Good morning. I had a fascinating conversation with Caltech computer scientist and mathematician Babak Hassibi, who delved into how his team created a large language model that achieves radical compression without compromising performance.

It reflects the still-accelerating pace of innovation in AI and points to one of the technology's likely outcomes: a new generation of small models that require less energy consumption, but still deliver the performance required for demanding applications on phones, laptops, industrial devices or robots. 

Hassibi’s team has launched PrismML, a company that came out of stealth Tuesday and open-sourced its 1-bit technology model, enabling others to use it.

PrismML has developed an extreme form of compression that allows AI to run locally on phones, laptops and other devices, and enables data-center build-outs that can do more with fewer resources and avoid ballooning energy costs, according to Hassibi.

Read the story here.

“We spent years developing the mathematical theory required to compress a neural network without losing its reasoning capabilities,” said Hassibi, chief executive of the venture. “We are creating a new paradigm for AI: one that adapts to diverse hardware environments and delivers maximum intelligence per unit of compute and energy.”

The company raised $16.25 million in a SAFE and seed round with investors Khosla Ventures, Cerberus Capital Management and Caltech.

AI’s future won’t be defined by who can build the largest data centers, but by who can deliver the most intelligence per unit of energy and cost, according to investor Vinod Khosla. “It’s a mathematical breakthrough, not just another tiny model,” Khosla said. “You can fit a much better model on a phone. That’s a big deal. Of course on your phone or a mobile device, energy consumption is a very, very big deal.”

The same efficiency gains that enable local deployment also allow data centers to operate more effectively, PrismML said.

The mathematics are proprietary, but Hassibi said the effect was much like compressing a digital photograph without losing visual fidelity.

By reducing the units of data, or model weights, to a single bit represented by +1 or -1, PrismML’s flagship 1-bit Bonsai 8B model can boost processing speeds by as much as eight times compared with a 16-bit model, Hassibi said. It can also achieve reductions in energy consumption of up to 75% to 80% on current hardware platforms.

Amir Salek, senior managing director at Cerberus Capital Management and a veteran of Google and Nvidia, said he was convinced PrismML achieved a major mathematical breakthrough with the potential to improve the economics of AI.

Does your company have a need for high-performing models that fit on phones and other small devices? Let us know.

Oracle told some employees in the U.S. and India on Tuesday that their jobs were being eliminated, WSJ reports. The cuts come as the company continues to build out costly AI data centers.

The full scope isn't yet clear, but some employees said internal metrics put reductions so far in the thousands. TD Cowen analysts earlier this year predicted Oracle would shed as many as 30,000 workers. Oracle had a global workforce of about 162,000 as of the end of May.

OpenAI closes Silicon Valley’s largest-ever funding round. The company raised $122 billion ahead of a blockbuster IPO expected by the end of the year, WSJ reports. The deal, which values OpenAI at $852 billion, came with an additional perk: greater access to individual investors. More from OpenAI's blog post on the news:

• "Momentum is just as strong on the enterprise side, which now makes up more than 40% of our revenue, and is on track to reach parity with consumer by the end of 2026."
• "With each new generation of infrastructure, we train more capable models, making each token more intelligent than before. At the same time, algorithmic and hardware improvements reduce the cost to serve each token, lowering the cost per unit of intelligence. That added intelligence makes AI useful for more complex workflows, which increases usage, drives compute demand, and accelerates the next turn of the flywheel."

Anthropic leaks part of Claude Code’s internal source code. “No sensitive customer data or credentials were involved or exposed,” an Anthropic spokesperson said, CNBC reported. “This was a release packaging issue caused by human error, not a security breach. We’re rolling out measures to prevent this from happening again.”

• "The leak also marks Anthropic’s second major data blunder in under a week. Descriptions of Anthropic’s upcoming AI model and other documents were recently discovered in a publicly accessible data cache," CNBC reported, citing a report from Fortune on Thursday.

Iran says it will target U.S. tech companies in Middle East. Iran’s Islamic Revolutionary Guard Corps (IRGC) said Tuesday that it plans to target major U.S. technology companies across the Middle East, including Apple, Microsoft and Google, on Wednesday amid the ongoing war, The Hill reports.

Former BP CEO goes digital. Data-center developer Prometheus Hyperscale has taken that tech conference chestnut, “data is the new oil,” to the next level, naming former BP chief Bernard Looney as its next chief executive, WSJ reports. Founded in 2020, Prometheus aims to build sustainably powered data centers with liquid-cooling technology that limits water use.

Microsoft plans to invest $5.5 billion in Singapore by 2029. Microsoft is on track to invest $5.5 billion in cloud and artificial-intelligence infrastructure in Singapore through 2029, as demand for AI skills and computing continues to grow in the city-state, WSJ reports. The U.S. tech giant’s vice chair and president, Brad Smith, said Wednesday that the investment will also go toward ongoing operations. The announcement comes a day after Microsoft said it plans to invest over $1 billion in Thailand.

The United Arab Emirates is preparing to help the U.S. and other allies open the Strait of Hormuz by force, Arab officials said, a move that would make it the first Persian Gulf country to become a combatant, after being hit by Iranian attacks. (WSJ)

Large corporate deals had their best quarterly showing ever, as companies forged ahead with tie-ups and investments despite the Iran war rattling markets. (WSJ)

The Supreme Court on Tuesday sided with a Christian counselor who challenged a Colorado ban on mental-health counseling that seeks to change young people’s sexual orientation or gender identity. (WSJ)

A federal judge in Washington, D.C., blocked construction of President Trump’s White House ballroom. Judge Richard Leon granted the National Trust for Historic Preservation’s request to temporarily stop work during its legal fight over the 90,000-square-foot project. (WSJ)

The WSJ Tech Council brings together CIOs, CTOs and CISOs advancing innovation and shaping the future. Join this trusted community where tech executives connect with peers to explore emerging trends and gain the perspective they need to stay ahead of disruption.

Request Information

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.