The Information AM

Save 25% on an annual subscription to read the most important news about technology and business first. For even more, Save $250 on The Information Pro for unlimited access to our proprietary org charts, databases and surveys.

Microsoft on Wednesday unveiled its latest effort to run AI in PCs powered by its Windows software, rather than running the AI in the cloud, which the company said would bring down costs for customers.

The announcement is Microsoft’s response to Apple’s Mac Mini, which has become a hit with AI customers and app developers because of its ability to run models directly on the device. Microsoft is also trying to show enterprise customers that AI can run securely on company-owned computers.

Microsoft said its forthcoming Surface computers powered by Nvidia chips, which the company previewed earlier this year, will start at $2,599, while a more powerful PC, Surface RTX Spark Dev Box, will start at $5,999. An even more powerful Nvidia-powered PC, DGX Station, will be capable of running large AI models (with as many as 1 trillion parameters, or settings tied to the model) on the device, such as Meta’s Llama 4 Maverick or DeepSeek’s V4-Flash. It will also be capable of streaming complex videogames like Call of Duty from the cloud starting next year. Microsoft didn’t disclose pricing or say when the device would be available.

Microsoft also said its Copilot AI software will use a model router to automatically run some AI tasks to run locally on the Nvidia-powered PCs, using Microsoft’s relatively cheap MAI-Code-1.1 model, for instance. And Windows customers will be able to direct Copilot to generate custom applications, such as desktop widgets or webpage prototypes, that the PC can host locally on their device, further driving down costs, the company said. Finally, Microsoft said it was adding new governance controls to Windows to let companies control how employees’ own AI agents, such as those powered by OpenClaw or Perplexity, can access certain files. The features will become available in Windows 11 later this month.

Elon Musk announced on Tuesday night that SpaceX’s AI unit will now use some AI models from competitors to power Grok Bot, signaling a shift away from relying exclusively on models developed in-house.

“Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is most likely to give you the best outcome,” Musk wrote on X.

The approach means the Grok Bot interface could act as a front-end that routes user requests to different AI systems depending on which model is best suited to a particular task. For example, SpaceX could use Claude for reasoning tasks and Midjourney for image generation. Grok Bot is designed as an always on AI agent that can perform tasks autonomously across websites, email, and other applications.

The move could have larger implications for SpaceX’s AI division, which has been trying to catch up with Anthropic and OpenAI and has lost many of its AI research staff over the last year. The company acquired the coding startup Cursor earlier this year to help bolster its coding efforts. SpaceX has increasingly begun selling its compute to rival AI labs, including Anthropic and Google since the spring.

Anjney Midha, a former general partner at Andreessen Horowitz, and former executives at Google, Apple and Nvidia have launched a company that aims to make it easier and more affordable for smaller companies or startups to access such compute. The move comes as Microsoft and other cloud providers have diverted server stockpiles to their internal teams or bigger customers.

To do so, the new company, National Compute, seeks to pool idle AI server capacity at cloud providers in the U.S. and U.S.-friendly countries. Companies that rent AI servers from cloud providers, but which may not utilize them 100% of the time, have committed to providing 750 megawatts of capacity to the pool of servers National Compute plans to sell to its own customers. (That amount of servers, such as Nvidia graphics processing units, would typically cost tens of billions of dollars a year to rent.)

Midha said National Compute already has more than $5 billion in reservations, or demand from potential customers.

The new effort spotlights a problem in the market for GPUs that is hitting smaller customers particularly hard. Midha said that sometimes, large customers that reserve dedicated GPU capacity at cloud providers might use only a small portion of it. National Compute would try to reduce such waste by finding renters for the unused capacity and could give both the large customer and the cloud provider a cut of the new rental revenue.

National Compute’s co-founders haven’t finalized its corporate structure, including who will own how much equity, nor have they settled on a specific business model. They’re seeking feedback on the plan and this week will submit an academic paper outlining National Compute’s strategy. AMP, another company Midha launched this year to make AI servers available to more companies, is putting together a group of companies to manage National Compute as a consortium, he said. AMP could also own a stake in the new company. (He declined to share the names of National Compute’s co-founders.)

Mike Kubzansky, the former CEO of philanthropic investment firm Omidyar Network, is launching a group for large investors that will facilitate information sharing about the AI industry to foster more trustworthy practices, he told The Information.

The investor network, called Andaris, is backed by about $25 million in initial grant funding. While existing networks such as the Council of Institutional Investors and the National Association of State Treasurers already serve some of this purpose, Kubzansky said Andaris is the first U.S.-based network focused exclusively on AI.

Andaris’ thesis is that sharing best practices about trustworthy AI among its members will help reduce risks that investors face in the AI boom, such as a bubble bursting or disruptions caused by AI labor displacement. “We would like to help prove the case that trustworthy AI will be more profitable AI in the long run because no one tends to buy technology they don’t trust,” Kubzansky said. For example, AI companies’ recent interest in embedding outside safety evaluators could help build trust in the technology, he said.

The members of the network are so-called “universal owners” such as pension funds, endowments and family offices whose portfolios include exposure to the whole U.S. economy. Early members include Roy Bahat, head of Bloomberg Beta, and Ian Fuller, co-founder and CEO of Westfuller Advisors. Attending members at its first convening at the end of the month will represent about $1.75 trillion in assets under management.

Andaris will help its members share information on topics such as round-trip financing in the data center buildout and corporate governance questions related to AI cyber attacks.

The goal is to help members steer the AI industry by asking better questions to their portfolio companies, selecting better fund managers and better directing their investments, according to Kubzansky. “Investors have a lot of tools at their disposal without waiting for policy,” he said. “They don’t have to wait for Congress to get its act together,” but “there is undoubtedly going to be a need for policy as well.”

Amazon laid off hundreds of people in its retail division today, a spokesperson confirmed.

“We’ve adjusted parts of our Stores business because we believe this structure will better enable us to deliver on our priorities,” a spokesperson said, adding Amazon is always looking at its team structures as it looks for ways to innovate for customers. “As part of these changes, we’ve made the difficult decision to eliminate a small number of roles.

The layoffs follow mass layoffs in the past year – Amazon laid off 14,000 people in October and another 16,000 in January. At the time Beth Galetti, senior vice president of people experience and technology, said it was not the company’s plan to make broad reductions every few months.

Still, the company has made some targeted cuts this year. Over the summer, Amazon cut staff working on its homegrown AI models, called Nova.

Sriram Krishnan, who advised President Donald Trump on AI policy before leaving the White House in June, is raising a venture fund that is targeting about $500 million to invest in growth- and later-stage AI companies, The Information reported.

Krishnan would be the sole general partner. The firm plans to focus on U.S.-based AI companies in areas such as national and cyber security, and to seek companies with opportunities for public-private collaboration, modeled on industries like aerospace, where firms often partner with agencies such as NASA. It couldn’t be learned which companies the firm is considering. The fund is in a formative stage.

One person said it plans to start making investments by the end of this year, and two said prospective investors include endowments and other institutional investors. At the White House he worked under AI and crypto czar David Sacks and helped lead the administration’s AI action plan, released in July 2025. Earlier he worked at Microsoft and Twitter, then joined Andreessen Horowitz in 2021, where he helped lead the firm’s $400 million investment in Twitter as part of Elon Musk’s $44 billion buyout and opened its London office in 2023.

The U.S. Treasury Department said on Wednesday it had fined the parent company of startup accelerator Plug and Play Tech Center, the first penalty issued under an outbound investment restriction program targeting China’s tech sector.

California real estate company Amidi Group was fined $200,000 for failing to notify Treasury of a Chinese subsidiary’s investment into Noematrix, a Shanghai-based robotics startup. The Outbound Investment Security Program, which took effect in January 2025, forbids Americans from investing in certain tech sectors in China and requires notification of investments in others.

In April, a Chinese fund controlled by Amidi invested $92,478 in Noematrix. The Treasury didn’t explain the basis for the calculations of the fine, which more than doubled the value of the investment. The department said the fine was issued in July, and didn’t require the investment to be unwound.

The parent company of Manus, the AI agent startup that recently separated from Meta Platforms, said Thursday that it has raised more than $500 million in a new funding round.

Butterfly Effect said in a social media post on WeChat that the funding round was led by new investors Boyu Capital and IDG Capital. Its existing investors Tencent, HSG and ZhenFund also participated in the round.

The fundraising comes after the startup completed the unwinding of a $2 billion sale to Meta. Manus, which was founded in China but later relocated to Singapore, was ordered by the Chinese government in April to undo Meta’s acquisition of the startup, which was announced in December. To reverse the deal, the startup’s original investors, Tencent, HSG and Zhenfund, bought its shares back from Meta. In June, The Information reported that Manus’s annualized revenue run rate had risen to between $400 million and $500 million, up from $100 million in December.

Manus faces growing competition from U.S. giants. Meta’s Muse personal AI agent app, launched a month ago, has become an instant hit, and OpenAI recently released its own personal agent called Dots. To expand its user base, Manus recently upgraded its main Manus app and also launched a new standalone app called Cue, which provides each user with multiple personal agents.