This week, I’m focusing on the increasing pressure on the AI industry’s wunderkind, OpenAI. I also look at the change in AI leadership at Apple, and at the music industry’s new cooperation with AI music generation apps. 

Sign up to receive this newsletter every week via email here. And if you have comments on this issue and/or ideas for future ones, drop me a line at su******@*********ny.com, and follow me on X (formerly Twitter) @thesullivan. 

Is OpenAI still the king?

The AI industry has always been very competitive, and it’s getting even more so. A relatively small group of AI labs are slugging it out to release the smartest models, and, by extension, the smartest chatbots. Ever since OpenAI released its ChatGPT chatbot three years ago, the upstart company has been seen as the leader, but that status has been called into serious question by Google’s new Gemini 3 Pro model (and the Gemini app).

ChatGPT has grown quickly. The official number is 800 million weekly active users. Google’s number is 650 million monthly active users for the Gemini chatbot. So, apples and oranges. SimilarWeb provides a somewhat better comparison, saying that Gemini’s share of web traffic grew from 5.7% a year ago to more than 15% today. Meanwhile, ChatGPT’s 87% share a year ago shrunk to 71.3% today.

OpenAI is feeling the pressure from Gemini (and probably from Anthropic’s new Claude Opus 4.5 model). CEO Sam Altman sent a memo to staff Monday declaring a “code red” effort to improve ChatGPT, according to The Information and other outlets. The effort includes reducing investments in enhancing the health information available on ChatGPT, as well as reducing work on the shopping experience, and the advertising that could go around that. “Our focus now is to keep making ChatGPT more capable, continue growing, and expand access around the world—while making it feel even more intuitive and personal,” ChatGPT product lead Nick Turley tweeted Monday.

In a wider sense, OpenAI is losing billions, and spending billions, a fact that must make its investors both nervous and curious. Leaked documents and analyst estimates show OpenAI will lose between $9 billion and $11 billion in 2025 (spending roughly $22 billion while bringing in about $13 billion in revenue). The company recently told investors that its spending through 2029 could rise to $115 billion. Altman has said his company, partners, and investors will commit as much as $1.4 trillion to infrastructure (chips, data centers, etc.) in the next eight years.

OpenAI is an aggregator, as the analyst Ben Thompson points out. The fact that it’s willing to de-emphasize its shopping and advertising experiences, which are potential revenue generators, shows that it’s still in the mode of growing users, and not yet in the mode of growing revenue. And the way that aggregators (like Facebook) grow is by becoming more things to more people in order to maximize attention and engagement on its platform, regardless of whether the users are paid subscribers. In the aggregator model, actually monetizing all those eyeballs comes later.

The confidence in that model, which requires constant growth toward a critical mass of users, has afforded OpenAI a certain swagger, and even a cavalier attitude about making returns for its investors. One of those investors, Altimeter Capital’s Brad Gerstner, asked Sam Altman during an October podcast (12:30 mark) how he explains to the markets spending more than a trillion on infrastructure when his company is operating deep in the red. Altman was exasperated. “Brad, if you want to sell your shares, I’ll find you a buyer,” he said. “I just . . . enough.”

But it’s no longer clear that OpenAI has the best models and the go-to chatbot. Setting aside the shopping and advertising work, OpenAI is right to reassign its talent to work on new models and new skills for ChatGPT. This also might mean taking talent off “fun” projects like the Sora app, which seems far afield from the mission of making ChatGPT the highest performing chatbot available.

On the other hand, things can change very quickly in the AI world. Reports say OpenAI is already set to release a new reasoning model codenamed “Garlic” that will overtake Gemini 3 on a number of key benchmarks. We’ll see if Garlic gets a better reception than GPT-5.

Apple must keep publishing AI research under Subramanya

This week Apple announced that its AI boss, John Giannandrea, will be leaving the company. Giannandrea had been a successful AI leader at Google, but his name is linked to Apple’s failure to seize on generative AI to improve its Siri voice assistant and make the iPhone and other iDevices smarter and more personalized. He’ll be handing the reins to another Google vet, Amar Subramanya, who once led engineering on Google’s Gemini chatbot, and is stepping down after seven years on the job. Apple’s stock price got a slight boost on the news, as some investors saw Apple signaling a new urgency to bring AI to its devices. Subramanya’s remit will be restoring Apple to some kind of parity with its peers in developing AI models and applying them in meaningful ways.

As Mark Zuckerberg can attest, achieving that goal will depend on recruiting and retaining top-shelf AI researchers. Giannandrea’s AI/ML group saw a lot of churn and lost a number of top shelf researchers to Meta and others, including Ruoming Pang and Robby Walker. One reason for this was the group’s habit of investing time and labor in technical approaches to problems only to see them scrapped. Another was the slow pace of developing and releasing new AI features for products like Siri.

Another problem is publishing. Apple is famously secretive about its R&D in all areas of the company. The company likes to talk about customer-facing products, and dislikes talking publicly about the technology that makes them work. AI researchers aren’t OK with that. They want to publish their research. They want the exposure and influence that can bring within an ultra-competitive industry.

When Giannandrea came to Apple, the company began allowing its AI talent to publish more of their research—to the extent they could do so without revealing trade secrets. Apple now has a “Apple Machine Learning Research” web page that lists published papers, technical reports, and conference submissions. It will be crucial that Subramanya keeps this practice going, or expands it. Otherwise Apple risks losing key researchers to competitors.

Record Labels are having their iTunes moment with AI

The Music Industry has stopped suingAI music generation apps—instead, it’s making deals with them: The three major record labels have now signed licensing agreements with AI music startups.

Warner Music Group, Universal Music Group, and Sony Music Entertainment have made licensing deals with an AI music startup called Klay Vision. The agreements grant Klay Vision permission to train its music generation models on music catalogs owned by the labels, replacing previous models that relied on scraped or unauthorized data.

AI-generated music is getting more popular. An AI-generated song using a simulation of a real human country singer’s voice recently hit number one on the Billboard Country Digital Song Sales ranking.

Suno, another AI music company that previously faced lawsuits from major labels, has signed what it calls a “first-of-its-kind partnership” with Warner Music. The deal moves the company toward licensed, artist-opt-in AI models.

The moment feels similar to the record labels’ decision in the early 2000s to sell digital music on Apple’s iTunes platform. The labels saw CD sales tank as consumers downloaded free MP3s from sites like Napster and Limewire.