What's up: OpenAI completes for-profit transition; jobs are disappearing as AI starts to bite; Here come the housecleaning robots.

Good morning. Beyond the technology announcements, the business partnerships, and the discussion of American reindustrialization, Nvidia’s first tech conference in Washington captured something about AI. Businesses are starting to figure out what to do with it.

Corporate adoption of AI technology has faced challenges, but demand is starting to pick up, according to Justin Boitano, Nvidia vice president, enterprise AI. “There were a lot of challenges in enterprise. Enterprises need to rethink their data centers for AI and it just took them a bit longer because it is this whole stack rearchitecting the data center,” Boitano told me. “We’re seeing demand actually start to take off…in the last couple quarters.”

That is consistent with the findings of a new report. Over the last year or so, research has found that most companies have struggled to drive sufficient value from AI. But one new study from Wharton Human-AI Research and GBK Collective suggests that is changing.

Three in four enterprise leaders report positive returns on their GenAI investments, with 72% of leaders saying their organizations now track metrics for GenAI tied to profitability, throughput or productivity, according to GBK partner Jeremy Korst. “Our third annual study shows a clear inflection point. GenAI is delivering measurable returns,” Korst said. Researchers said the study included 672 senior leaders from U.S. enterprises, each with annual sales surpassing $50 million.

Why now? AI infrastructure and models continue to improve, but so do the organizational and leadership frameworks that companies bring to the application of AI. To the extent that companies are beginning to get more value from AI, their evolving mindset is key.

Enterprises, which can include companies and other large organizations, have passed through several stages in the way they think about deriving value from AI. At first, enterprises thought they could take a generative AI model or chatbot and simply drive business value from it. In the second wave of adoption, they injected their own data for context, employing techniques such as retrieval-augmented generation. Now, with AI agents, “people are getting a lot higher accuracy, and as a result we see that driving adoption,” Boitano said.

Is your company’s value from AI accelerating or is it stuck in low gear? Let us know.

Earnings season continues with results due after the bell from companies including Alphabet, Meta and Microsoft.

A leaner new normal for employment in the U.S. is emerging. Large employers are retrenching, making deep cuts to white-collar positions and leaving fewer opportunities for experienced and new workers, WSJ reports. Nearly two million people in the U.S. have been without a job for 27 weeks or more, according to recent federal data.

Behind the wave of white-collar layoffs, in part, is the embrace by companies of artificial intelligence, which executives hope can handle more of the work that well-compensated white-collar workers have been doing

Recent cuts

Amazon.com said this week that it would cut 14,000 corporate jobs.
UPS said it reduced its management workforce by about 14,000 positions.
Target said it would cut 1,800 corporate roles.

When Silicon Valley goes to Washington. As par for the course these days for such events in Washington, D.C., Tuesday's Nvidia event included praise for President Trump's policies, WSJ reports.

“We are manufacturing in America again. It is incredible,” said Nvidia Chief Executive Jensen Huang. “The first thing that President Trump asked me for is, bring American manufacturing back…Nine months later, we are now manufacturing, in full production, Blackwell in Arizona.”

Tuesday’s speech marked the latest instance of Huang praising Trump while the president contemplates export policies worth billions of dollars to Nvidia.

AI supercomputing meets quantum. The view among researchers today is that for quantum to deliver, it needs to be integrated with high-functioning classical computers, like AI supercomputers, that perform calculations they can’t and fix the naturally occurring errors in their answers.

On Tuesday Nvidia unveiled NVQLink, an interconnect that links quantum processors to GPUs. “NVQLink is the Rosetta Stone connecting quantum and classical supercomputers,” said Huang.

Eli Lilley hopes Nvidia-chip powered supercomputer will help it find new drugs and accelerate lengthy R&D timelines. The companies on Tuesday said they would build a supercomputer powered by more than 1,000 Nvidia Blackwell GPUs to boost the discovery of new medicines, WSJ reports.

Nvidia will have a 2.9% ownership stake in Nokia. Nvidia will invest $1 billion in the Finnish telecom company as part of an agreement between the companies to work together on AI networking solutions, WSJ reports. Potential projects could include incorporating Nokia’s data center switching and optical technologies in Nvidia’s future AI infrastructure architecture.

OpenAI has successfully converted to a more traditional corporate structure, turning its for-profit subsidiary into a public-benefit corporation, in a move that clears an obstacle for a potential IPO, WSJ reports.

As part of the new structure, Microsoft, an early partner and among its largest investors, will own 27% and have exclusive intellectual-property rights to OpenAI technology until 2032. 

Behind the scenes. OpenAI’s conversion to a more traditional company structure came over the opposition from some of California’s largest labor unions, nonprofits and corporate rivals. The Journal reports how OpenAI CEO Sam Altman hacked a path, telling California Attorney General Rob Bonta in a meeting that the company would be willing to leave the state if Bonta blocked its plan.

A San Francisco startup looking to take on the Dutch behemoth ASML has raised more than $100 million, WSJ reports. Substrate said its core innovation, a compact machine that uses ultrashort-wavelength laser light to etch intricate and microscopic patterns on silicon wafers, will help it meet its three-year timeframe for producing its first chips. The company has hired more than 50 employees from IBM, TSMC, Google, Applied Materials and national laboratories.

A long-running joke in Silicon Valley is that most startups are focused on getting tech to do what their mommies used to do. The joke's on us. WSJ Tech Columnist Joanna Stern tries out Neo, a 5-foot-6-inch humanoid robot that can fold clothes, take dishes out of the dishwater and grab items from the fridge.

Amazon’s supercomputer for Anthropic gets nearly 500,000 custom chips

On Wednesday, Amazon Web Services announced that the first phase of the supercomputer it built for Anthropic, called Project Rainier, was completed in 10 months and holds nearly 500,000 of Amazon’s homegrown Trainium chips.

The so-called Ultracluster of chips is one of the largest in the world, and powers Anthropic’s flagship Claude model. It is being used for training Claude as well as inference, the process of sending queries and receiving answers back from the model, according to Dave Brown, Amazon Web Services vice president of compute and machine learning.

“Anthropic has not been waiting right from early January—this cluster has been in use for training and for inference of their workloads,” Brown told the WSJ Leadership Institute

Project Rainer is housed within data centers in Indiana and Mississippi, as well as other locations across the U.S. where power is available, AWS said. They are linked by Amazon’s proprietary networking technology, which allows the cluster to be physically spread out, Brown added.

By year’s end, the cluster will house over one million Trainium chips, according to AWS.

In addition to using Amazon’s Trainium chips, Anthropic also uses Google’s homegrown chips, called TPUs, as well as Nvidia’s advanced AI chips. Earlier last week, the AI startup said it would expand its Google Cloud partnership to access one million TPUs.

“I think anybody running the business wants to look and see what’s available in the market,” Brown said, referring to Anthropic’s deal with Google.

— Belle Lin

CORRECTION: Fireworks AI raised $250 million in its latest funding round. Tuesday's newsletter incorrectly said it raised $254 million. 

Hurricane Melissa made landfall in Cuba early Wednesday, a day after hitting Jamaica as one of the most powerful Atlantic storms on record. (WSJ)

Sticker shock over higher costs for Affordable Care Act plans is hitting many enrollees around the country who are now learning about hefty increases in what they might pay next year, putting the healthcare subsidies at the heart of the government shutdown in the spotlight. (WSJ)

President Trump said he expects to sign a trade deal with China when he meets Chinese President Xi Jinping on Thursday, potentially paving the way for a de-escalation in tensions between the two sides. (WSJ)

The White House fired all members of the Commission of Fine Arts, an independent federal agency that would be tasked with reviewing some of President Trump’s construction projects, a White House official confirmed. (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.