The emergence of ChatGPT late last year sparked a surge of interest in new generative artificial intelligence technologies, and today almost every major software vendor offers its solutions based on language models.
However, despite the tremendous enthusiasm from both technology providers and customers, no one has yet figured out how to capitalize on these powerful new products. On the contrary, as the Wall Street Journal reports today, companies such as Microsoft, Google and OpenAI that offer generative AI capabilities to their users are believed to be losing huge amounts of money on them.
For example, one of the first generative AI products was GitHub Copilot, a service used by programmers to create, fix and explain program code. Microsoft, which owns GitHub, claims the service has more than 1.5 million users and writes about half of the code they generate.
But even with the large number of customers, GitHub has become a huge money pit, a person familiar with the numbers told the Journal. GitHub charges users $10 a month to use Copilot, but on average loses about $20 per customer.
Generative AI is an expensive technology because it can take years to train and fine-tune models, and even after that they require huge resources for day-to-day operations. “They require enormous computing power,” said Jean-Manuel Isaret, head of the marketing, sales and pricing practice at Boston Consulting Group. “They require tremendous intelligence.”
In addition, there is an element of overkill in many use cases. For example, ChatGPT is powered by OpenAI’s GPT-4 model, which is considered one of the most powerful in the world. However, many corporate ChatGPT subscribers use it for very limited tasks. As the publication puts it, using GPT-4 to summarize an email is like delivering a pizza in a Lamborghini.
To stop the money drain, many companies are trying to develop less powerful models to perform simpler business tasks, while others simply plan to raise their prices. Microsoft, for example, is going to charge an extra fee of about $30 a month for the AI-enabled version of its Office 365 software suite. Currently, the cheapest version of Office 365 costs about $10 per month. The artificial intelligence features will be able to compose emails and PowerPoint presentations, automatically generate Excel spreadsheets and perform many other tasks.
Similarly, Google plans to offer users to pay $30 per month for generative artificial intelligence features in its productivity software, where the cheapest subscription currently costs just $6.
Meanwhile, other sources report that Microsoft is looking to develop less powerful models to address redundancy. For example, the company is building small and inexpensive AI models for Bing that will be designed for web searches only. Some of these models will reportedly be based on open-source AI from companies like Meta Platforms*.
Adobe is using a different tactic. It has created a credit system for its artificial intelligence imaging tool Firefly. According to the company, if a customer uses up their allotted monthly credits, the Firefly service will be slowed down to prevent overuse. “We’re trying to provide high value but also protect ourselves from a cost perspective,” Adobe CEO Shantanu Narayen told the publication.
The problem for Microsoft, Google and other companies thinking of raising fees for their artificial intelligence services is that they are delicately balancing supply and demand, as not everyone thinks the software is worth paying for. “A lot of customers I’ve talked to are unhappy with the cost of using some of these models,” said Amazon Web Services CEO Adam Selipski.
So far, the challenges of monetizing generative AI haven’t scared off investors, who have poured billions of dollars into the most promising startups this year. According to The Journal, OpenAI is discussing a stock sale that would bring its value to more than $90 billion, three times what it was worth at the beginning of the year.
However, many in the industry believe it is only a matter of time before investor enthusiasm wanes. When that happens, many will begin to scrutinize the costs of AI and how to profitably utilize the technology.
Is it okay to use runBlocking?
In this video I’ll talk about when it’s fine to use the runBlocking function from Kotlin coroutines and when you...
Mobile App Development Best Practices – 07.12
KSP2 Preview, Mastering in SwiftUI, How to implement gamification and more!
Gemini is the new foundation for artificial intelligence in Android
Foundation models are trained on a variety of data sources to create artificial intelligence systems that can adapt to a...
Google has released AlphaCode 2 based on Gemini
Google today, along with its Gemini artificial intelligence model, unveiled AlphaCode 2, an improved version of the AlphaCode code generator...
ColorfulX – Metal for crafting multi-colored gradients
ColorfulX is an implementation using Metal for crafting multi-colored gradients. ColorfulX Platform UIKit and AppKit platforms are generally supported. Due to MTKView not...
Mobile App Development Best Practices – 06.12
Power of enums, A New Foundation for AI on Android, developer dogmas and more!