The arrival of on-device AI could radically reshape the app/app store model that has ruled the tech industry in the smartphone era.

Why it matters: Apple and Google have dominated the mobile era by controlling the two major smartphone ecosystems and their associated app stores.

Driving the news: Qualcomm announced it is bringing the next generation of its Oryon chip to the smartphone — basically, putting the full power of a PC processor inside a phone.

• That, combined with improved companion processors, provides a lot of horsepower as Qualcomm looks to enable phone users to run AI models that previously had to operate via a data center.

At the same time, AI companies like Meta, OpenAI, Mistral and others have shown they can improve performance while decreasing the size of their models.

• Today, for example, models with only a few billion parameters can outperform the 175 billion parameter GPT 3.5, which powered the original ChatGPT.

The big picture: Qualcomm CEO Cristiano Amon told Axios that he sees on-device AI fundamentally reshaping tech's competitive landscape.

• "It could change some of the established constructs that we have built in our heads about OSes and app stores," Amon said in an interview on Sunday ahead of the company's Snapdragon Summit this week in Maui.
• "You may pick the agents of your choice, and once the agents understand what you want, the agent is free and not bounded by the OS, an app store or anything like that."

Zoom in: The on-stage demos Qualcomm showed yesterday hinted at what that future could look like.

• Honor, a Chinese device maker, showed how an AI agent could be used to take complex actions that would now require human intervention along with the use of several apps.
• In one example, an Honor executive used the agent to cancel online subscriptions, with the AI system looking across AliPay and WeChat and finding a music and cloud service that could be canceled.
• Qualcomm and its partners showed a host of other ways that on-device AI could be used, from improving pet photos to automatically checking a bank balance before making a purchase.

Between the lines: A video appearance from OpenAI CEO Sam Altman at Qualcomm's event was an indication of this moment's fluidity.

• The two companies did not announce any collaborations, and their current areas don't overlap.
• Qualcomm makes chips for devices that sit at the edge of a network — PCs, phones, cars, wearables — while nearly all of OpenAI's most important technologies run from the cloud.

But Altman's presence at Qualcomm's conference makes sense in light of comments he made at a different event last week.

• There, Altman spoke about the power of on-device AI to complement what can be done in the cloud — while also noting that cloud-based AI will always be a step ahead of what can be done on an individual device.
• "I am optimistic that we will make progress towards enormously capable models that can run locally," Altman said. "Long term, we will all have huge and powerful models we can run on our own devices, and I think that's great."

The other side: Apple and Google are unlikely to easily cede control of the lucrative smartphone ecosystems they created. Both are putting renewed effort into improving their own ubiquitous assistants, Siri and Gemini, which could be deployed in ways that further entrench their power.

• Apple, for example, has talked about the opportunity for developers to highlight "app intents" that a more powerful Siri could access. But the company will still want to take its cut of digital transactions — and to maintain its present control over content.

What to watch: Amon said apps won't necessarily go away, but they could fade more into the background as AI assistants take center stage.

• As for timing, Amon said he is sure that the transformation will happen over the next five years. "Will it happen in one year, two years, three years, four years? I don't know," he said.

Meta is gradually releasing facial recognition technology across Facebook and Instagram to help users locked out of their accounts and to spot scam ads that use a celebrity's image without consent.

Why it matters: This is the first major expansion of Meta's facial recognition system since its 2021 rollback.

Zoom in: Meta has started testing facial recognition to verify the identities of users who are locked out of their Facebook or Instagram accounts, Monika Bickert, Meta's vice president of content policy, told reporters.

• The company has also been running a test of the technology to spot so-called celebrity bait ads, which often rely on AI-generated images of famous people to entice users into buying into whatever scam they're promoting.

How it works: To recover their accounts, users will be able to upload a video of their face.

• Meta's facial recognition system will then compare that video to the profile pictures on the account the user is trying to access.
• Facial recognition isn't the only option. Users can still choose to upload photos of their ID cards or other official documents.

The intrigue: Meta is not deploying the technology in the European Union, the United Kingdom and a few other jurisdictions due to local data collection laws, Bickert said.

Flashback: In 2021, Meta shut down its decade-old facial recognition system and deleted the face data of more than 1 billion users.

• Back then, the technology was used to automatically identify people in photos and suggest "tags."

Between the lines: Meta has faced several lawsuits over its use of facial recognition.

• In July, the company agreed to pay $1.4 billion to settle a 2022 lawsuit with the state of Texas over the collection of users' biometric data.

What's next: Meta plans to roll out facial recognition for account recovery to all users over the next few months, depending on how initial testing goes.

A fascinating — and frustrating — throughline in the discussion of AI's energy demand is the stack of variables that make the future cloudy as hell.

Why it matters: Powerful data centers are fueling wider growth in electricity use, but estimates of how much and how fast are all over the map.

State of play: That's driven home by a new International Energy Agency commentary, which cites variables including:

• What AI users do with the tech. Creating video, for instance, is far more energy-intensive than text generation or AI-enabled search.
• The outlook for hardware and software efficiency improvements. (Side note: It's not always clear that more efficiency = less long-term energy use.)
• Inadequate electricity generation and grid capacity where it's most needed could become a check on growth.

Catch up quick: IEA raised eyebrows last week with estimates of future data center power needs that are pretty modest compared to several others.

• It sees data centers as a pretty small contributor to this decade's overall increase in global electricity thirst.
• That said, it also emphasizes how data centers can create major strains on local grids.

• New research shows that AI is helping election lies spread more quickly. (Axios)
• Dow Jones, which owns the Wall Street Journal, and the New York Post are suing AI startup Perplexity for copyright infringement. (Variety)
• OpenAI and Microsoft are funding a $10 million fellowship program to find new ways to use AI to bolster local news. (Axios)

I still need to post your comet and northern lights photos (feel free to keep sending them in). In the meantime, I think these are my favorite comet photos for pure comedic value.