By Dr. Tim Sandle
SCIENCE EDITOR
The aircraft carrier USS Gerald R Ford will be part of Washington's expanded military presence - Copyright AFP Jaime REINA
The U.S. government is pumping more money into AI research, especially in relation to defence, making AI central to the future strategy. U.S. Department of War (DoW), including the Department of Defense (DoD), is actively developing, testing, and fielding artificial intelligence (AI) capabilities at the edge, meaning the technology runs directly on local devices and in operational environments. The goal is to achieve a decision advantage over adversaries by processing information with greater speed and accuracy on the battlefield.
To help to understand these development, Digital Journal caught up with Sek Chai, CTO, Latent AI.
Digital Journal: With major investments in physical AI (or edge AI) infrastructure from NVIDIA, AMD, and Qualcomm, and others, what overall market evolution do you predict in the edge AI ecosystem by 2026? Which developments do you believe will have the greatest impact on real-world deployment?
Sek Chai: Investments in physical or edge AI are dwarfed by investments in hyperscalar data centres. Smart investors are now realising that $1T expenditures for massive data centres will not bring in immediate returns, if at all. Instead, investments in Edge AI are actually much more logical and less risky. New developments to enable the reliability and robustness of Edge AI will bring Edge AI to the forefront with real-world deployment.
Such a fundamental shift to Edge First approach will bring about standardisation, interoperability, and security to the Edge AI market. These issues have been key elements that are addressed with Latent AI’s product offering.
DJ: From your vantage point, is the software stack (inference runtimes, model compression tools, deployment tools, and secure update pipelines) finally maturing fast enough to enable developers to utilise new hardware in production by 2026 fully? What gaps remain?
Chai: The tools and software stack are maturing, but not fast enough. Especially in edge hardware, it is still a wild-west ecosystem with heterogeneous solutions that are not interoperable. This is where Latent AI shines, by offering a standardized AI runtime with services layers, much akin to how Java offers the necessary abstractions for software.
DJ: What structural bottlenecks inside organizations (data ownership, integrator lock-in, lack of model governance) will most limit edge AI adoption in 2026? How can these be overcome?
Chai: Enterprises fear the unknown and the uncertainty. Thus, they are not readily adopting an Edge First approach because the cloud still offers a platform that is familiar. This bottleneck is now being overcome when they realize the economic, logistical, and sustainability issues to rely solely on the cloud.
DJ: With hardware becoming more broadly available, do you expect 2026 to be the year when Department of War (DoW) can overcome integration, interoperability, and accreditation challenges in edge AI software platforms and finally move the current wave of Edge AI pilots to fielded capabilities?
Chai: DoW are already fielding AI capabilities on the edge. However, these systems cost hundreds of millions of dollars and with many years of development. Furthermore, there’s vendor-lock from integrators that sell entire platforms, from algorithms to hardware. DoW seeks to procure the best of breeds of AI algorithms, without being locked-in to any vendor. Edge AI will scale in deployment as DoW elevates the urgency with an alternative procurement strategy with interoperability in mind.
DJ: Which autonomy initiatives today are showing the clearest pathway from experimentation to fielding, and what inflection points do you expect in 2026 as these programs scale?
Chai: The battlefield is changing and there is an urgent need to build solutions that can adapt to new operational environments. These are evident in the new type of warfare currently in Ukraine, where adaptation is key to mission success. Adversaries now adjust signatures, tactics, and decoys on commercial timelines, not acquisition timelines, so static edge AI models that rely on long, centralized retraining cycles fall behind almost as soon as they are deployed.
The SecWar and DoW’s emphasis on speed, adaptability, and AI-enabled decision dominance highlights a clear pathway where adaptive AI is part of the fielding requirements.
SCIENCE EDITOR
DIGITAL JOURNAL
January 1, 2026
The aircraft carrier USS Gerald R Ford will be part of Washington's expanded military presence - Copyright AFP Jaime REINA
The U.S. government is pumping more money into AI research, especially in relation to defence, making AI central to the future strategy. U.S. Department of War (DoW), including the Department of Defense (DoD), is actively developing, testing, and fielding artificial intelligence (AI) capabilities at the edge, meaning the technology runs directly on local devices and in operational environments. The goal is to achieve a decision advantage over adversaries by processing information with greater speed and accuracy on the battlefield.
To help to understand these development, Digital Journal caught up with Sek Chai, CTO, Latent AI.
Digital Journal: With major investments in physical AI (or edge AI) infrastructure from NVIDIA, AMD, and Qualcomm, and others, what overall market evolution do you predict in the edge AI ecosystem by 2026? Which developments do you believe will have the greatest impact on real-world deployment?
Sek Chai: Investments in physical or edge AI are dwarfed by investments in hyperscalar data centres. Smart investors are now realising that $1T expenditures for massive data centres will not bring in immediate returns, if at all. Instead, investments in Edge AI are actually much more logical and less risky. New developments to enable the reliability and robustness of Edge AI will bring Edge AI to the forefront with real-world deployment.
Such a fundamental shift to Edge First approach will bring about standardisation, interoperability, and security to the Edge AI market. These issues have been key elements that are addressed with Latent AI’s product offering.
DJ: From your vantage point, is the software stack (inference runtimes, model compression tools, deployment tools, and secure update pipelines) finally maturing fast enough to enable developers to utilise new hardware in production by 2026 fully? What gaps remain?
Chai: The tools and software stack are maturing, but not fast enough. Especially in edge hardware, it is still a wild-west ecosystem with heterogeneous solutions that are not interoperable. This is where Latent AI shines, by offering a standardized AI runtime with services layers, much akin to how Java offers the necessary abstractions for software.
DJ: What structural bottlenecks inside organizations (data ownership, integrator lock-in, lack of model governance) will most limit edge AI adoption in 2026? How can these be overcome?
Chai: Enterprises fear the unknown and the uncertainty. Thus, they are not readily adopting an Edge First approach because the cloud still offers a platform that is familiar. This bottleneck is now being overcome when they realize the economic, logistical, and sustainability issues to rely solely on the cloud.
DJ: With hardware becoming more broadly available, do you expect 2026 to be the year when Department of War (DoW) can overcome integration, interoperability, and accreditation challenges in edge AI software platforms and finally move the current wave of Edge AI pilots to fielded capabilities?
Chai: DoW are already fielding AI capabilities on the edge. However, these systems cost hundreds of millions of dollars and with many years of development. Furthermore, there’s vendor-lock from integrators that sell entire platforms, from algorithms to hardware. DoW seeks to procure the best of breeds of AI algorithms, without being locked-in to any vendor. Edge AI will scale in deployment as DoW elevates the urgency with an alternative procurement strategy with interoperability in mind.
DJ: Which autonomy initiatives today are showing the clearest pathway from experimentation to fielding, and what inflection points do you expect in 2026 as these programs scale?
Chai: The battlefield is changing and there is an urgent need to build solutions that can adapt to new operational environments. These are evident in the new type of warfare currently in Ukraine, where adaptation is key to mission success. Adversaries now adjust signatures, tactics, and decoys on commercial timelines, not acquisition timelines, so static edge AI models that rely on long, centralized retraining cycles fall behind almost as soon as they are deployed.
The SecWar and DoW’s emphasis on speed, adaptability, and AI-enabled decision dominance highlights a clear pathway where adaptive AI is part of the fielding requirements.
Strange magnetism could power tomorrow’s AI
By Dr. Tim Sandle
SCIENCE EDITOR
In May, the most powerful geomagnetic storm to strike Earth in more than two decades lit up night skies in many parts of the world - Copyright AFP/File Sanka Vidanagama
Scientists from National Institute for Materials Science, Japan have confirmed that ultra-thin films of ruthenium dioxide (RuO2) belong to a newly recognised and powerful class of magnetic materials called altermagnets. These materials combine the best of two magnetic worlds: they’re stable against interference yet still allow fast, electrical readout—an ideal mix for future memory technology.
Altermagnetism is a newly discovered type of magnetism where magnetic moments (tiny magnetic fields created by electrons) align in opposite directions but follow a distinct rotated pattern.
The researchers also found that the performance of RuO2 thin films can be improved by carefully controlling how their crystal structure is oriented during fabrication.
Ruthenium dioxide has long been considered a promising candidate for altermagnetism
Why altermagnetism matters
Standard ferromagnetic materials used in memory devices allow data to be written easily using external magnetic fields. However, they are vulnerable to interference from stray magnetic fields, which can cause errors and limit how densely information can be stored.
Antiferromagnetic materials offer much better resistance to external magnetic disturbances. The challenge is that their internal magnetic spins cancel each other out, making it difficult to read stored information using electrical signals.
As a result, scientists have been searching for materials that combine magnetic stability with electrical readability and, ideally, the ability to be rewritten.
Slow progress
While altermagnets promise this balance, experimental results for RuO2 have varied widely around the world. Progress has also been slowed by the difficulty of producing high-quality thin films with a consistent crystallographic orientation.
The scientists overcame these obstacles by successfully creating RuO2 thin films with a single crystallographic orientation on sapphire substrates. By carefully choosing the substrate and fine-tuning the growth conditions, they were able to control how the crystal structure formed.
Using X-ray magnetic linear dichroism, the researchers mapped the spin arrangement and magnetic order in the films, confirming that the overall magnetization (N-S poles) cancels out. The tea also detected spin-split magnetoresistance, meaning the electrical resistance changes depending on the spin direction. This effect provided electrical evidence of a spin-split electronic structure.
The experimental results matched first-principles calculations of magneto-crystalline anisotropy, confirming that the RuO2 thin films truly exhibit altermagnetism.
These findings strongly support the potential of RuO2 thin films for next-generation high-speed, high-density magnetic memory devices.
Next steps
The scientists plan to develop advanced magnetic memory technologies based on RuO2 thin films. These devices could support faster and more energy-efficient information processing by taking advantage of the natural speed and density offered by altermagnetic materials.
The synchrotron-based magnetic analysis methods established during the study are also expected to help researchers identify and study other altermagnetic materials. This approach could accelerate progress in spintronics and open new pathways for future electronic devices.
The research appears in the journal Nature Communications, titled “Evidence for single variant in altermagnetic RuO2(101) thin films.”
By Dr. Tim Sandle
SCIENCE EDITOR
DIGITAL JOURNAL
January 1, 2026
In May, the most powerful geomagnetic storm to strike Earth in more than two decades lit up night skies in many parts of the world - Copyright AFP/File Sanka Vidanagama
Scientists from National Institute for Materials Science, Japan have confirmed that ultra-thin films of ruthenium dioxide (RuO2) belong to a newly recognised and powerful class of magnetic materials called altermagnets. These materials combine the best of two magnetic worlds: they’re stable against interference yet still allow fast, electrical readout—an ideal mix for future memory technology.
Altermagnetism is a newly discovered type of magnetism where magnetic moments (tiny magnetic fields created by electrons) align in opposite directions but follow a distinct rotated pattern.
The researchers also found that the performance of RuO2 thin films can be improved by carefully controlling how their crystal structure is oriented during fabrication.
Ruthenium dioxide has long been considered a promising candidate for altermagnetism
Why altermagnetism matters
Standard ferromagnetic materials used in memory devices allow data to be written easily using external magnetic fields. However, they are vulnerable to interference from stray magnetic fields, which can cause errors and limit how densely information can be stored.
Antiferromagnetic materials offer much better resistance to external magnetic disturbances. The challenge is that their internal magnetic spins cancel each other out, making it difficult to read stored information using electrical signals.
As a result, scientists have been searching for materials that combine magnetic stability with electrical readability and, ideally, the ability to be rewritten.
Slow progress
While altermagnets promise this balance, experimental results for RuO2 have varied widely around the world. Progress has also been slowed by the difficulty of producing high-quality thin films with a consistent crystallographic orientation.
The scientists overcame these obstacles by successfully creating RuO2 thin films with a single crystallographic orientation on sapphire substrates. By carefully choosing the substrate and fine-tuning the growth conditions, they were able to control how the crystal structure formed.
Using X-ray magnetic linear dichroism, the researchers mapped the spin arrangement and magnetic order in the films, confirming that the overall magnetization (N-S poles) cancels out. The tea also detected spin-split magnetoresistance, meaning the electrical resistance changes depending on the spin direction. This effect provided electrical evidence of a spin-split electronic structure.
The experimental results matched first-principles calculations of magneto-crystalline anisotropy, confirming that the RuO2 thin films truly exhibit altermagnetism.
These findings strongly support the potential of RuO2 thin films for next-generation high-speed, high-density magnetic memory devices.
Next steps
The scientists plan to develop advanced magnetic memory technologies based on RuO2 thin films. These devices could support faster and more energy-efficient information processing by taking advantage of the natural speed and density offered by altermagnetic materials.
The synchrotron-based magnetic analysis methods established during the study are also expected to help researchers identify and study other altermagnetic materials. This approach could accelerate progress in spintronics and open new pathways for future electronic devices.
The research appears in the journal Nature Communications, titled “Evidence for single variant in altermagnetic RuO2(101) thin films.”
No comments:
Post a Comment