There has been a breakthrough in artificial intelligence. A Chinese generative AI start-up called DeepSeek created a new AI model called DeepSeek-R1. This model, the company claimed, was built for about one thousandth the cost of the GenAI models built by the US tech giants. And because Western companies are not allowed to send cutting-edge chips and machinery to China, it was built using AI chips whose performance greatly lags the chips used by DeepSeek.
There has been skepticism about the veracity of these claims. As an industry analyst focused on supply chain solutions, I admit to some skepticism as well. As DeepSeek revealed details about how R1 was built, they reported that one tool they used involved compressing big AI models down into smaller ones. This made the models easier to run without, they claim, losing much in terms of performance. This technique has long been employed in supply chain planning optimization models. However, the history of this software market is that as computing performance improved, the models became much, much bigger. This is because bigger models did lead to significantly better optimization!
But Kevin Roose of the New York Times has had conversations with industry insiders. He points out that because the solution was built using open-source software, experts could examine its code. His conclusion was that while “it’s plausible that the company’s shoestring budget has been badly exaggerated, or that it piggybacked on advancements made by American AI firms in ways it hasn’t disclosed,” nevertheless, the “breakthrough was real.”
Much news coverage has focused on the geopolitical and stock market implications – Nvidia’s stock price plummeted.
But for companies using enterprise software, the news is not all bad. First, virtually every enterprise software firm is working to include AI, including GenAI, in their applications. In particular, GenAI will significantly improve user interfaces and can serve to find hidden connections other solutions have not been able to find. The vendors must pay OpenAI, Microsoft, and Alphabet (Google) to do this. And that cost gets passed down to the users. Cheaper AI means cheaper software.
Further, many large companies have been working with AI internally. Off-the-shelf software can’t cover all the nuances and operating processes of these companies. That is why customers are increasingly looking to work with enterprise software companies whose platforms can support “bring your own” AI. These internal skunkworks also become cheaper to operate.
Secondly, supply chain emissions are usually the most significant part of a company’s carbon footprint. As companies seek to track and reduce their greenhouse gas emissions, they commonly use an ESG framework that tracks Scope 1, 2, and 3 emissions. For a purchaser of data center services, the indirect data center emissions are considered Scope 2 because they primarily relate to the electricity purchased to power the data center. A generative AI–based prompt request consumes 10 to 100 times more electricity than a typical internet search query.
Data centers are also driving up electricity costs. They currently account for approximately 3% of total U.S. power demand, but this figure is expected to rise to 9% within the next decade due to the increasing use of artificial intelligence. The surge in energy demand from AI data centers could lead to a 70% increase in electricity bills for consumers and small businesses by 2029. This most affects locales where data centers are prevalent. As companies seek to open new facilities, particularly facilities where energy consumption is a significant portion of total costs, the cost of energy becomes a factor.
Mr. Roose also points out that “the results also raise questions about whether the steps the U.S. government has been taking to limit the spread of powerful AI systems to our adversaries — namely, the export controls used to prevent powerful AI chips from falling into China’s hands — are working as designed, or whether those regulations need to adapt to take into account new, more efficient ways of training models.”
Tariffs also play a central role in the Trump administration’s economic and geopolitical strategy, a stance that dates to their first term in 2016. Significant tariff increases on foreign goods have been proposed. Compliance with complex trade regulations has always been burdensome. The compliance organizations in firms engaged in global trade would welcome anything that makes their job easier.
