We are thrilled to unveil the Gemma 4 12B, our newest offering designed to integrate agentic multimodal intelligence seamlessly into laptops. This innovative model serves as a bridge between our edge-friendly E4B and the more sophisticated 26B Mixture of Experts (MoE), providing an impressive array of features within a compact memory footprint. Moreover, it marks our first mid-sized model equipped with native audio inputs.

Thanks to the contributions of the developer community, Gemma 4 models have achieved over 150 million downloads. From creating wearable robotic arms for personal assistance to high-level AI security solutions for businesses, the innovations you’ve introduced using our technology have been remarkable. We’re eager to see what new projects you will undertake with this addition to the lineup.

Here’s a glimpse of what differentiates Gemma 4 12B:

- **Innovative Unified Architecture**: This model eliminates the need for multimodal encoders, allowing vision and audio inputs to connect directly to the LLM backbone.
  
- **Enhanced Reasoning Capabilities**: It offers benchmark performance that approaches our 26B model, allowing for sophisticated multi-step reasoning and agent-based workflows.

- **Optimized for Laptops**: The Gemma 4 12B is designed to function efficiently on local devices with just 16GB of VRAM or unified memory.

- **Open Source Support**: Released under the Apache 2.0 license, it is fully supported across the developer community, promoting accessibility and collaboration.

- **Latency Reduction Tools**: This model includes Multi-Token Prediction (MTP) drafters, designed to enhance performance by decreasing latency.

These features collectively empower users to harness advanced multimodal capabilities on everyday hardware, all without compromising on speed or reasoning depth. Let’s delve deeper into how Gemma 4 12B makes this possible.

Experience cutting-edge agents directly on your device

Gemma 4 12B offers performance metrics that closely rival those of our larger 26B MoE model while keeping the total memory consumption below half that of its predecessor. Small enough to operate on standard consumer laptops with 16GB of RAM, it enables users to enjoy powerful multimodal and agentic experiences directly on their devices.

Unveiling Gemma 4 12B: a cohesive multimodal model without encoders.

Sunlight streams through tall windows of a minimalist research lab. At the center, a large multifaceted gemstone etched with “Gemma 4 12B” levitates above a wooden table. Around it orbit floating fragments—a torn page of prose, a pixelated image snippet, musical notes, and waveform strands—each seamlessly merging into the gem’s glowing core, symbolizing a unified, encoder-free multimodal AI.

Introducing Gemma 4 12B: a unified, encoder-free multimodal model - blog.google

- Gemma 4 12B offers multimodal intelligence for laptops with a reduced memory footprint.  
- It achieves benchmark performance similar to the larger 26B model for advanced reasoning.  
- The model is open-source, supporting developer innovation under an Apache 2.0 license.  

Today, we are introducing Gemma 4 12B, our latest model designed to bring agentic multimodal intelligence directly to laptops. Bridging the gap between our edge-friendly E4B and our more advanced 26B Mixture of Experts (MoE), Gemma 4 12B packages powerful capabilities inside a reduced memory footprint. It is also our first mid-sized model to feature native audio inputs.

Thanks to the developer community, Gemma 4 models have now crossed 150 million downloads. You’ve built everything from wearable robotic arms for physical assistance to enterprise-grade AI security. We're excited to see what you build with this latest addition.

Here’s an overview of what makes Gemma 4 12B unique:

Novel unified architecture: No multimodal encoders. The vision and audio inputs flow directly into the LLM backbone.

No multimodal encoders. The vision and audio inputs flow directly into the LLM backbone. Advanced reasoning: Benchmark performance nearing our 26B model, unlocking powerful multi-step reasoning and agentic workflows.

Benchmark performance nearing our 26B model, unlocking powerful multi-step reasoning and agentic workflows. Laptop ready: Small enough to run locally with just 16GB of VRAM or unified memory.

Small enough to run locally with just 16GB of VRAM or unified memory. Open and accessible: Released under an Apache 2.0 license with support across the developer ecosystem.

Released under an Apache 2.0 license with support across the developer ecosystem. Drafter-ready: Gemma 4 12B comes equipped with Multi-Token Prediction (MTP) drafters to reduce latency.

Together, these features bring advanced multimodal capabilities to everyday hardware without sacrificing speed or reasoning. Let's now take a closer look at how Gemma 4 12B achieves this.

Run state-of-the-art agents locally

Gemma 4 12B delivers performance nearing our larger 26B MoE model on standard benchmarks, but at less than half the total memory footprint. Small enough to run locally on consumer laptops with 16GB of RAM, it unlocks powerful multimodal and agentic experiences right on your machine.

Introducing Gemma 4 12B: a unified, encoder-free multimodal model

Engaging in infidelity might not always involve a third party, as emerging definitions of cheating suggest, notably in the realm of prenuptial agreements. Julia Rodgers, the CEO of Hello Prenups, revealed to Business Insider that some clients are asking for prenuptial contracts that broaden the conventional scope of infidelity to encompass interactions with artificial intelligence (AI). While specific examples from clients were not disclosed, an illustrative clause reflects trends in actual agreements, stating:

"knowingly and intentionally interacting with an artificial intelligence (“AI”) chatbot, virtual assistant, or similar AI-driven entity with the primary objective of developing or sustaining a romantic or sexual bond. This includes but is not limited to sharing sexually explicit messages or images, engaging in sexual roleplay or cybersex, expressing romantic feelings or exclusivity, or preserving an ongoing intimate relationship that a reasonable person would interpret as romantic or sexual."

As loneliness rates soar and AI companion applications like Replika and EVA AI gain popularity, more individuals are seeking emotional, romantic, and even sexual fulfillment from chatbots, some of whom are already involved in human relationships.

Legal experts specializing in prenuptial agreements are starting to receive inquiries about how to include these AI interactions in contracts. While many haven't personally crafted clauses addressing AI-infidelity, several view the growing concern as an extension of a trend where couples are redefining what constitutes cheating beyond physical encounters.

The evolving interpretation of infidelity points to a new frontier for couples before they commit to marriage.

Investigating 'AI Infidelity'

Hello Prenup manages an impressive volume of prenups annually, with about one-third incorporating infidelity clauses. Nonetheless, how courts will enforce AI-infidelity clauses remains unclear. Rodgers emphasized that "physical infidelity is relatively straightforward to define, but emotional virtual infidelity through a chatbot is much more challenging."

A recent study from Indiana University's Kinsey Institute highlighted that nearly 61% of single adults consider sexting or falling for an AI chatbot to be a form of cheating.

Lisa Zeiderman, a New York family lawyer, noted that this issue often arises within her clientele. Clients express frustration when their partners appear to prioritize AI companions over human relationships with family members.

Traditionally focused on asset management, prenups are now increasingly seen as agreements that set expectations for marital roles, according to attorney Aaron Thomas of Prenups.com, who likened them to partnership agreements in business.

Enforceability of infidelity clauses can become complicated even in absence of AI-related definitions, but demand for these clauses is on the rise as couples continue to expand what they view as infidelity.

For instance, a Kentucky appeals court recently upheld a prenup that specified cheating as "any form of contact with any female" outside of work, including various types of communication and even interactions during gaming.

Couples are increasingly attempting to define emotional infidelity within their agreements, and human-AI relationships are likely the next logical progression, according to Thomas. He anticipates AI-infidelity clauses will first surface in postnuptial agreements, which often tackle issues already threatening marriages.

Challenges in Enforcing AI Infidelity Clauses

California attorney Elizabeth Yang mentioned that her firm has received requests to incorporate AI-infidelity clauses into prenups but has advised against it due to potential issues with enforceability and the possibility of litigation down the road.

Drafting a clause addressing AI infidelity will require precise language, as highlighted by attorneys. A viable provision needs to include three essential elements: a clear definition of prohibited behavior, a means of determining if a breach has occurred, and specified financial implications. As Thomas stated, “a behavioral guardrail with a financial backstop.”

Possible definitions could revolve around spending thresholds on AI companions or time allocations spent in interactions with a chatbot. The agreement would also need to address evidence for breaches and the consequences, like adjusted alimony or asset redistribution.

The sample provision Rodgers revealed stipulated that if both parties agreed that one had committed infidelity, a specific financial penalty would be incurred. Moreover, it included language ensuring that even if the AI infidelity portion was unenforceable, the remainder of the prenup would still hold.

Also noting that a study found 61% of single participants considered interactions with AI as cheating, Zeiderman mentioned that AI-related cheating could potentially be managed under wasteful dissipation provisions, which restricts marital finances spent on outsiders, including romantic partners. Attorneys can use financial records to recover some of those expenses for the marital estate, suggesting a similar framework could apply to funds spent on an AI companion.

"The AI companion charge represents a new line item on credit card statements," observed Libby Leffler, CEO of First, reflecting a trend where expenses on such digital relationships arise periodically, previously marked by different platforms and spending habits over the years.

While measuring financial outlay might seem straightforward, quantifying the emotional toll of an AI affair poses challenges. "A person could find a cheap chatbot yet spend extensive hours engaged with it, which would present significant issues for a partner wanting to curb that behavior," Thomas noted.

Establishing Boundaries Regarding AI

Prenup attorneys discuss the advantages of crafting these documents with partners, emphasizing that the process often delves beyond the ultimate agreement. With AI companions still relatively new, couples may lack clarity on their mutual expectations regarding boundaries in a marriage. They may agree that AI-generated content is acceptable, while flirting with an AI companion crosses the line.

Most prenups don’t reach litigation, as one of their primary purposes is to prevent drawn-out legal disputes. Couples typically agree to terms that their attorneys then use to finalize divorce arrangements without the intervention of a judge.

In many respects, these prenuptial agreements serve as a means of setting clear expectations throughout the marriage, fostering alignment before either party initiates discussions with a new digital companion. 

If you have experiences or insights related to AI infidelity, reach out to the reporter at kvlamis@businessinsider.com.

Prenuptial agreements are facing a new dilemma: Is it possible to be unfaithful using AI?

Dark theme scene in a wood-paneled study, lit by a single warm desk lamp. A well-dressed couple sits tensely across a polished mahogany table from a lawyer. Centered on the table is a prenup document boldly titled “PRENUPTIAL AGREEMENT,” beneath a small glowing microchip encased in a glass dome—symbolizing AI. A fountain pen rests at the ready, long shadows heightening the air of uncertainty and modern intrigue.

Prenups Are Confronting a New Question: Can You Cheat With AI? - Business Insider

- Some prenups now consider emotional infidelity with AI as grounds for cheating.  
- 61% of single adults view romantic involvement with AI chatbots as infidelity.  
- Lawyers face challenges in drafting enforceable AI infidelity clauses within prenuptial agreements.

Cheating on your spouse may not require another person. That's according to some definitions of infidelity that are beginning to appear in prenuptial agreements.

Julia Rodgers, founder and CEO of Hello Prenups, told Business Insider some clients are requesting prenups that expand affairs to include encounters with AI. While she couldn't share specific customer clauses, she shared an example that reflects what's been used in real client agreements:

"knowingly and intentionally engaging with an artificial intelligence ("AI") chatbot, virtual companion, avatar, or similar AI-powered system for the primary purpose of establishing or maintaining a romantic or sexual relationship, including but not limited to exchanging sexually explicit messages or images, engaging in sexual roleplay or cybersex, expressing romantic love or exclusivity, or maintaining an ongoing intimate relationship that a reasonable person would consider romantic or sexual in nature."

With the loneliness epidemic and the explosive growth of AI companion apps like Replika and EVA AI, people are increasingly turning to chatbots for their emotional, romantic, and sexual needs — including some who are already in intimate relationships with other humans.

Lawyers who work on prenups told Business Insider they are beginning to field questions about how AI relationships could be addressed in those agreements. Most had not personally drafted an AI-infidelity provision, though several said the issue was a natural extension of a broader trend of couples redefining cheating beyond sex with another person.

While the legal questions remain, one thing's for certain: AI is giving couples one more boundary to define before they say "I do."

Related video How AI takes a toll on the freelancers who train, test, and refine it

Redefining 'infidelity' to include AI

HelloPrenup handles tens of thousands of prenups a year, a third of which include infidelity clauses, the company says. Rodgers said that while AI infidelity is already being included in prenups, how they will be enforced in court remains to be seen.

"Physical infidelity is comparatively really easy to define. Emotional virtual infidelity with a chatbot is not," Rodgers said.

A study conducted by Indiana University's Kinsey Institute last year found 61% single adults surveyed said sexting or falling in love with an AI chatbot counted as cheating.

AI companion apps like Replika AI are growing. OLIVIER DOULIERY/AFP via Getty Images

Lisa Zeiderman, a family attorney in New York, said the issue is already coming up in her work with clients seeking divorce who complain their spouse was spending more time with an AI companion than with them or their children.

Previously primarily about the handling of assets, couples are increasingly using prenups to set out their expectations for the marriage, according to Aaron Thomas, attorney and founder of Prenups.com, who said, "Couples are treating prenups like partnership agreements. If you're going into business with a partner, you lay out all the terms."

Infidelity clauses can be challenging to enforce even without the added question of a non-human lover, but some lawyers said they're seeing increased interest in them, with couples also expanding the definition of what might count as cheating.

An appeals court in Kentucky last month upheld a prenup and postnuptial agreement that defined cheating as, among other things, "any type of contact with any female" outside work, including phone calls, texts, emails, and interactions while playing video games.

"Increasingly, couples are defining emotional affairs to the best that they can as falling under infidelity," Thomas said, adding that human-AI relationships would be a logical continuation of that.

He thinks AI-infidelity clauses are likely to start appearing more in postnup agreements first, which commonly address issues that have already threatened the marriage.

Enforcing AI infidelity clauses could be a challenge

Elizabeth Yang, an attorney in California, said clients have asked her firm to include AI-infidelity clauses in prenups. The firm has discouraged them because of concerns about enforceability and the potential for future litigation.

Including AI infidelity in a prenup would require very careful wording, the lawyers said. Thomas said a workable provision would need three components: a definition of the prohibited behavior, a method for determining whether a breach occurred, and a stated financial consequence. As he put it, "a behavioral guardrail with a financial backstop."

Possible definitions could include spending more than a set dollar amount on an AI companion or exceeding a specified number of hours talking with a chatbot. The agreement would also need to say how a violation would be proven and what financial consequences would follow, like reduced alimony or a smaller share of assets.

The example provision for AI infidelity that Rodgers shared stipulated that if both parties agree one spouse had cheated, they would owe a specific dollar amount. It also included language that said if this part of the prenup was not enforceable, the rest of the document would still be valid.

A study found 61% of single adults surveyed considered sexting or falling in love with a chatbot as cheating. humanmade/Getty Images

Zeiderman said AI cheating could be addressed through wasteful-dissipation provisions, which cover marital money spent outside the marriage, including on romantic partners. Lawyers can use financial records to calculate the spending and seek to restore some of it to the marital estate, she said, adding that the same framework could theoretically apply to spending on an AI romantic companion.

"The AI companion charge is like the new mystery line item on the credit card statement," said Libby Leffler, founder and CEO of First, adding that five years ago it was OnlyFans and 10 years ago it was a gambling debt. "The format keeps changing, but the awkward discovery doesn't."

While spending might be easiest to measure, it may not capture the full harm of an AI affair. "If somebody can find a chatbot that's super cheap, but they're spending 20 hours a week going down the rabbit hole with this AI companion, that's still going to be hugely problematic for the couple or for the spouse that wants to stop that behavior," Thomas said.

Setting AI boundaries

Prenup lawyers say the benefits of drafting one with your partner go beyond the end result of the document.

With AI companions being so new, couples may not know where their partners stand on what's appropriate in a marriage. Maybe AI-generated porn is OK, but flirting with an AI companion with a name and face is not. Maybe the occasional sexting session with a chatbot is acceptable, but paying a hefty subscription for an ongoing relationship would be crossing the line.

Most prenups are never litigated, the lawyers said, noting that one of the points of the document is to avoid lengthy legal proceedings. Instead, the couple agrees to the terms in the prenup, and their lawyers use it to settle the terms of their divorce without a judge having to weigh in.

In some ways, they said the prenup is about more than enforceability during a divorce — it's about setting expectations for your marriage while you're married, so at least you're on the same page before either of you starts a new chat.

Do you have a story to share about AI infidelity? Contact this reporter at kvlamis@businessinsider.com.

Prenups are confronting a new question: Can you cheat with AI?

We are currently navigating a fascinating and complex business landscape. The rapid advancements in artificial intelligence over the past five years have dramatically reshaped our understanding of workplace dynamics, leaving many individuals apprehensive or disheartened about the future, particularly those who feel distanced from this transformative technology.

In contrast to the gradual integration of cloud computing, which ultimately facilitated greater convenience by allowing software products to be accessed online, the rise of AI brings a more profound shift. Many are grappling with the implications of empowering digital entities, which stands in stark contrast to traditional views of human value in the workforce. This is particularly evident in the United States, where job status significantly influences personal identity, posing challenges for those who are unemployed.

Given this evolving context, a growing number of professionals are rethinking the traditional career model. Instead of viewing themselves as integral parts of a corporate machine or members of small teams, some are embracing the concept of the “solopreneur.” This self-employed approach offers an appealing alternative to conventional employment paths, especially as AI disrupts established notions of job security.

A recent article featured on the AI Daily Brief, a platform hosted by podcaster Nathaniel Whittemore, sheds light on this trend. It highlights how AI is reshaping perceptions of career safety, noting that while AI reduces the barriers to starting a business, it also challenges the established risk perception of startups versus corporate jobs. With uncertainties surrounding the future of corporate roles, the once-clear distinction between “startup = risky” and “corporate = safe” is becoming increasingly blurred.

Supporting this shift towards solopreneurship are some notable statistics. For instance, data reveals that 63% of C-corporation filings in Q2 2026 featured a single founder. Furthermore, applications for solo businesses in sectors heavily influenced by AI surged by 27%. Though these figures are complex and contextual, they underscore the growing trend of individuals opting to establish their own enterprises.

Another intriguing statistic indicates a 20% increase in solo self-employment within AI-driven occupations between 2022 and 2025, specifically among founders of AI-centric ventures.

On a related note, Tesla has recently instituted a new spending limit of $200 per week for each employee, showcasing the different dynamics at play between lean startups and larger, more bureaucratic corporations. Startups are agile and can quickly adapt to market changes, unlike corporations, which often find themselves constrained by extensive organizational structures. The advent of sophisticated AI technologies that can operate on smartphones has leveled the playing field, enabling many solopreneurs to generate substantial revenue with minimal overhead.

As we observe these trends unfold, it appears that emerging graduates and industry advisors are increasingly encouraging a solopreneurial mindset. For those not pursuing trades such as welding or electrical work, embarking on a solo venture armed with creativity, determination, and intelligence might offer the most promising path forward.

As this narrative evolves, it will be intriguing to see how the future of work continues to transform, especially for those ready to embrace this new era of entrepreneurship.

AI Startups, Independent Entrepreneurs, and Real Data

A bright, sunlit co-working loft. At a wooden desk sits a focused solopreneur typing on a sleek laptop, surrounded by printed charts, a steaming coffee cup, and a small sculpted robot figurine representing AI startups. Behind them, a floor-to-ceiling glass whiteboard is filled with colorful bar graphs, pie charts and bold percentage figures titled “Actual Numbers.” Warm natural light streams through tall windows, casting soft shadows across the workspace and highlighting the energy of independent AI innovation.

AI Startups, Solopreneurs, And Actual Numbers - Forbes

- The rise of AI is reshaping career perceptions, prompting a shift towards solopreneurship.  
- C-corp filings show 63% of new businesses have a single founder in 2026.  
- Solo self-employment in AI-related roles is projected to rise by 20% by 2025.  

3D illustration of silhouettes of a group of gray persons walking leftward, and a single green person walking rightward, tilted Solopreneur. getty

It’s a strange time for business. Essentially, the dramatic rise of AI over the last five years or so has re-ordered so much of what we used to think about how enterprise works, that a lot of people who are less thrilled, or further from the technology itself, are rather demoralized, or grouchy, about it.

The cloud wasn’t like this. Its reach was more limited, and its power more incremental. It didn’t bother people so much that you could offer any software product through the web. In fact, it felt convenient.

Part of what’s bothering people now is this idea of giving agency to digital entities. It’s a direct challenge to the core value that we ascribe to humans, especially in America. We have this collective concept that you’re defined by your job – which is a problem if you don’t have one.

The Face of Work

In that context, some are starting to think that it’s a better idea to approach the work world from a different perspective: as a self-employed “solopreneur” than as a cog in a corporate wheel, or even a member of a smaller team in something like a C-corp.

I was taking a look at the AI Daily Brief web site, where my favorite podcaster, Nathaniel Whittemore, hosts these broadcasts on what’s happening in the AI world. There’s an article on the site with this blurb:

“AI is flipping which career path counts as 'safe' … AI lowers the activation cost of building a startup, but it also upends assumptions about what safe looks like on the corporate side. When people aren't sure what corporate roles will even look like post-AI, the old risk spectrum of 'startup = risky, corporate = safe' stops holding.”

That’s a big endorsement of the startup path, and the urge to “go it alone” – but unlike some of the other sources where I’ve read this kind of stuff, the AIDB site provides actual numbers. Let’s go over a few.

Findings on Solopreneurship

Okay, some of these statistics are a little complex – but they’re real indicators, and you’ll see that they clearly point to a rise in self-supported startups:

The first one is the percentage of C-corp filings with a single founder, in the second quarter of 2026, which just ended. That number is 63%. So that’s the percentage of filing businesses of this type.

Another is an increase, of 27%, in solo business applications, specifically in “high-AI” sectors. So 27% more of these AI-type businesses have one solopreneur. You can see how these numbers are segmented, partial and based on changing context. But you can also see that many new entrepreneurs are choosing to go it alone.

I must confess being puzzled about this one. The number: 25%. The metric: how much leaner AI-native startups run (compared to others? On what time frame?) at the same valuations.

I’m just going to leave that there.

The next one, an estimated 20% rise in solo self-employment in AI-exposed occupations, from 2022-2025, has a time frame, and we also know what we’re measuring: numbers of founders in AI-related businesses going solo.

Market Moves

And now for something completely different. Here’s a good one: Tesla’s new token spending cap, $200 per week, per employee.

That doesn’t say anything about solopreneurs, but some of these numbers were also mixed in above.

What you can see is that startups are lean, and corporations are brittle. Startups have the capacity to pivot quickly, and corporations are weighed down by bureaucracy and their own complex org charts. Corporations made more sense when you needed a lot of infrastructure to make an AI play. Now you can fit an evolved LLM on a smartphone. A lot of these spotlighted solopreneurships are making money hand over fist, by scaling quickly, without a lot of overhead. And that’s the name of the game in any business: keep costs low, and bill.

We’ll see how this kind of thing shakes out over the next few years. But what I’m hearing from new grads, and others who are advising them, is that if you’re not going to be a welder or an electrician or in skilled trades, it makes more sense to be a solopreneur, an army of one, armed with ideas, and grit, and determination, and smarts.

Go to it, young folks. Stay tuned.

AI Startups, Solopreneurs, And Actual Numbers

When it comes to your privacy preferences, there are a few options to consider. By selecting "Accept all," you agree to allow us and our partners to collect technical identifiers as well as browsing and search information. This data is utilized for analytics, personalized advertising and content, advertising measurement, and audience research, along with developing services. Should you click "Reject all," you are opting out of our use of cookies and personal data for these purposes.

If you prefer to tailor your privacy settings, you can choose "Manage privacy settings." Your consent is not permanent; you can modify or retract it at any moment by accessing the "Privacy and Cookie settings" or "Privacy dashboard" options available on our websites and applications. For additional details on how we handle your personal information, please refer to our Privacy Policy and Cookie Policy.

Your options for privacy

A cozy home office bathed in warm morning light. In the foreground, adult hands hover over a tablet screen showing a “Your Privacy Choices” dashboard: toggles for Location, Cookies, Ad Tracking, Camera, each with a small padlock icon. One switch is mid-motion, glowing softly. On the wooden desk, a pair of glasses and a notebook with “Privacy Matters” scribbled add a personal touch.

Fight over AI data centers reaches rural North Carolina, with a 24-month halt on the table - Yahoo

- Users can accept or reject cookies for personalized advertising and analytics purposes.  
- Customization of privacy choices is available through 'Manage privacy settings' option.  
- Consent can be withdrawn or changed anytime via 'Privacy and Cookie settings' links.

Your privacy choices

Accept all', we and technical identifiers and browsing and search data, for analytics, personalised advertising and content, advertising and content measurement, and audience research and services development. If you click '', we and our partners , including 250 who are part of the IAB Transparency & Consent Framework, will also store and / or access information on a device (in other words, use cookies) and use precise geolocation data and other personal data such asand browsing and search data, for analytics, personalised advertising and content, advertising and content measurement, and audience research and services development.

If you do not want us and our partners to use cookies and personal data for these additional purposes, click 'Reject all'.

If you would like to customise your choices, click 'Manage privacy settings'.

You can withdraw your consent or change your choices at any time by clicking on the 'Privacy and Cookie settings' or 'Privacy dashboard' links on our sites and apps. Find out more about how we use your personal data in our Privacy Policy and Cookie Policy.

Your privacy choices

EMERADO — During an event at Grand Sky on Friday, the AI company Armada presented its latest innovation, a transportable data center called Triton. Measuring 40 feet in length, this compact center can easily be towed by a semi truck and is touted by Armada as fully operational in a variety of global environments, from the arid deserts of Saudi Arabia to the frigid landscapes of interior Alaska, where temperatures can drop to minus 50 degrees.

“This tool will not only draw in major companies like Northrop Grumman and General Atomics, but it will also inspire innovators, including college students,” remarked U.S. Senator Kevin Cramer, R-North Dakota, during the launch.

The introduction of Triton coincides closely with the first meeting of North Dakota’s AI and Data Center Committee, which gathered two days prior to discuss the increasing demands placed on energy resources by these technologies. Data centers were responsible for approximately 4-5% of energy consumption in 2024, a figure projected to rise to between 9-17% by 2030. Rep. Mike Nathe, R-Bismarck, noted that the state may struggle to meet this impending demand.

Armada's Chief Operating Officer, Jon Runyan, voiced enthusiasm for promoting the use of Triton among various tenants located in both Grand Forks and at Grand Sky. “We invite everyone, from the Air Force to General Atomics and local universities, to explore the opportunities this unit provides,” he stated. The Triton will primarily serve to gather and process data for unmanned aircraft system missions within the aviation park, with the company generously offering its use free of charge for one year.

Additionally, Armada is collaborating with the federal government on President Donald Trump’s proposed Golden Dome missile defense initiative, a project that Cramer has championed to enhance North Dakota’s role in national defense operations. Earlier in the year, Cramer backed a Senate funding bill that includes a $5 million allocation for the Perimeter Acquisition Radar Attack Characterization System at Cavalier Space Force Station, which plays a pivotal role in monitoring objects in orbit integral to the Golden Dome project.

“The Golden Dome aims to integrate both new and current technologies to establish a protective barrier against incoming threats, whether they be missiles, aircraft, drones, or autonomous robots,” stated Cramer. “This kind of integration facilitates not only the collection of data but also its rapid transmission.”

The capabilities of each Triton are contingent upon the processing power of the computers it houses, according to senior project manager Tori McCaffrey. Customers have the option to provide their own GPU and CPU or source them from suppliers through Armada.

McCaffrey emphasized that Triton is energy "agnostic," meaning it can operate using power derived from a wide array of sources, both renewable and conventional. “We can utilize stranded energy resources effectively,” she noted, providing the example of deploying Triton on an oil rig in the North Sea, where a generator is often required.

She further highlighted that Armada’s data centers stand apart from traditional facilities due to their smaller footprint and reduced energy consumption, as well as their non-reliance on water for cooling purposes, though the centers do emit noticeable sound levels.

AI company launches mobile data center during Grand Sky event

A sunlit airfield scene at dawn, a sleek dark-gray semi-trailer parked before a hangar branded “Grand Sky,” its side panels folded open to reveal rows of high-density server racks basking in warm natural light. A small group of sharply dressed professionals stands nearby, pointing and discussing enthusiastically as the mobile data center hums to life against the backdrop of a clear sky.

AI firm unveils mobile data center at Grand Sky event - Grand Forks Herald

- Armada unveiled Triton, a transportable 40-foot data center, capable of global deployment.  
- Triton will support data processing for unmanned aircraft missions at Grand Sky, free for a year.  
- The data center is energy agnostic, utilizing various power sources, including stranded energy.

EMERADO — Artificial intelligence company Armada showcased its new transportable data center during an event on Friday at Grand Sky.

Named Triton, the 40-foot-long data center is compact enough to be pulled by a semi truck. Armada touts it as being functional anywhere on the planet — from the desert climate of Saudi Arabia to the minus 50 degree temperatures of interior Alaska.

ADVERTISEMENT

“This just becomes one more tool to attract, not just another Northrop Grumman or General Atomics, but another inventor — another college student,” said U.S. Sen. Kevin Cramer, R-North Dakota, in his remarks at the event.

Triton’s unveiling comes two days after North Dakota’s AI and Data Center Committee met for the first time to discuss the strain the technologies have on energy use. Data centers accounted for about 4-5% of energy use in 2024, but is expected to increase to 9-17% by 2030. Rep. Mike Nathe, R-Bismarck, said at the meeting that the state simply won’t be able to keep up.

“We want to encourage the use and testing of this unit by all of the tenants here in both Grand Forks and at Grand Sky,” Armada Chief Operating Officer Jon Runyan said. “Whether you’re the Air Force, the National Guard, General Atomics, the University, Northrop — we encourage you to come.”

Triton will be used to collect and process data used in the aviation park’s unmanned aircraft system missions. Runyan said his company is allowing Grand Sky to have the data center free of charge for a year.

Additionally, Armada is partnering with the federal government on President Donald Trump’s planned Golden Dome missile defense system — a project praised by Cramer, who had pushed to make North Dakota a major part in its operation.

Earlier this year, Cramer supported a Senate appropriations package that includes a $5 million investment at Cavalier Space Force Station’s Perimeter Acquisition Radar Attack Characterization System, which tracks objects in orbit and is expected to be a key part of the Golden Dome project.

“The Golden Dome is the integration of new and existing capabilities with the focus of creating a dome that protects us from incoming missiles, aircraft, drones or, you know, robots,” Cramer said. “When you’re talking about integration, you're talking about the ability — not just to collect data — but to move data quickly.”

ADVERTISEMENT

How powerful each Triton is depends on the processing power of the computers housed inside, said Tori McCaffrey, senior project manager. Customers source GPU and CPUs themselves or from suppliers through Armada.

McCaffrey described Triton as being energy “agnostic,” meaning it can run on power produced from a variety of sources, renewable or not.

“We’re able to take advantage of stranded energy sources,” she said. “For example, we’re deploying these on an oil rig in the North Sea. In that case, you probably need a generator because there’s not another option around.”

McCaffrey added that Armada’s data centers differ from the traditional brick and mortar centers in that they take up less space, use less power and don’t use water for cooling. However, the sound emitted from the data centers is still notable.

Visitors tour the Triton Modular Data Center during an event Friday at GrandSKY. Eric Hylden / Grand Forks Herald

AI firm unveils mobile data center at Grand Sky event

The discussion around AI surveillance cameras is intensifying, with significant concerns being raised about privacy and the potential for misidentification. Yet, are these AI-driven monitoring systems proving their worth in public safety?

At a lengthy city council meeting in Huntington that extended over seven and a half hours, Police Chief Phil Watkins presented a compelling case for investing $2 million into such technology. He emphasized that by integrating into a national data-sharing network, officers can access vital information more swiftly and effectively.

"This technology could drastically reduce the time it takes to apprehend suspects—from days down to hours or even minutes," Watkins remarked during the session.

Watkins clarified that their focus isn't solely on acquiring cameras; they're participating in a vast collaborative safety network aimed at eliminating safe havens for criminals in Huntington.

He pointed out that typically, law enforcement depends on 911 calls to determine the location of gunfire; however, callers can often be delayed or unsure about their exact locations. The use of AI cameras enables officers to quickly identify potential gunfire incidents and pinpoint their locations more accurately. This not only increases the chances of apprehending suspects but also facilitates prompt medical assistance to victims.

Mayor Patrick Farrell shared a recent success story during the meeting, where license plate readers played a crucial role in capturing a man linked to a shooting earlier in the month. This incident happened when a suspect returned to Huntington after allegedly shooting Jackie Harper. Thanks to license plate readers in another city, law enforcement was able to track the vehicle and arrest the suspect in Michigan.

"I've spoken with multiple mayors from various cities, and those that have implemented Flock cameras have reported significant drops in both violent and property crimes," Farrell stated, citing examples like San Francisco, San Jose, Oakland, and Las Vegas.

Despite these favorable outcomes, concerns persist. In April, residents in San Jose initiated a federal class action lawsuit against the surveillance system. Additionally, an audit in San Francisco revealed that data collected from Flock cameras was accessed by unauthorized entities.

In response to these concerns, Mayor Farrell assured that measures will be established to prevent misuse of the surveillance system, although he did not provide specific details on these safeguards.

Authorities claim Flock AI cameras might expedite arrests and reduce crime rates.

A sunlit urban street at dawn, a sleek Flock AI camera with its logo emblazoned on the housing perched atop a lamppost. Below, an adult suspect on a motorbike leaves dynamic motion trails as the camera’s lenses lock on, while a patrol car slows in behind—warm natural light flooding the scene.

Officials say Flock AI cameras could speed captures and cut crime - WCHS

- AI-powered surveillance aims to improve response times and enhance public safety in Huntington.  
- License plate readers aided in capturing a murder suspect linked to a recent shooting.  
- Concerns over privacy and misuse have arisen alongside claims of reduced crime rates.  

The debate surrounding AI cameras point out multiple issues, privacy issues and misidentification being two of the largest. But has AI powered surveillance actually been helpful?

During Huntington's 7 and a half hour long city council meeting on Monday night, Police Chief Phil Watkins said that by joining this network, officers can easily share and access data across departments on a national level. He noted this is a $2 million investment into public safety to ultimately help with quicker responses.

"If we had this technology as proposed, it could potentially shorten suspects' capture from days to hours or even minutes," Watkins said during Monday's meeting.

He said, they're not just buying cameras, they're joining the largest collaborative safety network available to deny criminals a safe haven in Hutington.

"We often rely on 911 to know where shots were fired but callers may be delayed or even uncertain of where the exact location is. This tool makes officers aware of potential gunshots faster and it gives us more precise locations where the incident has occurred. This enhances the likelihood of capturing the suspects but more importantly it also allows us to expedite providing medical treatment to any potential victims," he said.

Mayor Patrick Farrell said, these license plate readers caught a man suspected in the shooting death of someone from Huntington earlier this month in a different city, crediting the technology for bringing someone to justice.

"After the murder suspect came to Huntington, shot Jackie Harper twice in the head, and he left and he was going back to Michigan. Luckily for us, some other city had licensed plate reader cameras that caught the car, that gave us the lead that led to his arrest in Michigan," he said during the city council meeting.

He also noted he's spoke with multiple mayors from across the country and he said the cities that saw the largest decline in violent and property crimes use Flock cameras.

He noted cities like San Francisco, San Jose, Oakland and Las Vegas.

"Every one of those cities, they're using Flock. They were using it to catch the worst criminals and that led to significant reductions in violent crime, property crimes, and all other categories of crimes," he said.

However in April, three people living in San Jose filed a federal class action lawsuit over the mass surveilance system. In San Francisco, an audit found that the data from their Flock cameras was accessed by outside agencies.

Mayor Farrell said that they will be putting guardrails in place to ensure this system is not abused, but did not elaborate on what those are.

Officials say Flock AI cameras could speed captures and cut crime

According to sources speaking to The Post, the New York Mets were one of the teams identified as having taken advantage of a loophole tied to the utilization of artificial intelligence for making in-game decisions during the first half of the season.

To keep up with the latest on the Mets, consider joining Post Sports+, which offers subscribers exclusive features such as real-time texting with insider Mike Puma, providing the latest buzz on the team. There’s a free trial available.

In response to this situation, Major League Baseball has officially prohibited the practice, with the new rules taking effect starting Wednesday.

This crackdown originated from a memo sent to teams on June 11 but only went into effect just before the season's second half commenced after the All-Star break.

As reported by The Athletic, it was determined that certain teams were using league-issued iPads in the dugout for purposes that extended beyond their intended uses, such as offering recommendations on player substitutions, pitch selections, and other strategic decisions that are traditionally made by players and coaches.

The memo, written by MLB's executive vice president of baseball operations, Morgan Sword, was directed to general managers, assistant general managers, and video coordinators.

The Mets' dugout was photographed during a game earlier in the season, as reported by Corey Sipkin for the NY Post.

While teams can still use iPads in the dugout, these devices provide access to Statcast data and various video feeds, including relevant information related to the automated ball-strike system. However, any video content accessible during the game is available only on delay.

The custom applications that allowed teams to tap into other software will no longer be available under the new guidelines.

The regulation change may affect teams that have started to call pitches directly from the dugout—a practice that has gained traction this season.

This shift will likely have broader implications as organizations around the league increasingly integrate analytics into every facet of the game.

For deeper insights on the Mets, sign up for "Inside the Mets," an exclusive newsletter by Mike Puma available through Sports+. To subscribe, please provide a valid email address and ensure you agree to the Terms of Use and Privacy Policy. Additionally, check out more newsletters for ongoing coverage.

MLB has permitted the use of iPads in dugouts for over ten years and noted in the memo that, following a review earlier this year, teams were deemed "compliant" with regulations concerning sign stealing and electronic device usage, thus avoiding penalties.

Teams were effectively given a month to cease AI’s role in their in-game decision-making to make appropriate adjustments.

Mets among teams that exploited AI loophole with dugout iPads prior to MLB enforcement.

A dark, tense nighttime dugout lit by a single overhead bulb. A Mets logo is painted on the wooden bench where a coach in a team jacket crouches, secretly peering at a glowing iPad tucked under the seat. The iPad screen shows scrolling pitch analytics in warm amber light. Around him, players in shadows glance nervously toward the field, and an MLB patch hangs on the wall—capturing the hush of a looming AI loophole crackdown.

Mets among teams who used AI loophole with dugout iPads before MLB crackdown - New York Post

- The Mets exploited a loophole allowing AI use for in-game decisions this season.  
- MLB banned AI use in-game effective the day before the second half of the season.  
- Teams had a month to adjust after being notified of the new regulations.  

See more of our coverage in your search results.

The Mets were among the teams found to have exploited a loophole regarding the use of artificial intelligence for in-game decisions during the first half of the season, sources told The Post.

Access the Mets beat like never before Join Post Sports+ for exciting subscriber-only features, including real-time texting with Mike Puma about the inside buzz on the Mets. Try it free

Major League Baseball banned the practice beginning Wednesday.

Advertisement

The crackdown was issued in a June 11 memo to teams but wasn’t put in place until the day before the second half of the season began following the All-Star break.

As first reported by The Athletic, MLB found that some teams were using league-issued iPads in the dugout “beyond their originally intended purpose to include recommendations regarding substitutions, pitch calling, and other in-game decisions traditionally made by players and coaches.”

The memo, from MLB executive vice president of baseball operations Morgan Sword and obtained by The Post, was sent to general managers, assistant GMs and video coordinators.

The Mets’ dugout is pictured during an April 12 game. Corey Sipkin for the NY Post

Advertisement

The tablets, which still will be permitted in dugouts, give teams access to league-provided Statcast information, as well as multiple video angles and data related to the automated ball-strike system.

Video accessed during the game in the dugout is only available on delay.

The custom tab allowed teams to access other programs and will no longer be accessible.

Advertisement

The move could impact teams that began calling pitches from the dugout, a trend that has become more widespread this season.

It will no doubt have other impacts as teams throughout the league have relied more heavily on analytics in all aspects of the game.

Delivering insights on all things Amazin’s Sign up for Inside the Mets by Mike Puma, exclusively on Sports+ Thank you Enter your email address Please provide a valid email. By clicking above you agree to the Terms of Use and Privacy Policy. Enjoy this Post Sports+ exclusive newsletter! Check out more newsletters

Advertisement

MLB has permitted the use of iPads in dugouts for over a decade and noted in the memo that a review earlier this year showed that teams were “compliant” with rules regarding sign stealing and electronic device usage, so no penalties were handed out.

Teams were essentially given a month to stop using AI for in-game decision-making purposes “to make any necessary adjustments.”

— Additional reporting by Joel Sherman

Mets among teams who used AI loophole with dugout iPads before MLB crackdown

Sign in to share your thoughts on videos and immerse yourself in the excitement.

Investment specialist discusses his view on why he believes AI expenditures do not contribute to inflation.

A bright, sunlit modern boardroom with floor-to-ceiling windows overlooking a city skyline. A sharply dressed investment expert stands beside a sleek digital presentation screen showing two clean line graphs: one climbing sharply marked “AI Spending” with small chip icons, the other running flat labeled “Inflation Rate.” He gestures confidently at the screen, conveying calm authority and clarity.

Investment expert explains why he thinks AI spending isn’t inflationary - Fox Business

- Users can now log in to comment on videos for enhanced interaction.  
- The update aims to foster community engagement and user participation.  
- Comments will allow for a more dynamic and interactive user experience.  

Log in to comment on videos and join in on the fun.

Investment expert explains why he thinks AI spending isn’t inflationary

The smartphone industry is currently witnessing an intense competition focused on integrating artificial intelligence into devices to attract everyday users. However, Vertu, the luxury phone manufacturer established in the UK, is taking a unique route. Known for their handcrafted phones priced in the tens of thousands, Vertu has carved out a niche by offering opulent status symbols rather than engaging in the tech-spec race. Its latest model, the Alphafold, is aimed at affluent individuals, particularly executives, blending high-end materials with an AI assistant designed to streamline their daily tasks.

To explore Vertu's claims, I shifted my focus from typical smartphone metrics like benchmark performance and camera quality. Instead, I spent several days using the Alphafold as Vertu suggests its customers would, handling tasks such as document management, spreadsheet analysis, trip planning, and routine automation, while leveraging the AI assistant as a support tool throughout my workday. The core query was not whether the Alphafold is a competent smartphone, but whether it genuinely serves as a valuable tool for executives.

At the core of the Alphafold is Hermes Agent, an AI built atop the open-source Hermes project. This assistant is marketed as capable of file analysis, task automation across applications, and managing conversations while outsourcing intricate requests to a human concierge when necessary. In a departure from typical smartphone AI assistants that merely react to inquiries, Hermes is crafted to carry out complex workflows on behalf of users, positioning it as a crucial selling point alongside the device itself.

The Alphafold, priced starting at $6,880, exudes luxury. My review unit was enveloped in authentic calfskin leather and accented with titanium, distinguishing it from other foldable models that predominantly utilize glass or synthetic materials. It's evident that the phone targets users who view their devices as both functional tools and luxurious status symbols.

In comparison to the Samsung Galaxy Z Fold 7, which I used as a benchmark throughout my evaluation, the Alphafold is noticeably heavier at 264 grams, versus Samsung's 215 grams. While this added weight makes extended use feel more substantial, it retains comfort in handling. The Alphafold’s contoured design facilitates unfolding more seamlessly than the flatter edges of the Galaxy Z Fold 7, although Samsung's model feels more streamlined when folded, providing easier one-handed operation.

The Alphafold’s packaging mirrors that of a high-end jewelry box rather than a standard phone package, as it features an oversized case with neatly arranged drawers for accessories like a leather sleeve and charging cables. This packaging enhances the perception that Vertu is delivering a luxury experience rather than just a smartphone.

Beneath its premium façade, the Alphafold reveals notable similarities to the $1,100 ZTE Nubia Fold, particularly in hinge design, weight and button placements, and other components. The primary aesthetic differentiator is the luxurious leather back. Investigation into the system software also showed ZTE identifiers present within the software architecture.

In response to queries about these similarities, Vertu acknowledged to TechCrunch that the Alphafold was developed in partnership with ZTE/Nubia, utilizing their hardware and production expertise. However, they emphasized that they were in charge of the luxury elements, software optimization, and overall quality assurance. ZTE did not provide any comments when approached.

This is not an isolated instance for Vertu. A 2023 Wired review of the MetaVertu indicated that it had characteristics reminiscent of a ZTE Nubia phone, hinting at Vertu's approach of enhancing existing ZTE designs with upscale materials and tailored software.

Yet, the focus on hardware might obscure the true essence of the Alphafold. Vertu isn't merely aiming to produce a superior foldable; rather, it's betting on whether busy executives will invest in an AI assistant capable of enhancing their efficiency.

Throughout my testing, I employed the Alphafold as my primary device, shifting from simple commands to more complex executive-level tasks. I engaged Hermes in analyzing spreadsheets, organizing business travel, and managing my agenda, then juxtaposed this experience with Samsung's Galaxy Z Fold 7 featuring Google’s Gemini AI.

As the testing progressed, I encountered various experiences. Initial software iterations faced challenges with file uploads, image analyses, and connecting with Vertu’s concierge service. After highlighting these issues, Vertu implemented updates that rectified these functionalities, allowing me to complete the necessary tests.

The results revealed a more complex narrative than the company's claims conveyed. Hermes performed admirably with file and spreadsheet analyses, where Samsung’s Gemini often required manual uploads. Furthermore, Hermes displayed greater agility in automating inter-app actions, completing multi-step processes. However, this independence raised questions about when AI should operate autonomously versus when it should seek user clarification.

In one of the initial tests simulating a common executive scenario, I instructed Hermes to inform a contact about my delayed arrival, trigger navigation to the airport, activate Do Not Disturb mode, and remind me to contact the hotel in 15 minutes. Hermes successfully sent the message, engaged Do Not Disturb, and prompted Google Maps but failed to initiate navigation and set the reminder for the wrong time.

Conversely, running the same scenario with Samsung’s Gemini led to a more interactive and ultimately successful experience, as it checked in with follow-up queries before executing actions and accurately setting reminder times.

In another task aimed at arranging a business trip from Mumbai to Pune, Hermes suggested no direct morning flights and offered to escalate the request to its concierge service, but inaccurately entered the trip on my calendar for the wrong dates. Samsung’s Gemini, while unable to find a direct flight, continued planning by proposing alternatives instead of deferring the request.

The handling of business documents provided a mixed review as well. When requesting analyses of a locally stored financial spreadsheet, Hermes initially produced results but later failed to recognize the document in subsequent queries, requiring a fresh upload to proceed. In contrast, Gemini maintained context without needing re-uploads, which proved useful for follow-up discussions.

Overall, Hermes presents itself as an ambitious AI assistant still in development. Its willingness to take autonomous actions creates a more agent-like experience compared to Gemini, yet it also leads to partial workflows and inconsistencies. Observations during my testing indicated that Vertu is actively refining the platform, implying user experiences may evolve significantly over the coming months.

In addition to general functionalities, Hermes also includes specialized AI agents focusing on areas like legal and investment insights while offering the option to connect with a human concierge. This approach aims to position the Alphafold as more than just a premium device but rather as a comprehensive digital assistant crafted for executives.

However, these specialized agents should be viewed as preliminary sources rather than definitive advisors. Although they provide helpful summaries and suggestions, their AI-generated responses should not be blindly trusted, especially when it comes to critical legal or financial matters. The availability to refer specific requests to a human concierge reflects the current limitations of AI technology; human judgment is still essential.

Furthermore, Vertu aspires for the Alphafold to be recognized as a business tool rather than just another smartphone. The company demonstrated an enterprise resource planning (ERP) system integrated into the device, enabling executives to access business data and workflows. However, my assessment of this feature remained superficial, confined to a demo environment, which made it challenging to gauge its effectiveness in real-world applications or integration with existing business systems.

When it comes to security — a vital concern for many executives — Vertu contends that communications with Hermes Agent are encrypted and data handling protocols ensure anonymity in AI training. Users also have the autonomy to select where their data is processed, with enterprise options designed for heightened data protection. To bolster their claims, Vertu incorporates an “A5” security chip, which they assert adds hardware-level security for sensitive information and encrypted exchanges. While this could not be independently verified during testing, it represents a cornerstone of Vertu's value proposition for corporate users.

In everyday usage, the Alphafold behaves similarly to any advanced foldable, with battery life comfortably exceeding a day. However, the absence of wireless charging stands out as a surprising drawback, particularly given that the Galaxy Z Fold 7 includes both Qi charging and wired USB-C options.

The camera also features a document scanning mode labeled “Smart AI,” capable of recognizing and enhancing paperwork, which is beneficial for digitizing important documents. However, Samsung provides a comparable feature, making this aspect feel less like a unique selling point.

Ultimately, while the Alphafold aspires to be a groundbreaking AI-centered luxury smartphone, it falls short of justifying its hefty price tag. Although its exclusive materials and services are noteworthy, the underlying hardware does not offer significant advancements over more affordable foldable options, and the Hermes Agent remains a work in progress rather than a definitive reason to invest heavily. 

In conclusion, Vertu seeks to persuade consumers to pay a premium for its brand, craftsmanship, and suite of AI features layered on an established smartphone foundation. Based on my experience, the rationale for this significant investment is hard to support, especially when alternatives like the Samsung Galaxy Z Fold 7 deliver a more mature foldable experience with similar functionality at a significantly lower cost. With the anticipated release of the next-generation Galaxy Z Fold 8 approaching, the Alphafold's competitive edge becomes even more questionable.

Vertu expects executives to invest $6,880 in an AI agent — here’s the true performance analysis.

A sleek corporate boardroom bathed in warm, natural daylight. Three sharply dressed executives sit around a polished mahogany table. One executive pushes a gleaming gold card stamped “$6,880” toward a small, cube-shaped AI device on the table, its metal casing bearing a large Vertu logo. Behind them, a wall-mounted screen displays a performance gauge stuck at 50%, flickering. Tension ripples through the room as the executives exchange doubtful glances.

Vertu wants executives to pay $6,880 for an AI agent — here’s how it actually performs - TechCrunch

- Vertu's Alphafold targets affluent executives with luxury materials and an AI assistant.  
- Hermes Agent automates executive tasks but lacks the reliability of competitors’ AI.  
- High price tag challenges value proposition compared to more affordable devices like Samsung's.  

AI has become the smartphone industry’s latest battleground, with manufacturers racing to add AI-powered features to attract mainstream consumers. Vertu is taking a different path: the UK-founded luxury phone maker, known for hand-finished devices often costing tens of thousands of dollars, has built its business selling status symbols to the ultra-wealthy rather than competing on specs. Its Alphafold targets affluent buyers, particularly chief executives, pairing luxury materials with an AI agent designed to automate parts of an executive’s working day.

I decided to test that claim on its own terms. Rather than focusing on benchmark scores, camera comparisons, and media consumption — the staples of most smartphone reviews — I spent a few days using the foldable the way Vertu says its customers would: managing documents, analyzing spreadsheets and contracts, planning business trips, automating routine tasks, and relying on its AI agent as a digital companion throughout the working day. The question wasn’t whether it was a good smartphone, but whether it was a good executive smartphone.

At the heart of the Alphafold is Hermes Agent, a pre-installed AI agent built on top of the open-source Hermes project, which the company says can analyze files, automate tasks across apps, remember conversations, and hand off requests to a human concierge when needed. Unlike most smartphone AI assistants that largely just respond to prompts, Hermes is designed to execute multi-step workflows on users’ behalf, making it the centerpiece of Vertu’s pitch rather than the foldable hardware itself.

Physically, the Alphafold, which starts at $6,880, looks and feels every bit like a luxury device. The review unit I received was wrapped in genuine calfskin leather with titanium accents, setting it apart from mainstream foldables that largely rely on glass or synthetic finishes. It’s clearly built for buyers who see their phone as both a tool and a status symbol.

Compared with the Samsung Galaxy Z Fold 7, which I used as a reference device throughout this review, the 264-gram Alphafold feels noticeably heavier than Samsung’s 215-gram foldable. The extra weight is apparent during prolonged use, though it never feels unwieldy. The Alphafold’s curved frame also makes it easier to unfold than the Galaxy Z Fold 7’s flatter edges. Samsung’s design, however, feels sleeker and more comfortable to hold when folded, making it easier to use one-handed.

Vertu’s Alphafold with a Calfskin Leather back and Samsung’s Galaxy Z Fold 7 with a Glass Back Image Credits:Jagmeet Singh / TechCrunch

The Alphafold also arrives in packaging that feels more akin to a jewelry presentation case than a smartphone box. The oversized box opens to reveal neatly arranged drawers containing bundled accessories, including a leather sleeve and charging cables, reinforcing the sense that Vertu is selling a luxury experience rather than just a handset.

Vertu Alphafold’s with a luxury packaging Image Credits:Jagmeet Singh / TechCrunch

Beneath the premium materials, however, the Alphafold tells a different story. During the review, I noticed striking similarities between the device and the $1,100 ZTE Nubia Fold — from the hinge design and dimensions to the placement of the speakers, microphones, and the fingerprint reader. The most visible distinction is Vertu’s leather-clad rear panel, though. System information also revealed ZTE identifiers in parts of the software.

When asked about these observations, Vertu confirmed to TechCrunch that the Alphafold was developed through a specialist supply-chain partnership involving ZTE/Nubia’s hardware platform, component integration, and production engineering. However, the company said it was responsible for the luxury materials, software experience, quality control, and after-sales service. ZTE did not respond to a request for comments.

ZTE Nubia Fold Image Credits:YMobile.jp

This isn’t new for Vertu. In a 2023 review of the MetaVertu, Wired reported that the device appeared to be based on a ZTE Nubia handset, citing hardware similarities and comments from Counterpoint Research that Vertu had been adapting existing ZTE models with luxury materials and custom software.

Still, focusing solely on the hardware misses the point of the Alphafold. Vertu’s real bet is not on building a better foldable but on whether executives will pay for an AI agent that helps them get through the working day more efficiently.

Over several days, I used the Alphafold as my primary smartphone, replacing routine prompts with real executive-style workflows. Instead of asking Hermes to write emails or answer trivia questions, I tasked it with analyzing spreadsheets and contracts, planning business trips, managing my schedule, and automating actions across multiple apps. I then compared the experience with Samsung’s Galaxy Z Fold 7 running Google’s Gemini.

Vertu Alphafold and Samsung Galaxy Z Fold 7 Image Credits:Jagmeet Singh / TechCrunch

The testing evolved as I went. Early software builds struggled to upload files, analyze images, and connect to Vertu’s concierge service. After I reported these issues to Vertu, the company rolled out server-side fixes that restored the missing functionality, allowing the remaining tests to be completed.

What emerged over days of testing was a more nuanced picture than the company’s claims might suggest. Hermes impressed when analyzing local files and spreadsheets, areas where Gemini on Samsung’s foldable still relied on manually uploaded documents during my testing. It was also more willing to automate actions across apps and complete multi-step workflows. But that greater autonomy came with trade-offs, raising questions about when an AI should act independently and when it should ask for clarification.

Can Vertu’s Hermes Agent replace an executive assistant?

One of the first tests simulated a common executive scenario before leaving for the airport. I asked Hermes Agent on the Alphafold to message a contact that I was running 20 minutes late, navigate to the airport, switch the phone to Do Not Disturb, and remind me to call the hotel in 15 minutes. The agent sent the message, enabled Do Not Disturb, and opened Google Maps with directions to the airport. It did not, however, automatically begin navigation and instead set the reminder for 9:08 p.m., despite the request being made at 2:32 a.m. for a reminder 15 minutes later.

Image Credits:Jagmeet Singh / TechCrunch

Running the same request on Samsung’s Galaxy Z Fold 7 produced a different experience. Rather than attempting every action immediately, Gemini asked follow-up questions, including which airport I wanted to travel to and whether the reminder should be created in Google Tasks or Samsung Reminder. Once I made those selections, it created the reminder for the correct time.

Hermes was more willing to act autonomously, while Gemini preferred to confirm details before proceeding. As a result, Hermes completed more of the requested workflow, but Gemini ultimately produced the more accurate outcome.

Planning a business trip

A second test focused on a more open-ended task. I asked Vertu’s Hermes Agent to organize a business trip from Mumbai to Pune, including a morning flight, a hotel recommendation, and adding the itinerary to my calendar. The agent responded that there were no direct morning flights available for the requested journey and offered a Contact Butler button to escalate the request to Vertu’s concierge service. It also created a calendar entry for the wrong dates, scheduling the trip for 7 July instead of 18–19 July, leaving the workflow incomplete.

Image Credits:Jagmeet Singh / TechCrunch

Gemini on Samsung’s Galaxy Z Fold 7 took a different approach. After determining that no suitable direct morning flight was available for the requested journey, it continued planning the trip by suggesting alternative travel options rather than handing the task off.

Working with business documents

Business documents revealed a mixed picture, too. I asked both Hermes Agent and Gemini to analyze a locally saved financial spreadsheet, summarize the quarterly results, and determine whether third-quarter sales figures were included.

During my original testing, Hermes analyzed an uploaded sales spreadsheet and correctly summarised the Q2 figures. However, when I returned to the same conversation days later, it no longer recognized the previously shared document, instead responding: “I cannot access files stored directly on your local device. Please upload or attach the Sales spreadsheet here in the chat, and I will gladly analyze the Q2 data for you.”

Image Credits:Jagmeet Singh / TechCrunch

Gemini also required the spreadsheet to be uploaded initially, but retained the context of the conversation. Days later, it was still able to answer follow-up questions about the document, correctly identifying the North region as generating the highest sales without requiring the file to be uploaded again.

Taken together, the testing suggested Hermes Agent is an ambitious AI assistant rather than a finished one. Its willingness to act autonomously often made it feel more like an agent than Gemini on Samsung’s phone, but that same approach occasionally produced incomplete workflows, incorrect outputs, and inconsistent behavior. The pace of updates during the review also suggested Vertu is actively refining the platform, meaning today’s experience may not be the same one buyers encounter a few months from now.

Beyond general assistance, Vertu has built Hermes around a collection of specialist AI agents aimed at affluent professionals, including agents focused on legal advice and investment insights, along with the option to escalate certain requests to a human concierge. The idea is to position the Alphafold as more than a premium smartphone, instead presenting it as a digital assistant for executives.

In practice, however, the specialist agents should be treated as starting points rather than authoritative advisers. They can provide useful summaries and recommendations, their responses remain AI-generated and should be independently verified before being relied upon for legal, financial, or other high-stakes decisions. The option to escalate certain requests to Vertu’s concierge service underscores the current limits of AI agents. Human expertise still matters.

Vertu is also positioning the Alphafold as a business platform rather than just a smartphone. The company demonstrated an integrated enterprise resource planning (ERP) system designed to give executives access to business data and workflows from the device. My testing, however, was limited to a demonstration environment, making it difficult to assess how the feature performs in day-to-day use or how well it integrates with existing enterprise systems.

The security aspect

For Alphafold’s target audience, security may matter as much as AI. Executives are unlikely to use an assistant that analyses contracts, financial reports, and business plans if they are uncertain where that data is processed or stored.

Vertu says conversations with Hermes Agent are encrypted and are not used to train public AI models. According to the company, users can also choose where their data is processed, with enterprise deployments supporting private infrastructure for organizations that require greater control over sensitive information.

Vertu backs those claims up with a dedicated “A5” security chip, which it says provides hardware-level protection for sensitive data, encrypted communications, and digital credentials. Those claims couldn’t be independently verified during testing, but they’re central to Vertu’s pitch to executives and enterprises.

Living with the Alphafold

Away from AI, the Alphafold behaves much like any modern flagship foldable. The battery comfortably lasted more than a day during testing. However, the absence of wireless charging is a surprising omission at this price, particularly when Samsung’s Galaxy Z Fold 7 supports convenient Qi charging alongside wired USB-C charging.

Vertu Alphafold lasts for over a day on a single charge Image Credits:Jagmeet Singh / TechCrunch

The camera app also includes a document scanning mode under a “Smart AI” setting that can recognize paperwork and save them with enhancements, making it useful for digitizing contracts, receipts, and other business documents. Samsung offers a comparable scanning experience through its own camera software, so this is feels more like a parity feature than a differentiator.

Verdict

The Alphafold is an ambitious attempt to build an AI-first luxury smartphone, but the execution falls short of its price tag. Despite its premium materials and exclusive services, the core hardware offers little that cannot be found in significantly less expensive foldables, while Hermes Agent remains an evolving platform rather than a compelling reason to spend thousands more.

Ultimately, Vertu is asking buyers to pay a substantial premium for branding, craftsmanship, and an ecosystem of AI and concierge services built on top of an established smartphone platform. Based on my testing, that premium is difficult to justify, particularly when Samsung’s Galaxy Z Fold 7 offers a more mature foldable experience with comparable day-to-day functionality at a fraction of the price. With Samsung’s next-generation Galaxy Z Fold 8 expected very soon, the Alphafold’s value proposition becomes even harder to defend.

Vertu wants executives to pay $6,880 for an AI agent — here’s how it actually performs

During a signing ceremony in the Oval Office, President Donald Trump formally acknowledged the evolving landscape of artificial intelligence by signing an executive order aimed at establishing tighter regulations on future AI developments. Accompanied by Senator Ted Cruz of Texas and Commerce Secretary Howard Lutnick, the President underscored his administration's intent to manage who can access cutting-edge AI technologies.

Traditionally, leading American tech firms, notably Anthropic and OpenAI, have had the discretion to determine which organizations could employ their most sophisticated AI models. Companies like Anthropic have strategically partnered with select entities to showcase their advanced cybersecurity offerings, such as the Mythos model associated with Project Glasswing. Meanwhile, OpenAI was prompted by the administration to restrict access to its latest GPT-5.6 model, utilizing a specialized group known as Daybreak for similar cybersecurity purposes.

A representative from the White House clarified to CNBC that the administration does not grant approvals for AI technologies released by private firms. The official emphasized that any collaborations or consultations with government experts are purely optional, stating that the timing and extent of AI model releases remain firmly within the purview of the companies themselves, echoed in Trump's recent executive directive.

In a further development, the administration recently halted the release of Anthropic's Claude Mythos 5 and Fable 5, citing "national security concerns." Access was restored after extensive discussions with Anthropic. In line with government requests, OpenAI announced its decision to restrict the rollout of new AI models to "trusted partners" only.

Sources report that the White House is controlling access to advanced AI models, transferring influence away from technology leaders.

A dark, stormy scene of the White House at dusk under heavy clouds, the US flag waving atop. From the portico, a giant presidential decree scroll unfurls across the front steps stamped “ACCESS RESTRICTED.” Behind it, a heavy, barred door labeled “Frontier AI Models” glows faintly, with the Google, Microsoft and Amazon logos clustered and dimmed behind the gate. Moody natural lighting highlights the tension and power shift.

The White House is dictating access to frontier AI models, shifting power from tech giants, sources say - CNBC

- The Trump administration is increasing control over future artificial intelligence model releases.
- AI companies like Anthropic and OpenAI previously determined access to their powerful models.
- Trump's executive order mandates tighter collaborations while emphasizing security without hindering innovation.

President Donald Trump points his finger as he signs an executive order on AI next to Senate Commerce Committee Chairman Ted Cruz, R-Texas, left, and Commerce Secretary Howard Lutnick, in the Oval Office at the White House in Washington, Dec. 11, 2025.

The Trump administration has taken new steps to assert more control over the rollout of future artificial intelligence model releases by dictating which companies and entities are allowed access to the latest frontier models, two people familiar with the matter told CNBC.

Until now, that decision was in the hands of American AI giants.

Both Anthropic and OpenAI have decided which companies and agencies have access to their most powerful models, and have often included major enterprise customers.

Anthropic unveiled its most capable Mythos cybersecurity model to a handful of partners with Project Glasswing. OpenAI was asked by the administration to gate its recent GPT-5.6 release, and has a similar consortium called Daybreak for its cybersecurity model.

A White House official told CNBC that it doesn't provide approvals for AI releases from private companies.

The official said any engagements, testing or meetings with government experts are "voluntary" and that "decisions on timing and scope of releases rest entirely with the companies," referring CNBC to Trump's recent executive order.

"The Administration continues to collaborate with all of America's frontier labs to strengthen the security of this technology without stifling innovation," they wrote.

However, last month the Trump administration blocked Claude Mythos 5 and Fable 5 due to "national security concerns," reinstating access after weeks of intense negotiations with Anthropic. OpenAI last month said it would limit new AI models to "trusted partners" to comply with government requests.

The White House is dictating access to frontier AI models, shifting power from tech giants, sources say

There is a notable range of opinions regarding how many months behind the U.S. China is in terms of artificial intelligence development, with estimates coalescing around a six-month gap, which some experts consider to be a modest lead.

Moonshot plans to unveil comprehensive details about its Kimi K3 model by the end of this month.

Samm Sacks, a senior fellow at the Johns Hopkins School of Advanced International Studies, attended an A.I. conference in Shanghai on Friday where President Xi spoke. She noted that many specialists believe it is only a matter of time before a Chinese firm releases an openly available A.I. model that can rival Anthropic’s cutting-edge systems. Such advancements could pose significant challenges for regulators seeking to manage the release of potent A.I. technologies.

Recently, another Chinese start-up, Z.ai, launched a model called GLM-5.2, which approaches the capabilities of Anthropic's leading Fable 5 model. Unlike the American counterpart, GLM-5.2 became readily accessible to developers.

The model quickly gained traction among software developers and start-ups in Silicon Valley due to its lower cost compared to dominant American alternatives. Its introduction coincided with U.S. companies' urgent need to reduce A.I. expenses.

Kimi K3 is reported to be even more advanced than GLM-5.2, based on evaluations conducted by Vals AI. This launch comes at a time when the Trump administration is contemplating potential regulations on A.I. technologies. In response to various cyber threats and the possibility of these technologies being misused to create biological weapons, many leaders in Silicon Valley have started advocating for such regulatory measures.

China's Moonshot AI Launches Kimi Model, Challenging America's Dominance

A dark-themed clandestine AI war room bathed in crimson spotlights, long shadows cutting across sleek obsidian consoles. A colossal wall display shows the bold red “Moonshot AI” logo with a Chinese flag waving behind it. In the center hovers a glossy, helmet-like “Kimi” core, its pulsing red circuits like tentacles. Through a distant viewport, the moon’s cratered surface glows faintly, and the silhouette of the U.S. Capitol is barely visible, dwarfed by the looming red threat.

China’s Latest A.I. Breakthrough Threatens America’s Lead - The New York Times

- Experts estimate China is about six months behind the U.S. in A.I. development.  
- Chinese start-up Z.ai's GLM-5.2 model rivals American systems and is widely accessible.  
- Kimi K3 is reportedly more powerful than GLM-5.2, raising regulatory concerns.  

“Everyone has a different number about how many months China is behind the U.S., and the estimates cluster around six months,” he said. “That is not much of a lead.”

Moonshot said it would release the full details of Kimi K3 at the end of the month.

Samm Sacks, a senior fellow at the Johns Hopkins School of Advanced International Studies who attended Friday’s A.I. conference in Shanghai where Mr. Xi spoke, said that many experts believe it’s only a matter of time before a Chinese company releases a freely available model that matches Anthropic’s most advanced systems. Such a breakthrough, Ms. Sacks said, would raise difficult questions for regulators trying to contain powerful A.I. technology once it has been released.

Last month, another Chinese start-up, Z.ai, released a model called GLM-5.2 that was nearly as powerful as Anthropic’s Fable 5 model, the leading American system. And unlike Fable, this Chinese model was widely available.

Many software developers and start-ups in Silicon Valley quickly adopted the model, largely because it was much cheaper than the leading American systems. It arrived just as U.S. businesses realized they had to find ways to cut down their A.I. spending.

Kimi K3 is even more powerful than GLM-5.2, according to benchmarks run by Vals AI. The model arrives as the Trump administration is considering regulating the technology. Many Silicon Valley executives have begun to support such regulation, saying that leading technologies can enable malicious cyberattacks and potentially help build biological weapons.

Unveiling Gemma 4 12B: a cohesive multimodal model without encoders.

More from AI

Prenuptial agreements are facing a new dilemma: Is it possible to be unfaithful using AI?

AI Startups, Independent Entrepreneurs, and Real Data

Your options for privacy

AI company launches mobile data center during Grand Sky event