A $200 ChatGPT subscription could cost OpenAI $14,000 if you actually used it to its full potential

[Squid Surprise]

Nope…

HKEPC, a Hong Kong-based tech site, shared a video of Huang making his rounds through the Computex 2026 showroom floor, where he was signing and interacting with hundreds of industry attendees, fans, and media. However, questions arose online when Huang was given a wallet from a HKEPC editor to sign, but instead of signing the front or back, Huang opened the wallet to sign NTD (New Taiwanese Dollar) notes the editor had inside. In a series of X posts, the HKEPC editor explained that Huang emptied the wallet to sign NT$1,000 bills to give to showgirls working at the booth.

The editor wrote in a post, "First 2-3 bills? I was holding it together. By the 3rd? Started sweating. Then he dropped all 10 of them. I froze. 'Bro, at least leave me one for the taxi home...' " The editor added that shortly after the wallet was emptied, Huang returned and paid him NTD$1,000 from his own wallet, and also signed a NTD$1,000 note, along with the wallet itself. The editor added that they started with NTD$7,700 in their wallet, and Huang returned NTD$10,000, leaving the editor with a NTD$2,300 profit.

Read more: https://www.tweaktown.com/news/1120...wner-and-paid-him-back-with-profit/index.html

It’s easy to fact check - try it sometime

 

[Steveb8189]

You would need a pretty powerful computer to run a model that has decent capabilities. This is just offsetting the water usage from data centers to consumers. Everyone will need to hire a plumber to run a water hose to their PC which will be great business for plumbers, but not very cost effective for the consumer, especially if the water rates are high in the area.

You can build a "pretty powerful computer" for the $50k the OP mentioned!

4 x Ryzen AI Max+ 395 should do it without crazy cooling requirements or power consumption. You're still going to be limited to a 235B model though so not quite at Opus

 

[OortCloud]

It’s easy to fact check - try it sometime

You're right - I apologise. I'd heard it from Gamers Nexus I think?? that he basically stole peoples money and called them poor.. I watched the video, it's not the case. Having said that your final sentence is that awful form of communication only found on internet forums...

 

[Vanderlinde]

Obviously they train on these subscriptions, and shove in ads. It's not pure loss here. Many times I encounter an averege prompt to take over 5+ minutes to finish. We're really getting the money's worth.

 

[Watzupken]

Yeah but not everyone can keep AI pegged at 100% even during work hours. My own use case requires diligent supervision so when I'm off sleeping or motorcycling it's sitting idle. If you aren't constantly running agents to constantly prompt the AI company is doing fine.

The other factor is that inferencing is getting cheaper. More efficient models, processes, and hardware will keep driving down the cost.

Finally supply and demand. As AI companies adjust their tokens to realistic prices it incentivizes the biggest users to A use less or B switch to local models.

While not everyone will be using 100% of the tokens, the article mentioned that anything beyond 11% will cost the AI service provider. That also means that current subscription cost is way way below the actual cost of providing the service. As I mentioned in another article, the business model is not sustainable. It may be true that inference cost is decreasing (though I doubt that is the case unless there is evidence to support this and not some AI company CEO saying that), but that still does not mean it will come down meaningfully any time soon to make these companies profitable. We are currently in the 4th year where AI company CEOs have promised they will be profitable, AI will be so good/ become sentient, etc, but based on observation and experience, I don't think either happened. In fact, when using these AI solutions, it is good at looking up and summarizing data in general, but still often making mistakes.

 

[ScottSoapbox]

While not everyone will be using 100% of the tokens, the article mentioned that anything beyond 11% will cost the AI service provider. That also means that current subscription cost is way way below the actual cost of providing the service. As I mentioned in another article, the business model is not sustainable. It may be true that inference cost is decreasing (though I doubt that is the case unless there is evidence to support this and not some AI company CEO saying that), but that still does not mean it will come down meaningfully any time soon to make these companies profitable. We are currently in the 4th year where AI company CEOs have promised they will be profitable, AI will be so good/ become sentient, etc, but based on observation and experience, I don't think either happened. In fact, when using these AI solutions, it is good at looking up and summarizing data in general, but still often making mistakes.

There are many of articles that show inference cost is decreasing but it's basically the normal march of technology to solve problems that people will pay for the solution.

Nvidia, AMD, Intel and others are pushing the latest chips' cost per token for good reason. Obviously they each pick the most favorable metric in that regard but the trend is undeniable. Further, they are all trying to optimize on the software side, such as handing off easy tasks to smaller models etc.

Can they get to profits at $20/mo in the near term? Probably not unless they can convert many free users. But profit from $100-200/mo "prosumer" users is likely doable in the next year or so. The real issues are the hard core enterprise users that are coding agents to burn tokens for them and free people that cost a lot and will never turn into money.