❌

Normal view

This Deep Sea ‘Giant’ Can Go Five Years Without Eating—Now Scientists Are Unlocking the Genetic Secrets of Its Extreme Survival

6 June 2026 at 14:03


Scientists have uncovered the genetic secrets behind a novel energy-use strategy that allows a deep-sea “giant” to survive without food for over five years, in one of nature’s most stunning survival stories.

Supergiant bathynomids, also known as giant isopods, present a bizarre biological paradox: they exhibit pronounced body gigantism as megafaunal species yet inhabit a deep-sea habitat with extremely sparse nutrient availability.

In a recent paper in Cell, researchers from the Institute of Oceanology of the Chinese Academy of Sciences (IOCAS) revealed how these large deep-sea creatures subsist on such limited food, using a novel genetic strategy unique to bathynomids.

A Complex Genetic Approach

The IOCAS team employed a multi-omic strategy to address the problem, combining genomics with other disciplines, such as morphology, physiology, behaviorology, and metagenomics. Doing so led to a novel discovery: that supergiant bathynomids don’t have just one, but two techniques to manage their limited access to nutrients.

Bathynomus giganteus
One of the large species of Bathynomus seen in the Gulf of Mexico (Image Credit: NOAA)

First, isopods have a massive stomach that can store vast amounts of nutrients for later use. Second, they have an exceptionally low basal metabolic rate (BMR), which is genetically driven.

To diversify their research, the team focused on two species that occupy different depths. At around 300 meters, the Bathynomus doederleini is the shallower of the two, while the Bathynomus jamesi is found at 898 meters. Despite the difference in depth, both are technically considered “deep-sea” inhabitants, as the boundary is defined at 200 meters.

A Powerful Stomach

Roughly two-thirds of a bathynomid’s body is stomach—a figure far beyond that of related intertidal and shallow water species. The contents of those enormous stomachs are also unusual, filled with a mud-like mixture of highly digested food. This finely ground nutrient sludge has a low proportion of digestive bacteria, but a high amount of lipid-storing Chlamydiae. 

With a bacterial ratio leaning toward preservation rather than just digestion, this creates a food reserve that can be slowly digested and utilized as the creatures lower their BMR to endure long periods between feedings.

Genetic Findings

Another major finding was that bathynomids benefit not only from vertically transferred genes between parents and offspring, but from the horizontally transferred ND1 gene they acquired from an exogenous symbiotic bacterium. Intriguingly, the transfer defied typical horizontal transfer limitations, achieving an ultra-high level of expression in the bathynomid genome. 

These isopods use epigenetic histone modifications to regulate energy-related gene expression, thereby improving efficiency, conservation, and control. An example of this is histone acetylation regulating ND1 expression at an ultra-high level, which the team investigated in a laboratory test. Researchers suspected this gene was essential to energy metabolism, as it is a component of Complex I in the electron transport chain.

ND1 Genetic Tests

Using zebrafish, nematodes, and human 293T cells, the IOCAS team tested the gene’s impact on energy usage. At normal temperatures, the team found that ND1 actually decreased starvation tolerance by increasing energy metabolism. Yet when the team lowered temperatures to reflect those found in a typical deep sea environment, they found the opposite effect, with ND1 decreasing mitochondrial activity and energy metabolism rates, to improve zebrafish starvation tolerance an astounding 37%.

The IOCAS team concluded that ND1 provides a solution to the long-standing paradox of bathynomid gigantism in the context of their nutrient-poor environment. By fine-tuning the degree of metabolic depression, the gene can adjust the mitochondrial metabolic network to preserve energy during extended periods without sustenance.

Never before has such a strategy for energy efficiency, combining horizontal gene transfer and epigenetic optimization, been identified in a deep-sea megafaunal species.

“Our work not only deciphers the mystery of ultra-long starvation tolerance in deep-sea isopods,” said lead author Yuan Jianbo, “but also provides an important paradigm for understanding how life balances growth and survival in extreme environments.”

The paper, “Deep-Sea Megafauna Co-Opts Microbial Energy Metabolism Genes to Withstand Ultra-Long Starvation,” appeared in Cell on June 5, 2026.

Ryan Whalen covers science and technology for The Debrief. He holds an MA in History and a Master of Library and Information Science with a certificate in Data Science. He can be contacted at ryan@thedebrief.org, and follow him on Twitter @mdntwvlf.

“A Game Changing Warfighting Capability”: DIU-Funded Hermeus Says Autonomous, High-Mach Payload Release Test Flights to Begin

5 June 2026 at 16:36


Days after successfully flying its Quarterhorse unmanned aircraft at supersonic speeds, an industry-first defense aviation company, Hermeus, has received the largest-ever contract awarded by the Defense Innovation Unit (DIU) to demonstrate high-Mach flight combined with high-Mach payload release.

The new $ 159 million award, which increases the original contract to a total of $219 million, will allow Hermeus’ engineers to expand on previous high-Mach test flight milestones in direct partnership with the U.S. Air Force and the U.S. Navy, with the goal of providing the American warfighter with the most advanced arsenal capable of delivering munitions and payloads around the world at record-setting speeds.

“By delivering flight-ready aircraft and demonstrating payload release at speed, we will prove this technology can create a decisive military advantage on a timeline that matters,” explained Hermeus CEO and co-founder, AJ Piplica. “Our focus is on providing the Air Force and Navy with the validated data they need to transition these platforms into the future force.”

High Mach Payload Release: “A Game Changing Warfighting Capability”

When announcing the new DIU award, Hermeus executives noted that the program will include a series of “intensive flight tests.” The first is expected later this year, while the final flights are expected sometime in 2027.

Although few specifics were provided, the company said the flights will “prove the aircraft can maintain high speeds and expand its performance limits” while also demonstrating the successful carry and release of payloads at high velocity. The company also noted that all flight data gathered during the development and testing phases “will directly inform service experimentation and future acquisition decisions.”

high-Mach flight
In May, Hermeus’ Quarterhorse Mk 2.1 unmanned aircraft successfully completed its first supersonic flight, reaching a top speed of Mach 1.21. Image Credit: Hermeus.

Although the U.S. Military has several aircraft capable of exceeding the speed of sound, and several more hypersonic (exceeding five times the speed of sound) aircraft and missile prototypes, delivering payloads from high-Mach aircraft has remained a challenge. However, some military experts argue that the U.S. would benefit greatly from the capability.

When discussing the need for high-speed payload release, DIU Military Deputy Major General Joe “Solo” Kunkel said that the largest problems the military is facing in the future are “generally around time and distance,” and the associated challenges with getting to distant locations “in a timeframe that matters.”

“I care about the warfighting capability, providing advantage to our partners and us in the fight, and making sure that we can walk into every fight with swagger,” General Kunkel said, adding that mass producing the technology developed by Hermeus could move it from a test platform to a “game-changing warfighting capability.”

“I think we found a significant number of use cases where it can be used as a weapon,” the General added. 

Moving High-Mach Capability Out of the Lab

When discussing the company’s approach that convinced the DIU to offer its largest award in the organization’s history, including what they describe as “reclaiming the lost art of rapid iterative prototyping to build the fastest aircraft in the world today,” Hermeus noted that directly partnering with the USAF and USN allows their engineers to align tests with future mission needs and shorten the path from testing to deployment.

“By prioritizing relentless hardware iteration, we deliver high-speed systems at the pace of the modern battlefield,” the company’s statement explained. “We work with the Department of War to provide the high-speed capabilities our nation and its allies need to maintain a durable, asymmetric advantage.”

High-Mach flight
Image Credit: Hermeus.

When discussing the significance of the upcoming tests and the overall need for high-Mach flight payload deployment, including the benefit of the performance data, the company CEO heralded the DIU’s award, noting that the financing will help them move their successful demonstrations into real-world high-Mach platforms.

“This program is about moving high-Mach capability out of the lab and into an operationally relevant environment,” Piplica said. 

Christopher Plain is a Science Fiction and Fantasy novelist and Head Science Writer at The Debrief. Follow and connect with him on X, learn about his books at plainfiction.com, or email him directly at christopher@thedebrief.org.

Singtel advances drive for widespread internal AI use

5 June 2026 at 16:09

Singtel Group positioned a partnership with government-linked body Digital Industry Singapore as helping it accelerate a sweeping AI transformation of its business and strengthen digital infrastructure in the country.

The operator noted the pact, which comes with a grant for the company, will contribute to a multi-year plan to up adoption of AI across its business including within infrastructure, customer platforms and for workforce development.

In the first phase Singtel plans to develop and hire AI talent, establish governance and tech frameworks and build “enterprise advisory and implementation expertise to support AI deployment at scale”.

Among its aims is to become a “world-leading AI-native telco” with advanced AI and machine learning capabilities in all business units.

The operator highlighted its push aligned with a wider national strategy in Singapore which aims to strengthen sovereign digital infrastructure, increase skills of the local workforce and develop “globally competitive leaders” in the AI sector.

Singtel CEO Yuen Kuan Moon added it was “championing enterprise transformation with AI” in its home market.

“As an adopter, provider and enabler of AI, Singtel is embedding intelligent technologies across our operations, delivering trusted enterprise-grade solutions that help customers move from pilot to scale, and providing secure, sovereign platforms and digital infrastructure to support adoption across industries.”

He continued: “Ultimately, our goal is to create real impact not just for our business, but for Singapore, future-proofing our industries to be more competitive and resilient in the AI era.”

The post Singtel advances drive for widespread internal AI use appeared first on Mobile World Live.

Why Apple may be winning again

5 June 2026 at 14:14

As we lean into WWDC, three strategically brilliant Apple moves have been exposed in the last couple of weeks, two of which will have immense consequences in the coming year, while one sets the scene for essential future growth.

In each case, Apple’s leadership has found counter-intuitive gambits that actually secure the company’s future. Let’s start with Vision Pro.

It came from the future

News this week is that incoming CEO John Ternus has made some tough decisions around Apple’s approach to spatial computing, terminating development of Vision Pro (even as leaked images of a black model emerge) while focusing R&D on two smart glasses projects to compete with Meta. 

The intention is to introduce XR and AR glasses priced at around $300 to $500 each. While not as richly-featured as the Vision Pro, they will be within the reach of more people and draw deeply on the huge R&D effort that went into the original Apple AR visors. Apple hopes a focus on trust and privacy will be enough to push Meta aside, helping Cupertino dominate this part of the category. If that plan succeeds, don’t be at all surprised to see plans for Vision Pro 2 return to the table, though that’s not the focus now.

Apple frequently described the Vision Pro as a product “pulled from the future,” a device for enterprise users and early adopters. For many such users, the existing product will be useful in their work for time to come. 

What’s strategically solid about this is that Apple has now defined a good future for spatial computing and is bringing components of that future to the mass market on the basis of a provable technology you can already try for yourself in any Apple Store. This is a long game, and while it will take time to play out, it’s a game the company has proved it can join.

Privacy as a standard

Apple has apparently chosen to use Nvidia and Google technologies to support at least some of the Apple Intelligence/Siri improvements to be announced at WWDC next week. This seems to fly against the company’s general approach to privacy on its platforms, with the silicon, operating system and — thanks to Private Cloud Compute — the cloud all in its control.

How, you might ask, can Apple ensure privacy when using third-party infrastructure to manage some AI transactions? How can it do so without damaging its trusted brand?

One direction that makes sense is to consider that Apple and its partners have reached consensus on what privacy should be and how it should be delivered. That’s a very important consensus, as it suggests Apple is building an approach that makes privacy an attestable standard.

The company has been pushing governments for years to agree to such standards, but all it seems to have had in return are continued government attempts to erode personal privacy. That’s particularly evident in the UK government’s egregious move to undermine encryption to the detriment of all. (The UK isn’t really alone in preferring surveillance above liberty.)

Android developer Google has had the same experience, and while its approach to privacy differs from Apple, both companies understand the need for encryption. As such, any form of consensus on some form of privacy standards is welcome — and while I’d very much prefer an enforceable, verifiable approach, some industry agreement has to be better than nothing at all. 

While I don’t believe Apple’s approach to privacy in the new breed Siri/Apple Intelligence will be introduced in this way at WWDC, it will be interesting to see what does emerge from the new tech triptych (Apple, Google, Nvidia) in the coming months. Certainly, all three have a great interest in guarding encryption against Quantum attacks for which hidden backdoors would be easy pickings.

Winning the PC War

Apple has won the PC war.

That last point needs explanation: Apple’s entry-level MacBook Neo also has just 8GB RAM, but it also has custom-designed Apple processors and an operating system optimized to run on the hardware. That means those Macs use memory far more efficiently than their competitors. 

So, you can pick up some mass market Windows laptops for $800 that hold just 8GB memory, or spend $699 for a MacBook Neo with double amount and that can also run Windows in VM extremely well. 

Component prices are not going to shrink back for a while, any more than further magical thinking is going to end the war in Iran. At this point in the cycle, Apple has the PC market advantage. Millions are purchasing its entry-level Mac and the vast majority of those new users will love the platform, as new users usually do. That’s going to lead to a spike in Apple services sign-ups, and prompt solid future upgrade and accessory sales cycles.

Apple accomplished this by selling a low-cost Mac at a time when competitors face existential problems maintaining their grip on the mid-range market. The longer Apple holds prices on the device, the greater the advantage it builds, while applying huge pressure on PC competitors.

Summing up the goods

With the Mac hitting its iPod moment as it achieves mass market sales, Apple finally seeing something like progress in its attempt to secure privacy in a digital age, and a strong position from which to grow in the wearables market, the company has played a fine hand. 

That’s even before it introduces us to its improved Apple Intelligence, and an era of AI access in which many everyday tasks take place token free directly on the device. Indeed, when it comes to AI, if it gets things right at WWDC, Apple appears to be making money, while AI competitors are bleeding financial oxygen as their inflated bubble heads to its inevitable demise. What about the enterprise? Take a look at this chart.

You can follow me on social media! Join me on BlueSky,  LinkedIn, Mastodon, and follow The Core.

Jason Momoa’s 1920s Harleys become electric hybrids in wildest EV conversions yet

5 June 2026 at 12:49

UK-based EV conversion specialist Electrogenic has converted five of Jason Momoa’s cherished vintage vehicles to electric power — including three 100-year-old Harley-Davidson motorcycles reimagined as plug-in hybrids. The builds are featured in Season 2 of Momoa’s HBO documentary series On The Roam, streaming now.

The project includes two extremely rare classic Land Rovers converted to all-electric drive with 150+ miles of range, a battery-equipped off-grid camping trailer, and the three Harleys that can switch between electric, petrol, and combined hybrid power on the fly. Electrogenic is now selling conversion kits for both the Land Rovers and the Harleys.

The Friday File: Anthropic; EU; FCC

5 June 2026 at 11:57

Mobile World Live brings you our top three picks of the week as Anthropic widened access to its Claude Mythos model despite security concerns, the European Commission (EC) unveiled a fresh digital sovereignty push and the FCC commenced its first spectrum auction in four years.

Anthropic expands Mythos access to 150 new companies

What happened: Anthropic expanded access to its controversial Claude Mythos AI model under the Project Glasswing to 150 additional companies in sectors including power, healthcare and communications, after initially restricting it to a group of private technology players.

Why it matters: Anthropic said the latest cohort brings in sectors underrepresented in the first wave. In commentary to security publication CSO Online, experts noted the expansion could add to security concerns around the model. Carmi Levy, an independent technology analyst, questioned what Glasswing will be able to accomplish by adding 150 more participants, noting the initial point was to allow the AI player to work closely with a small, fully vetted group of vendors to develop stronger defences against cybersecurity risks. “Expanding access into the hundreds may very well bring in more minds to build better defensive measures, but it simultaneously introduces significant concerns around potential leaks.”

Research director for AI security at IDC, Grace Trinidad, added that Anthropic’s announcement pointed out that each of the 150 new participants “will need to meet our security requirements before they gain access”, which also did not build confidence. “Nobody knows what those security requirements are.”

Earlier this week, Anthropic also confidentially filed IPO paperwork with the US Securities and Exchange Commission (SEC), ahead of rival OpenAI’s rumoured float.

EU targets AI, chips in fresh sovereignty drive

What happened: The EC unveiled a fresh digital sovereignty package targeting semiconductors, AI, cloud, open source and energy infrastructure in a bid to accelerate Europe’s push for digital sovereignty.

Why it matters: The package includes a proposed a revamped Chips Act 2.0 and a Cloud and AI Development Act to streamline data centre deployment and introduced measures to expand open source use, support startups and digitalise the energy system. EC president Ursula von der Leyen said Europe “cannot afford to depend on others for the technologies that keep our hospitals running, our energy grids stable and our services secure”.

Jennifer Okafor, a UN and Global Health strategist and AI and data analyst, said the policies represent “a comprehensive approach to balancing growth, stability, and long-term resilience across the EU”. However, president of German digital industry group Bitkom Ralf Wintergerst told Reuters it is “crucial that these efforts do not stop at mere announcements”, while Keegan McBride, director of science and technology at non-profit think tank Tony Blair Institute argued Europe “can’t regulate its way to competitiveness, it must build”. He added, “there’s still much more to do ​if Europe wants to close the gap with the US and China”.

FCC kicks off first spectrum auction in 4 years

What happened: The FCC opened Auction 113, its first spectrum auction in four years, covering spectrum in the 1695MHz to 1710MHz, 1755MHz to 1780MHz and 2155MHz to 2180MHz bands.

Why it matters: AT&T, T-Mobile US, Verizon and potentially SpaceX are among likely bidders. The licences cover territory home to more than 100 million people across 48 states and two US territories. FCC chair Brendan Carr declared: “Finally! The FCC is back in the game,” adding spectrum auctions are “the lifeblood of licensed wireless service”. Carr argued “more spectrum means more building, lower prices and stronger competition”.

Proceeds from the auction will fund the FCC’s “rip and replace” programme targeting Huawei and ZTE equipment in US networks. Indeed, the auction also bolsters the regulator’s broader Build America Agenda, which targets 800MHz of spectrum by 2034.

The post The Friday File: Anthropic; EU; FCC appeared first on Mobile World Live.

“Temos a solução para as empresas criarem as infraestruturas para a nova onda de Inteligência Artificial”

5 June 2026 at 11:21

Nate Snow, diretor da unidade de negócio da Asus, diz que empresas procuram cada vez mais soluções combinadas entre hardware e software que permitam implementar inteligência artificial de forma prática e segura, reduzindo custos de utilização de tokens e com facilidade para escalar o negócio.

The post “Temos a solução para as empresas criarem as infraestruturas para a nova onda de Inteligência Artificial” appeared first on Tek Notícias.

What Safari reveals about Apple’s AI strategy ahead of WWDC

4 June 2026 at 17:27

Apple’s latest Safari privacy campaign is more than pre-WWDC marketing. It is an early signal of how the company plans to frame artificial intelligence (AI): as something that only works if users trust the platform behind it.

The week before WWDC is often significant, as Apple tends to make announcements it simply can’t fit into the keynote itself. This year’s first pre-show reveal is a new campaign focused on privacy that shows how much more private Safari is than rival browsers; there’s even a highly entertaining video that makes the point.

Privacy on Safari

Apple has been building privacy protections into Safari for years. The browser protects you from malicious scripts that might attempt to access passwords or credit card information. Safari also tells you what data an extension wants to access and can restrict access to match your settings. It blocks third-party cookies by default, detects and removes trackers, and has measures in place to prevent data companies from identifying — and following — you through device characteristics. 

That’s even before Apple’s powerful Private Browsing mode, which includes meaningful protections. The company has put together a page packed with resources to explain the privacy protections it has in place across its platforms.

Privacy is critical to Apple — not only because the company regards it as a human right, but because it correctly recognizes that to make new generations of sensor-laden technologies it must ensure privacy is protected. Without privacy and trust, people won’t use the technology.

Trust is the product, not you

The truth is that people are becoming increasingly concerned about how the digital devices we depend on for convenience are now being used for different kinds of surveillance, and we need to be convinced that our personal data is protected. We do not want every aspect of our life to become fodder to feed a digital dystopia, even as we still want the positive solutions technology promises.

Think about the Apple Watch. Consider the data it gathers: distance walked, calories burned, and more — it’s a rich trove of personally identifiable data that no one really wants to share with others without consent. Apple Watch is not the only Apple device that is gathering information, even your web browser captures a great deal of it. Hence, the focus on Safari in Apple’s new campaign.

Privacy will become an even greater concern as AI spreads. Data brokering services already make extensive use of AI to analyze and identify patterns in the online data they harvest. AI deployed without strong privacy protections poses serious risks to the way we live, while the consolidation of AI ownership in the hands of a few companies risks creating dangerous imbalances of power. That’s the context in which private data needs to be protected, making privacy an essential component of a positive tech-augmented future. 

Why the AI era raises the stakes

Apple’s focus on privacy is far from new; it has been consistent in this work for many years. Competitors often accuse Apple of hypocrisy, but the company has been arguing for privacy’s importance for more than a decade. Others have adopted some of the same principles, though not all of them — and while Apple may sometimes use privacy as a moat for its own products and services, that does not diminish its value.

It’s with all this in mind that I consider Apple’s latest privacy ad campaign and its rollout just before WWDC, where it is expected to introduce new AI services. That Apple’s new privacy campaign seems not to have made the final cut for the show tells me the company has much more to discuss on the topic, particularly around Apple Intelligence.

What Safari’s signals suggest

When Apple introduces its new AI features at WWDC it will do so while celebrating the privacy built into them. The current privacy ad campaign will be part of an overall push as the company explains that its ecosystem can run third-party AI services while also offering its own bespoke Apple Intelligence AI to do really useful things in complete privacy.

This isn’t just a competitive moat, it’s a realistic assessment in practice. It shows that Apple understands that in the age of AI, privacy matters more than ever. As AI becomes central to everyday digital experiences, privacy is no longer optional — and Apple is prepared to make the case to support it.

You can follow me on social media! Join me on BlueSky,  LinkedIn, Mastodon, and follow The Core.

SoftBank’s PayPay set for life insurance buy

4 June 2026 at 15:26

SoftBank Group’s mobile payment subsidiary PayPay inked a deal to acquire a 70% stake in T&D Financial Life Insurance Company for JPY134 billion ($840 million), part of an aim to broaden its range of financial services.

When launched in 2018 PayPay focused on providing contactless mobile payments but has since widened this to include credit cards and other banking services. It had a base of more than 74 million users as of May 2026.

Announcing the deal, PayPay said the move to buy the life insurance company from T&D Holdings was part of an ambition to “provide comprehensive financial services tailored to each stage of users’ lives”.

The company intends to grow T&D Financial Life’s business through use of its existing platforms and by creating “new customer experiences in the digital life insurance domain”.  

In a linked deal, asset management company One Investment Management intends to buy a stake of almost 15% in the insurance company from T&D Holdings. The current owner will retain the other 15%, though a call option on this is being inserted into the acquisition agreement.

The deal is subject to regulatory approval.

Alongside the proposed acquisition, PayPay agreed a “comprehensive business alliance” agreement 3with T&D Holdings, largely concerning the latter’s insurance brand Taiyo Life.

The pair will “explore the development and provision of services designed to enhance customer convenience and help address social issues, while also drawing on AI, digital technologies, and other capabilities of SoftBank Corp”.

The pact is set to include: offering T&D Holdings’ Taiyo Life insurance products on the PayPay app; exploring call centre enhancement and operational efficiency for Taiyo Life; and jointly exploring a “smart senior city concept”.

The post SoftBank’s PayPay set for life insurance buy appeared first on Mobile World Live.

T-Mobile US expands globally with India tech hub

4 June 2026 at 14:57

T-Mobile US officially opened a global capability centre (GCC) in Hyderabad, India, with plans to hire around 1,000 staff by 2027, its first such facility outside of its home market.

The Indian government stated T-Mobile, through its TMUS Global Solutions Technology subsidiary, opened a site spanning 250,000 square feet in the city, which is situated in the state of Telangana.

It will operate as a strategic innovation hub within its global network, focusing on software engineering, DevOps, product development, cloud technologies, AI, data analytics, cybersecurity and next-generation digital solutions.

Posting on X, minister for IT Sridhar Duddilla said T-Mobile’s GCC represented another significant milestone in Telangana’s growth as a technology and innovation destination.

“The decision by T-Mobile to expand its presence here reflects the confidence that global companies have in Telangana’s talent, business-friendly environment, and strong digital infrastructure.”

Chandra Gupta, VP IT operations at TMUS Global Solutions, added the company decided to locate the facility in Hyderabad as it offers “a combination of technology talent and an established innovation ecosystem aligned with the company’s long-term goals”.

The Economic Times of India reported the company has already onboarded more than 500 people at the facility,

According to Reuters, India’s GCCs have evolved from low-cost outsourcing hubs to offices for global companies, supporting parent companies in several functions.

The post T-Mobile US expands globally with India tech hub appeared first on Mobile World Live.

Google brings local AI agents to laptops with Gemma 4 12B

4 June 2026 at 11:06

Google has released new tools that allow developers to run agentic AI workflows locally using Gemma 4 12B, a 12-billion-parameter model from Google DeepMind.

In a blog post, the company said the model, combined with the Google AI Edge stack, can be used to build and test applications on everyday machines. The model-runtime combination supports capabilities such as autonomous data processing, visual insight generation, webpage creation, and tool use.

The release includes Google AI Edge Gallery for macOS, where developers can use Gemma 4 12B to generate and run scripts for tasks such as data analysis. Google also said its Eloquent voice dictation and editing app now runs fully on-device on macOS, with support for local transcription and voice-driven text editing.

Google has also expanded LiteRT-LM, its lightweight command-line tool for running language models locally, with a new serve command. The company said this allows the CLI to act as a local LLM server and lets developers connect Gemma 4 12B to standard tools, SDKs, and frameworks through a local endpoint.

“Your data stays on your device while maintaining reliable responsiveness, utility, and cost efficiency,” the company said in the blog post.

The announcement comes as enterprises are looking beyond large, general-purpose models for some AI workloads. Gartner predicted that by 2027, organizations will use small, task-specific AI models at least three times more than general-purpose large language models, citing demand for more contextualized and cost-effective AI systems.

Challenges to overcome

But running agents on employee devices brings a number of problems. Companies must work within the limits of endpoint hardware, which can restrict the size of models that run effectively and the number of model instances that can operate at one time.

“While the AI can now fit on a laptop, enterprise IT infrastructure is largely unprepared to manage it,” said Rishi Padhi, principal analyst at Gartner. “Even highly optimized models like the Gemma 4 12B require around 16GB of unified memory or VRAM to run alongside standard applications. Many standard-issue enterprise laptops lack the memory bandwidth and NPUs/GPUs required for fluid, multi-turn agentic execution.”

Anand Joshi, AI analyst at TechInsights, said local deployment also changes the nature of the workloads. On a PC, search may mean finding information across internal folders and files. In a data center, the same function could involve searching the internet or querying a large database such as SQL.

“The framework for local deployment of agentic AI is different from that of a data center,” Joshi said. “The models are smaller; you can run only one instance of a large model at a time. You are limited by memory, CPU, and so on.”

Security and governance are also likely to become bigger concerns as AI agents move closer to enterprise endpoints. Agentic AI is designed to take actions, creating new security risks when local models are given access to employee files or allowed to interact directly with applications and scripts.

“Sandboxing these agents without breaking their utility is still a major operational challenge,” Padhi added. “And all this while enterprises need to audit AI usage for compliance and security. When inference happens entirely offline, capturing logs, tracking model drift, and ensuring employees are using the approved, compliant ways for a model becomes incredibly difficult.”

The cost tradeoff

Running AI agents locally could reduce some cloud inference costs, but the savings may be offset in the near term by higher spending on endpoint hardware and management.

“First and foremost, it is an OpEx-to-CapEx shift, as it shifts that financial burden by forcing accelerated hardware refresh cycles for premium PCs or edge devices,” Padhi said. “It would require buying expensive, high-memory laptops for employees at a time when memflation in the hardware industry is already driving up end-user average selling prices for laptops.”

Many enterprises refreshed PCs in 2025 to support Windows 11, but at that point, most AI inference still ran in the cloud, and the case for on-device AI remained unclear, Padhi said.

Enterprises may therefore move cautiously, buying AI-capable PCs only where local inference has a clear business case.

Over time, however, on-device AI could make enterprise AI spending more predictable by reducing exposure to variable cloud inference bills. The tradeoff is that companies may face a higher baseline cost for equipping and managing employees’ devices.

Complementing cloud AI

For enterprises, local AI is unlikely to replace cloud-based AI outright. Analysts said local AI is more likely to be used for workloads that benefit from endpoint processing, especially when applications must operate offline or when privacy and response times are critical.

“For local agentic AI to proliferate, the use cases on edge will have to complement data center/cloud use cases,” Joshi said. “I don’t expect local agentic AI to replace cloud AI, but it has potential to take a slice away from the cloud, and models like Gemma are significant steps towards enabling that.”

The market, Joshi added, is still determining where local AI fits best. “I estimate that use cases that require privacy or have strict latency needs will move to local node first, with further migration of others in the next 2-3 years,” he said.

Padhi said model placement will depend on the privacy requirements of a workload, the computing power it needs, and where the relevant data resides. Tasks such as code generation or analysis of local files could increasingly run on employee devices, while enterprise-wide RAG systems and more complex AI workflows are likely to remain cloud-based.

The article originally appeared on InfoWorld.

❌