What's up: Phantom data centers haunt the grid; AI's gender gap; AI-powered drone swarms enter the battlefield.

Good morning. As U.S. AI companies spend billions on infrastructure and talent in their quest for artificial general intelligence, Chinese companies are taking a more pragmatic approach, narrowing their focus on smaller, less expensive practical applications built on open-source AI models.

It could turn out to a winning path, especially given recent stumbles with GPT-5, initially touted as a major steppingstone on the path to AGI. 

The Journal reports: “Already in China, domestic AI models similar to the one that powers ChatGPT are being used, with state approval, to grade high-school entrance exams, improve weather forecasts, dispatch police and advise farmers on crop rotation, say state media and government reports.”

U.S. tech companies are developing plenty of practical applications using current AI. But China’s AI push is being backed by Beijing. In January, the central government unveiled an $8.4 billion AI investment fund focused on supporting startups. Read the story.

Whether one path or another is better is up for debate. In the end, all roads lead to AI. More on that below.

Phantom data centers haunt the grid. The Journal reports that data-center developers and tech companies are peppering utilities around the country with requests for service. Often, the same potential projects are being double, triple or quadruple counted by different utilities.

552: The number of requests Texas utility Oncor has received from large customers such as data centers or industrial facilities in its interconnection queue by the end of June, up 30% from the end of March.

OpenAI is scouting locations in India for a new data center of at least 1-gigawatt capacity, Bloomberg reports. The country is ChatGPT’s second-largest market by user numbers.

Amazon.com plans to invest about US$4.4 billion in New Zealand on the construction and operation of data centers, supporting the launch of a new cloud services region in the country, WSJ reports.

Cutting-edge AI was supposed to get cheaper. Good news for business users, the cost of AI inference is getting less expensive. While training huge models still sucks wallets dry, getting answers from existing models–inference–is going down by a factor of 10 every year, Ben Cottier, a researcher at Epoch AI, tells WSJ Tech Columnist Christopher Mims.

But while the price of tokens continues to drop, the number of tokens needed to accomplish many tasks is skyrocketing. Why? Reasoning. Those touted “thinking” models can deliver better responses, but spend a lot more tokens in the process, Mims notes.

That can have businesses, including software vendors that “buy AI by the barrel,” burning their AI credits more quickly, leading to tough choices: Eat the cost or pass it on to customers–neither situation is ideal.

One possible outcome: “Consumers could just use cheaper, less-powerful models that require fewer resources.”

Why do some workers refuse to work with AI? Skills gap and lack of access to training are often seen as primary reasons for why workers don’t immediately pick up the AI tools their employers poured so much effort and $$ into. But are there other factors at play?

AI and social dynamics. Researchers from King’s College London, Peking University, Hong Kong Polytechnic University and University of Hong Kong asked over 1,000 engineers from the same company to evaluate an identical code snippet.

The results. When participants believed an engineer had used AI, they rated that engineer’s competence 9% lower on average. The penalty was worse when they thought a woman used AI to write code.

Most revealing, the researchers found, were the reviewers most critical of the AI use: Engineers who hadn’t adopted AI.

Follow-up surveys revealed a fuller picture. “Many engineers actively anticipated this competence penalty and strategically avoided using AI to protect their professional reputations," the researchers wrote. "Those who most feared competence penalties in the tech industry—disproportionately women and older engineers—were precisely those who adopted AI least. (read the full story in Harvard Business Review.)

Women are not adopting AI at the same rate as men. Further studies found that women were about 20% less likely than men to use generative AI. “To close the gap, employers could make generative AI usage mandatory,” Rembrand Koning, an associate professor at Harvard Business School and one of the authors of a recent paper, tells the Journal. If men are the primary users, generative AI could exacerbate gender biases or stereotypes, he said.

The next frontier of battle. Ukraine is using artificial intelligence to allow groups of drones to communicate and make decisions independent of their operator, the WSJ reports.

U.K. digital banks Revolut and Starling are weighing acquisitions of nationally chartered banks in the U.S., at a time when many large U.K. fintechs are winning customers in their home market at a slower rate than before, the Financial Times reports.

The Journal’s Tim Higgins makes the case that Apple's iPhone, criticized for its glacial adoption of AI, remains as critical to the overall tech ecosystem as it ever was. And we can thank Elon Musk for that. His AI company and his social-media platform sued Apple recently, claiming it violated antitrust laws by giving preferential treatment to OpenAI.

News of a new AI chip and strong quarterly results helped send Alibaba shares soaring more than 19% Monday, CNBC reports. Alibaba said Friday that cloud revenue grew 26% in the April-June quarter on the back of surging demand for AI services.

After more than a decade together at Kraft Heinz, the food giant said it plans to split its business into two companies, unwinding an industry megamerger that married two packaged-food behemoths. (WSJ)

Nestlé said it dismissed Chief Executive Laurent Freixe with immediate effect following an investigation into an undisclosed romantic relationship with a direct subordinate that breached the group’s code of conduct. (WSJ)

The Trump family notched as much as $5 billion in paper wealth on Monday after its flagship crypto venture opened trading of a new digital currency. (WSJ)

A new Wall Street Journal-NORC poll finds that the share of people who say they have a good chance of improving their standard of living fell to 25%, a record low in surveys dating to 1987. (WSJ)

The WSJ CIO Journal Team is Steven Rosenbush, Isabelle Bousquette and Belle Lin.

The editor, Tom Loftus, can be reached at thomas.loftus@wsj.com.