Big tech companies like Microsoft and Google are taking on the challenge of turning AI products like ChatGPT into profitable enterprises. report Wall Street Journal. Businesses are investing heavily in AI technology that can generate business notes and code, but the cost of running advanced AI models is proving to be a major hurdle. Some services, such as Microsoft’s GitHub Copilot, cause significant operational losses.
Generative AI models used to create text are not cheap to operate. Large-scale language models (LLMs), such as those that power ChatGPT, require powerful servers with high-end, energy-hungry chips. For example, we recently cited a Reuters report with an analysis claiming that each ChatGPT query can cost as much as 4 cents to run. As a result, Amazon Web Services CEO Adam Selipsky told the Journal that many business customers are frustrated by the high running costs of these his AI models.
The current cost challenge has to do with the nature of AI computations, which often require new computations for each query, unlike standard software, which enjoys economies of scale. This makes flat-rate models for AI services risky, as increased customer usage can increase operating costs and lead to potential losses for businesses.
Some companies are looking to reduce costs, while others continue to invest more deeply in technology. While Microsoft and Google have introduced more expensive AI-assisted upgrades to their existing software services, Zoom has made improvements to the costs by occasionally using less complex in-house AI models for some tasks. It has been reported that attempts have been made to reduce Adobe is tackling the issue of activity caps and pay-per-use, while Microsoft and Google typically stick to flat fees.
Chris Young, Microsoft’s head of corporate strategy, believes it will take longer to see a return on investment in AI as people figure out how best to use it. “We’re clearly at a point now where we have to translate the level of excitement and interest into real adoption,” he told the outlet.
Notably, the WSJ report notes that Microsoft’s GitHub CopilotHelping App Developers with Code Generation operates at a loss, despite attracting more than 1.5 million users and being integrated into nearly half of coding projects. Users pay a flat fee of $10 per month for the service, which costs Microsoft an average of more than $20 per month per user, according to people familiar with the matter. In some cases, individual power users are costing companies as much as $80 per month.
One reason why AI services are so expensive is that some companies are reaching for the most powerful AI models available. For example, Microsoft uses OpenAI’s most complex LLM, GPT-4, for many of its AI capabilities. GPT-4 is one of the largest and most expensive AI models and requires significant computing power to operate. WSJ quipped that using the model for basic tasks like email summarization is “like having a Lamborghini deliver your pizza,” and that using the most capable AI models is simple. This suggests that it may be excessive for certain tasks.
Along these lines, Microsoft has conducted research. cheaper alternative For the Bing Chat search engine assistant, including Meta’s Llama 2 language model. However, advances in AI acceleration hardware may reduce the cost of operating these complex models over time. It’s unclear whether these advances will come quickly enough to match this year’s AI hype cycle.
While there is still excitement in the sector, we may be nearing the peak before the reality check kicks in, WSJ reports. Some experts predict a more rigorous financial approach in the near future, emphasizes May Habib, CEO of the generative AI company Writer. He told the same media, “I think next year will be the year when the slush funds for generative AI will disappear.” This suggests that the industry may soon move from enthusiasm and experimental budgets to a phase where the focus is on whether these AI models can actually contribute to enterprise profitability.