Do shake-ups at 2 high-profile AI companies signal more to come? 

And so, after shake-ups at two high-profile AI companies last week, we have a trend. The question is whether the shake-ups are isolated to the two businesses in question or a signal of more upheaval to come in the AI industry.

The fireworks started last week with a major turmoil at Inflection AI, which had built a chatbot assistant that’s supposed to juggle the everyday chores of its users like scheduling appointments while also giving them emotional support. The company, led by former Google DeepMind’s cofounder Mustafa Suleyman, had received a super-sized $1.5 billion in funding and was said to be valued at $4 billion. 

But on Tuesday, everything changed. Suleyman and his top lieutenants suddenly decamped to Microsoft and Inflection pivoted to focusing on corporate users—i.e. customers who are more likely to pay big bucks.

The abrupt reshuffling doesn’t exactly scream success. 

Then, late Friday, it was Stability AI’s turn to go into convulsions. CEO Emad Mostaque, who founded the company best known for its text-to-image generator Stable Diffusion, announced his exit after battling with investors, grappling with a staff exodus, and once claiming to have worked as a secret agent for the U.K. government. 

Even before Mostaque’s departure, Stability AI had faced growing questions about its direction. Now, with its public face gone, the skepticism about its business, valued at $1 billion in 2022, has reached a new level. 

So where does that leave the AI industry? You could argue that the changes at the two companies are isolated incidents that speak only to their own internal dynamics. But my take is that the industry is undergoing a bit of a reality check after living off of easy money since at least the end of 2022 when the debut of OpenAI’s ChatGPT opened the venture capital floodgates.

It’s no secret that most AI companies, despite all their buzz and sky-high valuations, actually have minuscule revenue while burning wads of cash to train their models. One recent estimate is that it costs $1 billion of computing to train one large language model. 

In every tech boom, investors inevitably start asking questions about big spending by their portfolio companies and when significant revenue will start pouring in. For at least some AI startups, that time has come. 

Another truism is that the world doesn’t need dozens of AI chatbots and image generators. Perhaps now we’re finally seeing that obvious reckoning. Of course, OpenAI and its Big Tech rivals have the deep pockets to fund their ambitions and can offset their money-losing AI products by bundling them with their non-AI products that make money. Smaller fry, with their more limited resources, will have a tough time competing unless they’re focused on a niche that the whales ignore. 

Don’t get me wrong. The venture capital floodgates are still wide open for AI and there doesn’t appear to be an imminent mass culling of AI startups. And the recent pop of AI-adjacent IPOs Reddit and Altera Labs shows that Wall Street has plenty of appetite for artificial intelligence stocks. But in some cases, pressure is growing on companies to pivot and replace leadership after their struggles become impossible to ignore. 

After all, two is a trend—even when it comes to AI, an industry that has so far defied most of the usual business realities. 

With that, here’s more AI news. 

Verne Kopytoff
verne.kopytoff@fortune.com

AI IN THE NEWS

Tech firms try to loosen Nvidia’s grip. Companies including Google, Intel, and Qualcomm are planning to make the AI sector less dependent on Nvidia by building an open-source suite of software and tools that would serve as an alternative to Nvidia’s CUDA software platform, which most developers use to build AI applications. Building off Intel’s OneAPI tool, the new UXL Foundation will “promote productivity and choice in hardware,” Google high-performance computing chief Bill Magro told Reuters. The consortium’s members are promising nailed-down tech specs in the first half of this year, which would suggest an urgent effort.

AI builds AI ETF portfolio. Canadian asset manager Evolve Funds Group has launched an ETF with an AI portfolio that’s partly picked by AI. According to Reuters, the Evolve Artificial Intelligence Fund uses a model from consultants Boosted.ai, which seeks out smaller companies that it expects to benefit from the AI boom—though the fund will also buy stocks of megacaps like Nvidia and Microsoft.

Global AI resolution. The United Nations General Assembly unanimously adopted a resolution on Thursday that calls for personal-data protection and the upholding of human rights as AI becomes more widespread—as well as the monitoring of AI for emerging risks. As Ars Technica reports, the AI industry seems happy with this resolution, which lacks any teeth. It’s not the first international agreement on AI, given the existence of the (also quite vague) Bletchley Agreement from late last year, but it is the first truly global agreement on the subject.

AI energy debate. AI’s soaring energy usage isn’t going to be leaving the news agenda anytime soon. John Pettigrew, the CEO of the U.K.’s National Grid, today told a conference that the AI boom would place pressure on the power network, with data centers’ energy demands increasing sixfold over the next decade. But also today, the Financial Times reported about energy companies’ own use of AI to improve network efficiency and potentially reduce their carbon footprint.

Creative AI pushbacks. The actors guild SAG-AFTRA won heavy protections for human voice actors in animated series, with a studio agreement that says only people can qualify as voice actors. This goes beyond the contracts agreed for live-action actors following last year’s monthslong Hollywood strike. Per Variety, the cartoon contracts allow the use of AI, but actors can withhold permission for the reproduction of their voices. Meanwhile, BBC viewers complained about the use of generative AI in promotional materials for the upcoming season of Doctor Who, leading the British broadcaster to promise not to do it again, Deadline reports.

FORTUNE ON AI

In 40 years as a founder-CEO, Michael Dell turned his dorm-room PC company into a tech giant. Can he cash in on the AI boom?, by Michal Lev-Ram 

How AI can boost sustainable investing, by John Kell 

The U.S. could need to equivalent of 40 new nuclear plants in the next 5 years—blame power hogs AI and cannabis, by Irina Ivanova

Country music heartland Tennessee becomes the first state to protect musicians from AI, by the Associated Press

Italy's AI watchdog hasn't been able to hire for any of its 4 open positions: 'The search process went worse than our low expectations', by Ryan Hogg

AI CALENDAR

April 15-16: Fortune Brainstorm AI London (register here)

May 7-11: International Conference on Learning Representations (ICLR) in Vienna

May 21-23: Microsoft Build in Seattle

June 5: FedScoop’s FedTalks 2024 in Washington, D.C.

June 25-27: 2024 IEEE Conference on Artificial Intelligence in Singapore

July 15-17: Fortune Brainstorm Tech in Park City, Utah (register here)

Aug. 12-14: Ai4 2024 in Las Vegas

EYE ON AI RESEARCH

Detecting attempts to wake virtual assistants. Apple researchers published an as-yet non-peer reviewed paper Friday, explaining their new technique for using AI to bypass the need for a trigger phrase when activating a virtual assistant. These trigger phrases serve an important purpose—without them, devices would need to be constantly listening for instructions rather than a particular “wake word” (“Siri” for Apple’s assistant, “Alexa” for Amazon’s) which would have implications for privacy as well as for unintentionally given commands. But Apple’s researchers have come up with a multimodal detection system, built on a pretrained version of OpenAI’s lightweight GPT-2, that they say can “distinguish device-directed utterances from background speech in the context of interactions with a virtual assistant.”

As MIT Technology Review notes, three of the paper’s authors work on Apple’s Siri team. However, there could be serious downsides if Apple does choose to ditch the “Siri” invocation—which had to also include a “hey” until last year—given its previous legal troubles over allegations that Siri continues to listen even when not activated.

EYE ON AI NUMBERS

47%

That's the share of the world's top AI researchers who got their undergraduate degrees in China as of 2022, according to a study by MarcoPolo, a research organization operated by the Paulson Institute, which promotes closer ties between China and the U.S. The U.S., in contrast, produced just 18%.