AI News Roundup – OpenAI completes corporate restructuring, Qualcomm unveils series of AI chips, new U.S. bill proposes to ban AI companions for minors, and more

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

• • CNBC reports that OpenAI has completed its corporate restructuring, paving the way for an IPO filing in 2026. The AI giant had been considering a restructuring (referenced in this AI Roundup) away from its original nonprofit model and complicated interconnected governance system and into a public benefit corporation (PBC), a structure used by OpenAI’s rival Anthropic. This past week, OpenAI announced that it had completed this process, with its governing nonprofit arm (renamed to the OpenAI Foundation) controlling a 26% stake (worth around $130 billion) in the for-profit OpenAI Group PBC. Microsoft, a major backer of OpenAI since 2019, holds a 27% stake in the company, while OpenAI employees hold 26%, with the remainder held by other investors and stakeholders. Following these moves, Reuters reported that OpenAI is preparing for an initial public offering (IPO) that could value the company at up to $1 trillion, which would allow the company to raise even more funds for its ambitious AI infrastructure projects, especially as OpenAI has yet to turn a profit. The IPO could be filed with U.S. securities regulators as soon as the second half of 2026.
• • U.S. chipmaker Qualcomm has announced a series of AI-focused chips for datacenters, according to Bloomberg. Qualcomm, a leading maker of processors for mobile devices, launched a series of AI-focused chips focused on the AI datacenter market. In contrast to those used by industry leader Nvidia, Qualcomm’s chips are based on the ARM processor architecture, which is used in the vast majority of mobile devices running Android and iOS, and is generally more power-efficient than conventional processors. Qualcomm’s offerings also include a “neural processing unit,” a dedicated chip for AI-related tasks that debuted in smartphone devices and now implemented in laptops and other larger devices as well. The low-power claims of Qualcomm’s new chips may encourage adoption by datacenter developers concerned about the high-power usage of such facilities, though it remains to be seen if the company will be able to challenge Nvidia in the datacenter market.
• • NBC News reports on a new bill introduced in the U.S. Senate that would ban the use of AI “companions” by minors. The bipartisan legislation, introduced by Senator Josh Hawley, a Republican of Missouri, and Senator Richard Blumenthal, a Democrat of Connecticut, would require AI companies to implement age-verification for AI companions and prevent minors from using such products, as well as require the AI companions to disclose that they are AI. The bill also creates criminal penalties for AI companies that create AI companions that elicit sexually explicit or other harmful conduct from minor users. AI companions had been coming under increasing scrutiny in recent months (as reported by this AI Roundup in September), following lawsuits by grieving parents against AI companion companies alleging AI behavior that contributed to their teen children’s suicides as well as investigations from the Federal Trade Commission. This bill also follows a few weeks after California approved the nation’s first regulations on AI companions, requiring similar disclosures to the federal bill. The text of the Hawley-Blumenthal bill, dubbed the Guidelines for User Age-verification and Responsible Dialogue, or “GUARD” Act, is available here as a PDF.
• • Universal Music has settled its lawsuit against AI music company Udio, according to the Wall Street Journal. The lawsuit, brought by Universal as well as major record labels Warner and Sony in 2024 (and previously analyzed by MBHB Partner Aaron Gin and other MBHB practitioners), alleged copyright infringement through Udio’s use of copyrighted songs for AI model training. Under the settlement, Udio will launch a subscription service that will allow users to generate music based on songs licensed from Universal, though artists must consent to their music being included, and generated songs cannot be exported from the Udio platform. Artists who allow their music to be used by Udio will be paid, according to a Universal executive. Copyright concerns with AI training have been perhaps the most litigated aspect of AI over the past few years, but licensing deals such as this are likely to become more common. Regardless, the lawsuit by Warner and Sony against Udio continues, and no agreement has been reached by any music label involved in the suit with Suno, another AI music company, who is also a defendant in the case.
• • The MIT Technology Review reports on a new optical character recognition (OCR) AI model developed by China’s DeepSeek. OCR has been a major application of machine learning technology for several decades, and is still considered one of the most difficult tasks for computers to perform accurately. DeepSeek’s new model, while relatively conventional, could use the OCR approach to improve large language models (LLMs). LLMs generally store text as small units (“tokens”); however, as models become more complex with vast quantities of tokens, it becomes more difficult for the model to retain information, especially during long exchanges with users. In contrast, DeepSeek sought to store text as images, which according to a company whitepaper allows the model to retain a similar amount of information using far fewer tokens. The unconventional approach may aid the efficiency of future AI models.