The Morning Download: OpenClaw Shines on Nvidia’s Stage

Good morning. Nvidia’s massive GTC gathering in San Jose has all of the electricity and excitement we have come to expect, but this time the glow is coming from an external force.

Nvidia co-founder and CEO Jensen Huang bestowed the limelight on open-source AI agent framework OpenClaw. Nvidia’s new NemoClaw adds elements of enterprise-level support and safety.

“The importance is profound. It’s the most popular open-source project in the history of humanity,” he said. This is just getting started, but WSJ Leadership Institute’s Isabelle Bousquette explains what OpenClaw agents are doing today.

ServiceNow incorporated elements of NemoClaw architecture into AI specialist agents that it released internally last month, according to Amit Zavery, president, chief operating officer and chief product officer. Some customers are already using the agents, which also employ governance and orchestration from ServiceNow’s AI Control Tower product.

A general release is planned around May, according to Zavery. The agents have been achieving 80% to 90% resolution rates. “We will see a lot more of those things,” he told Isabelle and me during a conversation at GTC.

At Nvidia’s on-site Build-a-Claw experience, technologists were building “claws” for everything from road trip planning to job applications, research and content marketing, Isabelle writes. Nvidia executives were standing by to help. Highlights from her story:

• Claws are autonomous agents and can plan and execute tasks on their own, and, critically, spin up their own subagents to tackle specialized tasks, access files and themselves delegate tasks to other subagents.
• They represent a big leap beyond question-and-answer-style AI chatbots as well as recent iterations of AI agents, which typically have narrow use cases and run for a set amount of time—although claws also come with a new set of security concerns. For claws to work as a true personal assistant, they need access to all of a user’s data.
• Daniel Buchmueller, co-founder and chief technology officer of defense logistics startup Gallatin AI, said he was looking at building a claw that could ingest all the daily newsletters he received, and then rewrite the information as a single newsletter, customized to his personal preferences, possibly delivered to his inbox at market close every day.
• This type of asynchronous, always-on capability separates claws from conventional AI chatbots where work happens in real time through back-and-forth exchanges. “Your computer doesn’t need to sleep at night. It can do work for you,” he said about the advantage of claws.

It quickly became clear to Isabelle and me that using OpenClaw isn’t intuitively obvious in the way that ChatGPT was right from the start. The average person will probably have to work harder to get value from them. As one demonstrator told us, it’s more like hiring an assistant who doesn’t know anything about you and building up a good working relationship over time.

Have you experimented with OpenClaw? Let us know what you learned.

Nvidia co-founder and CEO Jensen Huang teased, prompted and prodded reporters during a press conference that went on Tuesday for nearly two hours. In a long-standing ritual, Huang puts the journalist with the last question of the session on the spot in front of hundreds of peers.

Huang: Now, just so you know, the last question is really important…Are you ready? You have got a great question…Let’s hear it everybody. Let’s all be the judge.

Journalist: You say that you want to bless everyone with nothing but suffering…What suffering would you recommend?

Huang: …You know something, there is plenty of suffering to go around. There’s no question about it. It is the essential part of striving to be perfect. It is the essential part of striving to be the best you can be. And so long as you feel that you are striving to be the best you can be, you know, that you have done your best, that you’ve left everything on the field, then the suffering must have happened. All right, thank you everybody. Good job.

— Steven Rosenbush

The AI economy is scaling fast...

Token tracking. As AI adoption accelerates, companies are tracking a new productivity metric: token consumption. While token pricing has gone down, token costs can be higher for some newer, more sought-after models—and companies’ use is generally going up, the Journal’s Katherine Bindley reports.

Companies sometimes opt for pay-as-you-go plans: “It’s a little bit like giving people a fire hose of fuel,” says Vercel CEO Guillermo Rauch, who gives his employees an unlimited token budget.

Others might buy enterprise plans that include a certain amount of use per worker.

The challenge is that high token use can signal either exceptional performance or inefficiency, forcing managers to evaluate AI spending the way they would any other business cost.

Vercel’s Rauch says his highest token spenders are also his top performers. He estimates that $10,000 in tokens spent one day to analyze a research paper and build a new critical-infrastructure service based on it probably saved him millions.

SAP shifting to AI consumption model. SAP CEO Christian Klein tells Bloomberg that he is overhauling the company's business model, shifting from the subscription-based pricing model to AI consumption charges. As part of the pivot, the company will also create "forward deployed engineering" teams to work with customers to develop its AI applications on top of SAP’s technology. SAP hasn't been immune to the stock selloff hitting software companies in the wake of the release of new AI tools from Anthropic and others.

Samsung bets on the AI chips. The South Korean tech giant plans to invest over $70 billion this year to vie for dominance in manufacturing AI chips, the WSJ reports. Solidifying its position in the highly lucrative memory market is a core aim as well, coming as an AI-driven supply crunch sends memory prices surging. See below for more.

AI hunger for memory chips continues. Memory-chip maker Micron Technology’s second-quarter revenue nearly tripled as AI-related demand for memory is outpacing supply. 

Apple on track to surpass $1 billion in AI revenue this year. But it's not from its own AI products, but by acting as a toll road for rivals, the Journal reports. ChatGPT, Grok and other generative AI apps paid Apple nearly $900 million in App Store fees in 2025, with ChatGPT alone accounting for three-quarters of that.

True, Apple's own AI efforts remain a work in progress. But it may not need to win the AI race outright, the Journal explains. While competitors spend on chips and data centers to build frontier AI models, Apple is betting on the personal data stored on its devices and its own custom chips to power an on-device AI strategy. If privacy and personalization prove to be the decisive edge, Apple's approach could look savvy.

When AI agents go rogue. An agent triggered a serious security breach at Meta Platforms last week, the Information reports, autonomously posting advice on an internal forum without user approval. The incident exposed sensitive data to unauthorized employees for nearly two hours. Meta classified it as a Sev 1 incident, its second-highest severity level. No data was misused.

Snowflake launches AI platform. The AI cloud data company on Wednesday launched Project SnowWork, an autonomous AI platform that connects enterprise data with teams across sales, finance and HR to automate business workflows, Axios reports. "We are entering the era of the agentic enterprise, ushering in a fundamentally new way to work,” CEO Sridhar Ramaswamy said, announcing the effort. The product, currently in limited research preview, runs on governed enterprise data within Snowflake's own platform.

Escalating attacks on Persian Gulf oil-and-gas infrastructure are sending the U.S.-Israeli war with Iran into a dangerous new phase that threatens to worsen the crisis over global energy supplies. (WSJ)

Oil topped $115 and European natural-gas prices surged more than 20% as attacks on Middle Eastern energy infrastructure fueled anxiety over supply. (WSJ)

Chinese leader Xi Jinping had hoped next month’s summit with President Trump would cement China’s status as a peer superpower. But the U.S. request to delay it served as a frustrating reminder that Washington still drives the global agenda. (WSJ)

Democratic lawmakers stormed out of a closed-door briefing with Justice Department leaders about the Jeffrey Epstein files, saying they didn’t trust Attorney General Pam Bondi to answer their questions fully unless forced to do so under oath. (WSJ)