately, I think Altman needs to take a serious maths lesson, because none of this makes sense, and these "solutions" will actually make OpenAI's problems twice as bad.

The Economics

OpenAI has an interesting problem. The only thing their product can do is mimic what already exists, as that is what generative AI does: it takes data and extrapolates trends within it to create something derivatively "new". It can inherently only create things that are worse than, or nearly as good, as things that are already out there. So, how can you compete in a market with a worse product? Well, by either undercutting the competition in price, flooding the market to drown out the competition, or doing both. That is how Uber and Airbnb grew their businesses, despite the fact that their product is arguably worse than booking a proper taxi or hotel. OpenAI is attempting to do exactly the same thing by allowing the rollout of"erotica" to verified adults, as well by launching its Atlas Browser and Sora social media site.

But here is the big issue: Uber & Airbnb's costs per unit of service decreased as they grew, so this costly, aggressive growth eventually paid off when economies of scale set in and they achieved profitability. The opposite is true for OpenAI; its costs increase linearly as it grows. The cost per output of an AI is the same no matter the number of users. It doesn't have any economies of scale. So, growing bigger is not a solution to the profitability problem and will, in fact, make OpenAI's losses larger.

Even if that weren't the case — and OpenAI could completely dominate these industries — there isn't enough revenue available to solve their problem!

Let me show you what I mean.

Atlas

Just over a week ago, OpenAI launched Atlas, an AI browser. It is pitched to directly compete with Google's Chrome. Basically, rather than having Google Search embedded in the browser, it has ChatGPT's search functionality. You can download Atlas and use these functions for free. But if you pay $20 a month, you get access to its other features, such as a ChatGPT-powered page summary and some very terrible agentic features.

These paid-for features have some serious problems. Even die-hard AI fanatics have ridiculed them, as they are riddled with security issues and the "agentic" features pretty much don't work, with no version of ChatGPT being even remotely up to the task. No one will pay for this level of inconvenience. But that is okay, because OpenAI is almost certainly losing a tremendous amount of money here. Again, OpenAI loses hundreds of dollars per month on its $200-per-month ChatGPT subscriptions. And, because people use their browsers far more than they use chatbots, that means OpenAI is almost certainly losing even more for every $20 Atlas subscription.

Honestly, I don't think the only reason OpenAI provides these features is because they want to say they have offered an agentic AI as promised, but by locking it behind a $20 subscription, they have effectively ensured only gullible AI bros can access it, presumably in hopes that the general public won't realise that this AI isn't "agentic" at all. In short, it's a PR stunt.

Really, this is an attempt to dethrone Google Search. The vast majority of users access search engines through their browser's default search engine, which is almost always Google Search, which makes $48.5 billion in revenue per year. By poaching users at the source — browser-based search — OpenAI can take a slice of this pie. Indeed, OpenAI has been trying to embed adverts and paid promotions into ChatGPT Search for a while now so they can take on Google at their own game.

But there is an immeasurably large problem with this plan.

Currently, each word ChatGPT generates costs them $0.0003, and a ChatGPT search responds with 30 words on average, meaning a single search query costs OpenAI $0.01. Embedding ads will increase the number of words and the cost per search, but let's be generous and stick with this $0.01 figure. Google receives more than five trillion searches per year. Let's assume Atlas is wildly successful and poaches all of these users — well, processing that many search queries would cost OpenAI over $50 billion per year!

In other words, it would cost OpenAI at least $1.5 billion more to replace Google Search than Google Search makes in revenue per year.

Now, let's also not forget that embedding ads in ChatGPT Search will increase search query complexity and the average number of response words, significantly increasing costs per search. Advertisers are also deeply wary of advertising within chatbot search engines, as the integration is clumsy, and there isn't the infrastructure to accurately target customers or ensure your advert isn't placed in a potentially damaging context. Again, AI gets this kind of stuff wrong constantly! So, to draw advertisers away from Google, OpenAI will have to undercut them substantially. In reality, we can expect these losses to be more like $10 billion a year.

This market is only worth $48.5 billion. But if OpenAI dominated it, because of how expensive their models are and how inefficient using them to advertise would be, it wouldn't make a profit; instead, it would generate huge losses.

"Erotica"

From December, OpenAI will allow verified adults to generate "erotica" stories and imagery from ChatGPT. But let's call this what it is: porn. For now, video-based porn generation is still off the table. However, I guarantee you that it won't be for long.

One of the significant issues with AI porn is that it can be used to create violating deepfakes. This is one of the main reasons OpenAI hasn't allowed porn generation. Even when that isn't the user's intention, ChatGPT will generate it anyway. The AI tries to create an identity by amalgamating the faces and bodies it has been trained on, but it often looks identical to a person in its training data. And, whether a deepfake is generated of a celebrity or a random person, it is no less violating and leaves OpenAI vulnerable to huge legal recourse.

So, the fact that they are now letting people generate images of porn heavily suggests they are testing the waters to see if they can get away with rolling out video porn generation. After all, the potential downside of porn images and porn videos is functionally the same.

And, if OpenAI wants to make money from this, they need video. $65 billion of the $100 billion annual porn industry revenue comes from video subscriptions. The rest is live streaming, advertising and the like, which OpenAI can't compete with. While an estimated four million pornographic images are shared online each day, the majority of the traffic and revenue comes from the estimated 73 million pornographic videos uploaded to the internet every year.

So, what if OpenAI pulled an Uber on the porn industry and dominated it? How much money would they make?

Well, it costs OpenAI customers $0.25 per image generated and $0.50 per second of video generated. Keep in mind, the true cost to OpenAI is far higher than this, as these services are being sold at a major loss, butwe can use this figure as a baseline. Unfortunately, AI-generated images and videos often require around five attempts to produce something passable. So, the cost per usable image is $1.25, while the cost per five-minute video is $750!

And how much would it cost to flood the market?

Well, generating four million passable pornographic images would cost $5 million, which isn't much in the grand scheme of things. However, in this market, these images are not revenue-generating. Indeed, AI-generated porn is already flooding the market and making nothing from it.

But generating 73 million five-minute pornographic videos would cost $54.75 billion!

So, you'd think that if OpenAI could flood the market and dominate the porn industry, it could make $10.25 billion in annual profit ($65 billion in revenue minus $54.75 billion in overhead).

But don't forget that it costs OpenAI significantly more to make these videos, as they are selling them at a loss. Indeed, the Financial Times' recent optimistic analysis of OpenAI's finances suggests that for every dollar in revenue, OpenAI spends two dollars. So, generating this many videos could cost OpenAI up to $109.5 billion, meaning an annual loss of up to $44.5 billion.

Let's also not forget that the overwhelming reason people pay for porn is to support their favourite pornstar. You can't do that with AI, as it struggles to reproduce the same "person" from one video to the next. Even "AI Actors" like Tilly Norwood don't look the same from one image to the other. With such a fluid, hollow identity, there is no connection that would make people pay. So, even if OpenAI flooded the market, they won't actually poach these customers.

And, like with the Search, trying to embed adverts in these AI-generated videos is not going to be viable. Advertisers don't want to be associated with deepfake porn or potentially problematic pornographic material. The current porn market makes so little from advertising, which means AI porn will make even less.

Then there is the fact that OpenAI will face legal recourse for making deepfakes. Again, they can try to prevent celebrity deepfakes, but due to the unoriginality of AI, it will certainly produce a deepfake of someone. Settling these court cases could easily cost millions, if not hundreds of millions a year, if done at scale.

Ultimately, can OpenAI make money off erotica?

Well, the porn industry makes nothing off pornographic images and stories, so OpenAI won't make anything here. But if they try to enter the lucrative video market, their costs will be too high, and even if they take 100% of the market, they would at best break even but most likely make annual losses in the tens of billions of dollars. Regardless, they can't take 100% of the market, as their product isn't a replacement for the online sex work this market mostly comprises, which is overwhelmingly driven by human pseudosocial connections — i.e., something OpenAI can't replicate.

Sora App

I have saved the worst until last. A month ago, OpenAI launched the free-to-use Sora 2 app. It's essentially a TikTok clone, but rather than user-filmed videos, users generate videos with OpenAI's latest Sora 2 model.

By now, you get the shtick. Let's see how many videos are uploaded to TikTok a year and how much revenue they generate, then let's see if OpenAI can flood the market, do the same thing and make a profit (hint: it can't).

As of 2025, 23.56 million videos are uploaded to TikTok per day, or 8.6 billion a year! This is generating $23 billion in annual revenue for TikTok.

The typical TikTok video is 30 seconds long. At Sora 2's current pricing, this length means it would cost $15 per video (again, the actual cost for OpenAI is likely higher). But again, it will take roughly five attempts to generate a passable video, so the cost per usable video is $75. Generating 8.6 billion videos will cost $645 billion.

In other words, if OpenAI tried to muscle TikTok out of the market and was able to dominate 100% of its market, it would run an annual loss of $622 billion a year!

Atlas, "Erotica" & Sora App Summary

Okay, so even if the Atlas browser dethrones Google, it will at best only lose a little per year, but more realistically, it will cost them over $10 billion a year in losses. If their erotica dominates the market, it won't fare much better. Erotic stories and images don't make the current porn industry any meaningful revenue, so they won't for OpenAI either. If they expand and try to generateAI-generated pornographic videos, as many others and I expect them to, and dominate that market, they will at best make just over $10.25 billion in profits. With a more realistic cost of generating these videos, it shows they would lose up to $44.5 billion a year. But, taking the reality of the market into context, OpenAI isn't likely to poach customers, as their product is different from what people are paying for. However, if their Sora app dethrones TikTok and dominates its market, even with unrealistically optimistic video-generation costs, they will run a staggering loss of $622 billion a year!

So, a realistic worst-case scenario for these three attempts to plug OpenAI's revenue gap will generate $676.5 billion in losses per year. The best-case scenario is just $10.5 billion in profit per year, where they shut down unprofitable Atlas and Sora app operations, somehow hit the wildly unrealistic market domination of the porn industry, and achieve unrealistically low video-generation costs.

Fundamentally, none of these new ventures are profitable for OpenAI. It doesn't matter if they are small-scale or if they try to flood the market and dominate; they will lose money. In fact, if they do dominate these industries — which they want to do, as that generates hype which pumps up the value of the company — they will face losses of over half a trillion dollars per year!

Summary

I have estimated that OpenAI requires $500 billion in additional revenue by 2030 to break even on its current AI infrastructure investment. This is backed up by a separate study, which found that the AI industry, which OpenAI utterly dominates, will have $800 billion in revenue in the short term or break even by 2030.

This is why OpenAI is getting desperate. Their models are not getting better, and studies are showing that just shoving more cash, data and infrastructure at them, like OpenAI is doing, won't make them better. Meanwhile, studies and real-world examples show they can't automate jobs or even augment them successfully. That was meant to be their route to revenue growth, to fill this fatal $500 billion hole, and everyone can now see that was just a fantasy all along.

So, OpenAI is desperately trying to find other markets to fill the hole. It doesn't matter what market it is. And the best they could come up with is already a dead end. In fact, if they dominate these markets, as their investors and market value expect them to, it won't fill the hole but more than double it.

OpenAI is not stupid. They know this. They know Atlas, erotica and the Sora app won't save them from their inevitable cash crunch. But they know it will delay it. These aren't products or business plans; they are fodder for the AI hype machine. Their promises of worker replacement are dead on arrival (read more here); criticism, accountability, and a value crash are just around the corner. These new directions provide a new narrative, a way to escape the crushing reality. They are just as much, if not more, of a dead end. But that doesn't matter. For a while, it will mitigate scepticism and accountability while enabling more speculation, pushing their value up even higher, enabling them to borrow more, sell more stocks, and keep this sinking ship above the water for a little longer. When this speculation is crushed by reality, there will be another direction, another hollow promise for the future. And this cycle will only end when the AI bubble pops and the cash rug is pulled from under OpenAI.

There is only one word that sums this all up. Pathetic.

Thanks for reading! Don't forget to follow me on YouTube, Bluesky, and Instagram, or support me over at Substack.

(Originally published on PlanetEarthAndBeyond.co)

Sources: TR, Futurism, Bain & Co., SEL, CIOC, Themeisle, Wired, SEL, SEJ, SEL, BBC, GlobalGPT, Zipdo, SendShort, BOA, Will Lockett, Will Lockett