Uber CEO Dara Khosrowshahi speaks at the Future Investment Initiative Institute Priority Summit in Miami Beach on Feb. 21, 2025. (Photo: Chandan Khanna/AFP/Getty Images)

Uber’s top executive says Elon Musk is unwilling to make Tesla’s planned robo-taxis available on his ride-hailing company’s platform.

“They want to build it alone, so to some extent in Austin, we and Waymo will be competing with Tesla when they launch,” CEO Dara Khosrowshahi said Friday at the FII conference in Miami. 

Uber and Alphabet-owned Waymo are currently working together on robo-taxi service in Phoenix and announced in September an expansion to Austin and Atlanta in “early 2025” using all-electric Jaguar I-Pace vehicles.

Meanwhile, CEO Elon Musk pledged to investors during a call last month that Tesla would begin its own robo-taxi service in Austin in June using its self-driving-equipped consumer models (e.g. Models 3 and Y) rather than the Cybercab it announced in October.

There are several factors to consider in the Texas ventures. 

For Uber, investor pressure on its human driver-heavy business model means the company (and rival Lyft) is chasing an autonomous future; it has entered into several partnerships—as well as made investments—in recent years to facilitate it. 

For Tesla, there is skepticism that the company will fulfill its lofty promises (Musk previously promised a million robo-taxis on the road by 2020), actually have its self-driving tech ready for use, and be able to supply the service with enough demand to support it. 

“We estimate that the U.S. market alone is a trillion-dollar opportunity as you commercialize [autonomous vehicles] at scale and bring the unit economics down,” Khosrowshahi said during a call with investors earlier this month. And when that happens? “You're going to see many, many more players get over the finish line.” —AN

For years, the U.S. chip sector complained that Washington’s chip controls would hurt their business and spur the creation of a Chinese competitor—yet these warnings rang hollow as revenue kept hitting records thanks to the AI boom. 

But recent earnings reports from equipment manufacturers show the U.S.’s broadside against China’s chip sector may be starting to bite.

Applied Materials, the largest U.S. chip equipment manufacturer, issued a lukewarm revenue forecast last week, citing the risk of new export controls from Washington. Revenue from China, the company’s largest market, dropped 25% in the most recent quarter to reach $2.2 billion.

Meanwhile Lam Research’s China revenue over the last three months of 2024 dropped 10% year on year to reach $1.4 billion. “Our sales to customers in China, a significant region for us, have been impacted and are likely to be materially and adversely affected by export license requirements and other regulatory changes, or other governmental actions in the course of the trade relationship between the U.S. and China,” it warned.

Equipment manufacturers, for now, are still reporting an increase in overall revenue thanks to the AI boom. But analysts warn these companies will sorely miss what China has: a lot of semiconductor manufacturing, and a consumer market eager to snap up the products that use those chips. 

“China is not just a semiconductor manufacturing hub,” says Bain & Company partner Moonsup Shin. “Around or more than 50% of the semiconductors manufactured by Chinese manufacturers are consumed in China.”

Worse, there’s no ready-made alternative waiting. Though regions like Southeast Asia are investing in local manufacturing to position themselves as neutral territory in a chip war, in practice, these up-and-comers lack both China’s manufacturing and consumer demand. —Lionel Lim

China’s DeepSeek released its market-roiling models as “open source,” but that term hits different in the AI world. Sure, DeepSeek allows free use and modification of the models, but it’s not like they published the code.

Well, now they say they’re going to. DeepSeek announced Friday that it would be open-sourcing five of its code and data repositories, starting this week.

“These humble building blocks in our online service have been documented, deployed and battle-tested in production,” it said. “As part of the open-source community, we believe that every line shared becomes collective momentum that accelerates the journey. Daily unlocks are coming soon. No ivory towers—just pure garage-energy and community-driven innovation.”

What does this mean for the projects that have been trying to replicate what DeepSeek achieved (i.e. low-cost training, with a novel technique for squeezing the smarts of a big AI model into a much smaller model)? That depends on what exactly DeepSeek releases.

“We will be very excited if DeepSeek open-source their reinforcement learning pipeline as the community is still grappling with several technical challenges such as figuring out how to scale the training to very large models,” Lewis Tunstall, a Hugging Face researcher who’s working on the Open-R1 project, told me.

“It's unclear whether they will be open-sourcing code for inference or training (or something else entirely), but in either case it's a net win for the open source community!” Tunstall added. “In general, open-source benefits from having multiple approaches to the same problem.” —David Meyer