The Cost of Genius

From assignments to relationship advice and everything in between, AI and more specifically GPT tools, have taken the world by storm. However, the saga of Artificial Intelligence could very well be over before it even starts. The exponentially large running costs of operating Generative Pre-Trained Transformers (GPT) make it a field unlikely to be profitable in the foreseeable future, as even pioneers in the field struggle to stay afloat.

Released in November 2022 to the general public, ChatGPT emerged as a messiah, an ever-dependable assistant you could rely on for both basic and advanced tasks. Reaching 100 million weekly viewers within a staggering sixty days, it soon propelled itself to the top of the AI assistant market, constituting about 62% of the market. It has been on an ever-steepening rise, recently hitting 400 million active weekly users. However, beyond the numbers themselves lies the market’s struggle to simply sustain itself.

An AI of its kind requires a large amount of data training to reach that level of intelligence. According to some estimates, it cost OpenAI a whopping $100 million to train its latest model, GPT-4. But that’s only part of the story with these whopping numbers being only a chunk of the costs inflicted on the company. Facilitating such a large influx of requests and prompts on a daily basis requires a massive number of servers to function. Located in San Antonio, the servers have raised daunting questions about the sustainability of the model, something yet to be addressed by its parent company, OpenAI. Every single prompt is estimated to cost at least 10 ml of water, and when operated on the scale it runs at, inexplicable amounts of valuable resources are wasted. The horrors do not end here, though. Every day, the company loses in the range of $100,000-700,000 just to stay afloat. These numbers do not make for very optimistic reading, since most estimates indicate OpenAI to be nearing bankruptcy this year.

While the accessibility of ChatGPT has been lauded and been the driving force behind its large market size, it might be exactly where the hard work comes undone. The business model of ChatGPT doesn’t exactly facilitate profit prospects. While the subscription-based model may attract some customers with its increased intelligence and efficiency, most users prefer to use the standard, free-to-use version instead. According to OpenAI, they have 10 million subscriptions worldwide, which amounts to only a fractional value compared to the weekly number of people who use these services. The donations they receive from corporations such as Microsoft may be enough to keep their head out of the water and help them remain operational for now, but are unlikely to support long-term survival prospects.

To put it plainly, OpenAI must ensure that they are able to monetse their customers, because the services they provide are certainly worthy of remuneration. While current subscription models may be too expensive to lure a majority of their audience, an alternative model may be more effective- where payment plans are cheaper to use, but required to function the tool. This, though it may lure away some customers, will help them retain their accessibility without compromising on essential financial stability.

Their financial woes do not necessarily extend to the broader field. A Chinese company trained their AI model, Deepseek, with a budget of just $3 million- an astonishingly impressive number compared to ChatGPT 4’s $100 million expenditure. The field is also exponentially growing- with 149 new models being launched in 2024. This competition means that ChatGPT is bound to lose some of its audience, with questions being raised regarding its speed and comparative intelligence compared to other supposedly ‘more intelligent’ models.

However, the truth is hard to ignore- OpenAI must rework their financial model or risk depriving the world of its favourite problem-solving assistant, not because of its lack of usefulness but because it simply isn’t sustainable.