The AI Build-Out Is Inflation’s New Driver

SPEAKER_1 (0:00)

The world of business is constantly evolving, and Comcast Business keeps you in step with secure AI-backed networking in more than 100 countries, powering 90 percent of the Fortune 500 and millions of small businesses. And behind it all, thousands of experts answering your call at 2 a.m. like it's 2 p.m. One partner, that's it, powering how business gets done for companies around the globe.

When you add it all up, no one does business like Comcast Business.

Luke Vargas (0:33)

Deadly earthquakes rock Venezuela. We'll get the very latest. Plus, Anthropic accuses China's Alibaba of illicitly accessing Claude to train its own AI systems and Meet Inflation's new catalyst, the AI Buildout.

Justin Lahart (0:49)

There's a lot of things that go into these data centers. There's cooling equipment, there's electrical equipment, there's batteries.

So, it puts cost pressures on a number of businesses and those costs are likely to get passed on to consumers.

Luke Vargas (1:04)

It's Thursday, June 25th. I'm Luke Vargas for The Wall Street Journal. And here is the AM Edition of What's News, the top headlines and business stories moving your world today.

Rescue workers in Caracas are working to free survivors in collapsed buildings after back-to-back earthquakes rocked Venezuela yesterday evening. The US Geological Survey said that a 7.2 magnitude earthquake was quickly followed by a 7.5 magnitude quake about 200 miles west of the capital.

Venezuela's acting president, Delci Rodriguez, said that 32 people were killed and at least 700 injured with those numbers expected to rise. Rodriguez declared a state of emergency and said Caracas' main international airport is closed because of damage from the quakes which were felt as far away as northern Brazil. She said Mexico, the Dominican Republic, El Salvador and Qatar have dispatched a rescue teams to Venezuela and she also thanked President Trump who said the US was ready to send help. American AI lab Anthropic is accusing China's Alibaba of carrying out a brazen campaign to harvest the capabilities of its frontier AI model, Claude. Reporter Jason Chow has been reviewing a letter that Anthropic sent to a pair of US senators about what it's calling a distillation attack. Jason, I'm not certain we've talked about distillation attacks here on the podcast directly before. Just how do they work and what exactly is Anthropic alleging here?

Jason Chow (2:48)

So distillation is simply the process of training AI models by learning from the responses of another usually more advanced model to replicate their capabilities. And so Anthropic is accusing Alibaba and its AI lab, Quinn, of training their model on Anthropic's Claude model illicitly.

Luke Vargas (3:11)

And this is happening on a mass scale. This is not just a few accounts doing this.

Jason Chow (3:15)

Yes. So in the letter, Anthropic said that they found that Alibaba created almost 25,000 fake accounts to access Claude, and that involved nearly 29 million exchanges between late April and early June this year. And so they're accusing Alibaba of targeting some of Claude's most valuable capabilities. So that includes agentic reasoning, software engineering, and long horizon tasks, which are some of the most difficult tasks that an AI model can perform.

Luke Vargas (3:47)

And Jason, Anthropic writes to these senators turning billions of dollars of American R&D into a subsidy for a geopolitical competitor. I'm having flashbacks to the sort of Chinese IP theft narrative, which maybe sort of diminished a bit in recent years. I mean, just as a way of trying to forecast how this letter is likely to be received in Washington, what is the mood like there on this issue? Have we seen the Trump administration weigh in on distillation?

Jason Chow (4:11)

Yes. So American AI labs, and that's not just Anthropic, you know, OpenAI as well, for example, have accused Chinese AI labs of accessing their models and training Chinese models on American frontier models without permission. And so that's been signaled to the Trump administration before.

The White House has issued a memorandum earlier this year, sort of warning Chinese organizations of creating these proxy accounts, and then using all types of jailbreaking techniques to distill American models. And they've already said that's unacceptable. But we haven't seen sort of further, more severe sanctions or penalties that's been imposed to some of these Chinese firms.

In a way, Chinese AI labs are roughly about, depending on who you ask, roughly about six months to at most a year behind American labs. And that gap is widely considered to be shrinking. And I've probably get said that it's a priority for them to make sure that the Chinese AI models do not surpass American models. And this is the latest sort of escalation of those tensions.