Organize | Su Mi
Listing | CSDN（ID：CSDNnews）
From joy to app frenzy to negative press, ChatGPT's nine-month history doesn't seem to be a good way to survive.
Now, according to Analytics India Magazine, if OpenAI continues to burn money at the current rate, it is very likely to go bankrupt by the end of 2024.
There are three main factors for making such a judgment.
At a daily cost of $700,000, who will maintain the huge overhead of a large model?
All along, OpenAI has never revealed the true cost of running large models.
Everyone only knows that OpenAI, which was originally bent on promoting and developing friendly artificial intelligence to benefit humanity as a whole, shifted from non-profit to for-profit in 2019 amid confusion and accusations from the outside world.
Because at that time, whether it was developing general-purpose robots or researching large models, it required huge overhead, and in the summer of 2018, just training OpenAI's Dota 2 robot needed to rent 128,000 CPUs and 256 GPUs from Google for several weeks.
So, if it hadn't turned to profit, OpenAI might have been on the verge of bankruptcy at that time.
Today, Analytics India Magazine points out that just running the artificial intelligence service ChatGPT costs OpenAI nearly $700,000 a day.
Although Altman, the helm of OpenAI, is trying to make money by commercializing GPT-3.5 and GPT-4, OpenAI is still far from breaking even, and profitability is still a fantasy.
In fact, in May this year, The information reported that because large models require huge computing power resources and data, and the operation behind them also has to pay a lot of expenses, in 2022 alone, OpenAI lost a total of $540 million, in stark contrast to its revenue of only $28 million.
The carnival receded, ChatGPT traffic dropped significantly, who will promote the revival?
With operational costs of $700,000 per day, ChatGPT traffic continues to decline.
According to the SimilarWeb survey, a website data analytics tool, ChatGPT traffic growth has been declining over the past few months, with a month-on-month growth rate of 131.6% in January, 62.5% in February, 55.8% in March, a sharp decline to 12.6% in April, and a month-on-month growth rate of only 2.8% in May.
Since then, global desktop and mobile web traffic to ChatGPT sites (chat.openai.com) has declined by a total of 9.7% from May to June. Some people think that this is because a large number of students who use ChatGPT began to have summer vacation during this period, so there were many fewer users for a while; On the other hand, it may also be because OpenAI released the ChatGPT API, which led many third-party developers to build their own conversational bots instead of using ChatGPT prototypes.
However, the data further shows that ChatGPT still had 1.7 billion users in June, but by the end of July, the number of users had fallen to 1.5 billion, down 12% month-on-month.
Based on this, users from the X (formerly Twitter) platform revealed that the main reason why ChatGPT is becoming less and less used today is that many companies have begun to prohibit employees from using ChatGPT for work, but allow them to use APIs to apply large models to different workflows.
Meta and Microsoft hold hands, how to resist the external competitive pressure brought by the open source big model?
In the past few months, looking at the entire industry, countless startups have released large-model products, bluntly saying that they want to benchmark against OpenAI, but in fact there are very few that can really compete with it.
However, the rise of open source models has indeed made OpenAI feel great pressure.
A month ago, Meta released a new AI model, Llama 2, which is not only open source, but it can be used for research and commercial purposes for free.
What's even more unexpected is that Meta partnered with Microsoft to promote the adoption of Llama 2.
In this case, third-party developers seem to prefer the open, free, and free Llama 2 version to the paid, proprietary, and limited versions offered by OpenAI.
At this point, Analytics India Magazine also said in the report that AI developer Santiago tweeted, "I have spoken to two startups who are migrating from proprietary models to Llama 2." ”
Self-help: IPOs don't seem to work
Under internal and external troubles, what is the way OpenAI can help itself?
Analytics India Magazine reports that it is too early for any leading AI company such as OpenAI, Anthropic or Inflection to enter the initial public offering (IPO) market to inject more cash into the business. A report by Investopedia states that an IPO requires at least 10 years of operation and $100 million in revenue to be successful.
And, in addition to the above three challenges, Sam Altman has repeatedly publicly mentioned that the lack of GPUs hinders the company's ability to train new models.
Although Microsoft previously announced that it will expand OpenAI with "multi-year, multi-billion" investments, Analytics India Magazine analyzed that Microsoft's $10 billion investment in OpenAI is only enough to maintain the company's current normal operations.
In addition, OpenAI expects annual revenue to reach $200 million in 2023 and $1 billion in 2024, which seems unlikely because the losses will only grow.
To ask why losses are getting bigger:
- On the one hand, OpenAI's move to a paid version may make some money, but the expected revenue may come from users of the API and users of GPT-4-based chatbots or other products such as DALL-E2, where financial data remains murky.
- Secondly, OpenAI originally gathered a lot of top talents in the field of AI, and the remuneration paid by employees was also quite high.
- Moreover, according to The Verge, Microsoft's investment in OpenAI is premised that "Microsoft has the right to receive 75% of OpenAI's profits in the early stage until it earns back the $10 billion invested and Microsoft's early investment in the company of an additional $3 billion; Subsequently, when OpenAI's profit reached $92 billion, Microsoft's percentage of OpenAI profit dropped to 49%; Finally, when OpenAI's profits reach $150 billion, OpenAI will get 100% of the company." Therefore, when OpenAI is profitable in the early stage, the main profits will go into the pockets of Microsoft and other recent investors.
- In addition, OpenAI has begun to apply for a trademark for "GPT-5", which indicates that OpenAI may plan to launch another large-scale language model soon, which will require a large amount of money.
Therefore, in such a "burning money" situation, how OpenAI maintains daily operations and retains talent is a very challenging thing.
Based on this, Analytics India Magazine believes that if OpenAI does not get more funding soon, then the company may have to file for bankruptcy by the end of 2024.
User Take: Microsoft may take over
However, in the eyes of some users, OpenAI is currently making money through APIs, and many companies are also using it, and its running costs should not lead it to direct bankruptcy.
Some users also noted that "a lot of people here don't take into account other costs such as training, fine-tuning, inflated salaries, and most importantly competition from other cloud-based AI models, especially $0 or close to free AI models and their services." Eventually, the latter will catch up.
GPT-5 will take longer to train for OpenAI, and at the same time, they are currently unable to raise prices as OpenAI's AI models begin to degrade.
While a 2024 collapse is unlikely, the only outcome I can predict is that OpenAI will have to go public in the next 4-6 years or be 100% acquired by Microsoft. ”
Finally, some netizens are worried about the success of prediction and provide suggestions for OpenAI:
"If you reduce the functionality of the free model and increase the price of the Plus, then the revenue and profit margin forecasts within OpenAI will be very interesting."
For those of us who think it's a very useful, very cheap service, the current strategy doesn't seem to have any other goal than to make OpenAI the Google of AI.
I'd like to see OpenAI survive with a reasonable, realistic fee model, and I'm curious what they think the path ahead is – bankruptcy is clearly not their ultimate intention."
Reference:
https://analyticsindiamag.com/openai-might-go-bankrupt-by-the-end-of-2024/
https://news.ycombinator.com/item?id=37109394