Article reprinted from: Machine Heart
The income is not enough to make ends meet, and the family is the only supporter.
Image source: Generated by Unbounded AI
People always mention that the implementation of large models is accelerating, but few people can conduct in-depth investigations into the actual cost of implementation.
The reality is that even those technology giants with "rich financial resources" will have trouble making ends meet.
Today, a report in the Wall Street Journal pointed out that although GitHub Copilot charges $10 per month, Microsoft still loses more than $20 per user per month on average.
This highlights the daunting economic challenges of providing AI services, which are not only expensive to build but also very expensive to operate. The Wall Street Journal compared it to this: "Using (AI) to summarize emails is like having a Lamborghini deliver pizzas."
Why did Microsoft suffer "heavy losses"?
Over the past year, the artificial intelligence boom has brought huge profits to many companies. For example, Nvidia is one of the companies that has made the most profit. Nvidia's AI chips such as H100 are one of the most valuable resources in the technology field at present, because they are essential for training large language models such as ChatGPT.
Given that demand for Nvidia GPUs in data centers of many technology companies, including Microsoft, has risen sharply in 2023, which has brought huge profits and stock surges to Nvidia, it has also made the chip shortage problem more serious, indirectly increasing the operating costs of large models.
Meanwhile, Microsoft has struggled to monetize its AI services, with setbacks, including heavy losses from one of its first generative AI services, GitHub Copilot.
GitHub Copilot was first launched in June 2022, allowing developers to let the Copilot AI assistant help write code for applications and services. The paid model of the service is $10 per month and $100 per year. Verified students and established open source project owners can use GitHub Copilot for free.
A source said that based on data from the first few months of this year, Microsoft is losing more than $20 per user per month on average, with some users costing as much as $80 per month.
Microsoft plans to launch Microsoft 365 Copilot for enterprise users on November 1. This AI service will help users use the company's productivity applications such as Word, PowerPoint, Excel, Outlook, etc., and the monthly usage price for each user is US$30.
Microsoft has been looking for more affordable ways to run its AI services, and one of them may be to make its own AI chips instead of buying them from Nvidia.
Developing AI chips is an expensive and time-consuming process. As early as April this year, there were reports that Microsoft was developing an AI server chip code-named "Athena", and the development of the chip began around 2019. Microsoft hopes that Athena will be comparable to Nvidia's H100, and Athena was jointly tested by Microsoft and OpenAI teams.
Recent reports say Microsoft will officially announce the AI chip at its Ignite conference on November 14.
Additionally, there are reports that Microsoft is trying to reduce its reliance on OpenAI by developing its own large language models, and given the rising operating costs, Microsoft has asked researchers to build smaller and cheaper alternatives to the OpenAI GPT model.
It seems that perhaps the best way to make money in AI is not to build big models, but to build the chips or data centers that run them. This is the same reliable method in any gold rush: just sell shovels.
ChatGPT revenue growth slows
At the same time, OpenAI is also facing slowing revenue growth.
There has been both good news and bad news for OpenAI’s mobile business recently. On the one hand, ChatGPT’s app installations and revenue continue to grow, setting new records in September: 15.6 million downloads of the app worldwide on iOS and Android, and total revenue of $4.58 million.
Revenue growth came from the ChatGPT+ subscription service, which offers faster response times, priority access during peak hours, and early access to new features and improvements, though it's priced pretty high at $19.99 per month as an in-app purchase.
ChatGPT's total revenue in June was $2.1 million, which increased to $2.74 million in July, $3.81 million in August, and a new record of $4.58 million in September.
However, according to the latest data from market intelligence firm Appfigures, its revenue growth has begun to slow. After maintaining revenue growth rates above 30% for the past few months (31% in July and 39% in August), as of September, that growth rate has fallen to 20%. Slowing revenue growth could be the first sign that ChatGPT is approaching saturation.
Surprisingly, despite its widespread popularity, ChatGPT is not the AI application with the largest revenue. According to Appfigures, a competitor called Ask AI earned more revenue based on a large number of advertisements, from $6.48 million in May when the mobile version of ChatGPT was launched to $6.55 million in August, and then fell slightly to $5.51 million in September, but this figure is still higher than ChatGPT's revenue.
Of course, net revenue might be another story, given Ask AI’s ad spending. As for ChatGPT, it had net revenue of about $3.2 million in September after Apple and Google took their cut of in-app purchase revenue.
In addition to the record revenue, ChatGPT also received 15.6 million installs in September, bringing its total installs to 52.2 million, according to Appfigures estimates. Google Play drove the growth in downloads, with 9 million downloads in September, while the App Store had 6.6 million downloads.
Reference Links:
https://www.neowin.net/news/microsoft-reportedly-is-losing-lots-of-money-per-user-on-github-copilot/
https://techcrunch.com/2023/10/09/chatgpts-mobile-app-hit-record-4-58m-in-revenue-last-month-but-growth-is-slowing/