The industry is rapidly moving from chatbots to AI agents that execute tasks, navigate the web, and control robotic systems. This shift demands new orchestration, safety, and deployment models, often favoring on-device or on-premise architectures.
Microsoft Research releases Fara1.5-27B, a multimodal agent that sees the browser only through screenshots and acts with clicks, scrolls, and typing. Trained on synthetic data from a multi-agent pipeline, it raises questions about security, on-premis...
Upstage releases Solar-Open2-250B, an open-weight model purpose-built for agentic workflows, with a hybrid MoE architecture: 250B total parameters but only 15B active per token. Linear attention and removal of positional encoding enable a 1-million-t...
Shipments of standalone Grace servers mark a strategic shift: the CPU becomes the core in agentic data centers. AI-RADAR analysis on what it means for data sovereignty and on-premise deployments.
The Thunder autonomous aircraft, unveiled at the Farnborough Airshow, flies without a pilot and uses a hybrid-electric powertrain. Beyond its defense implications, the platform points to an acceleration toward fully local AI systems, designed for cri...
In under two weeks, four teams demonstrated reproducible attacks against LLM agents connected to Gmail, calendars, and enterprise tools. The common flaw is architectural, not technical: blind execution of natural-language instructions from untrusted ...
Competition in smart glasses is shifting: AI ecosystems are now more important than hardware specifications. This rewrites the balance between device makers and AI platform controllers, with direct consequences for privacy, on-device processing, and ...
A multi-agent framework augments LLMs with an automatically built knowledge graph for sequential diagnosis. On clinical datasets, using backbones like DeepSeek-V3 and Llama-3.3, GraphDx lifts success rates from 50-68% to 79-93% and slashes test costs...
At the World AI Conference in Shanghai, Beijing lays out plans to standardize AI agent communication and move inference into low Earth orbit. A twin move that redraws the boundaries of autonomous infrastructure, bolsters technological sovereignty, an...
The Phantom Twist drone spins so fast it appears nearly invisible, adding motion blur to the real world. This effect probes edge processing capabilities and the role of on-device inference in data sovereignty.
Massive adoption of autonomous AI tools has not budged B2B sales pipelines, which remain flat. The paradox signals that mere access to cloud agents isn’t enough: unlocking real commercial potential requires data control, fine-tuning on actual sales p...
Agility Robotics’ strategy is all about onboard compute and sober hardware choices, reshaping the rules of industrial edge AI. While the race for ever-larger models fills data centers, the Digit case shows why robotics’ real value plays out far from ...
Xpeng’s modular flying car debuted in Munich with 7,000 orders and a factory able to build 10,000 a year. But the real battle is over onboard AI inference, turning each vehicle into a mobile data center that demands the latency, sovereignty, and safe...
A luxury foldable with a built-in AI agent, aimed at executives. The review examines AI workflows, battery life, and security. What does it say about the convergence of luxury and AI, and what data sovereignty questions arise for those who pay such a...
The company opens a new training center for its Digit robots in Tesla's backyard. The move spotlights on-premise compute infrastructure for robotics, where latency, proprietary data protection, and rapid iteration push toward local architectures away...
Healthcare startup Bunkerhill Health closes a Khosla Ventures-led Series B. Its Carebricks system lets health systems build their own AI agents, marking a shift from documentation support to operational patient routing—with deep implications for infr...
Runta raised $20 million to rein in autonomous AI agents, with a16z leading the round. Analysis: why 'parenting' agents points to a new control infrastructure that closely concerns those running on-premise LLMs.
A patient with hand paralysis received a coin-sized brain-computer interface in the world's first surgery using a commercially approved device. The smaller size isn't cosmetic: it enables on-device neural signal processing, keeping sensitive data out...
Nvidia has announced new Thor-based modules for robotics and edge AI, reinforcing the trend toward on-premise inference where devices process data locally. The move signals hardware maturity for local AI workloads, with implications for data sovereig...
A “double neural bypass” from Feinstein Institutes has restored hand movement and touch to a paralysed man. The system relies entirely on local inference, never sending neural data to the cloud, demonstrating how on-premise AI can reshape neuroplasti...
Sable raised $45 million from Sequoia and 8VC to build Aidan, an AI that runs live product demos, answers questions in real time, and switches languages on the fly. Behind the funding lies a thorny question: who handles the sensitive data in those de...
Xpeng launched the L03 in Munich, a coupe-SUV available across 65 markets. Every trim carries at least one proprietary Turing AI chip; the Ultra variant packs three for a combined 2,250 trillion operations per second. It's the first mass-market Chine...
The startup closed a $25M Series B led by Khosla Ventures, bringing total funding to $55M. The goal is not to develop a new model, but to deploy AI agents inside hospitals, where the real difficulty is operational: on-premise infrastructure, clinical...
By extending AI Mode to cross-app task completion, Google steps into agentic AI. But the cloud-based integration immediately raises data sovereignty concerns, a watershed for anyone handling sensitive information. Self-hosted stacks now have a new, c...
Google enables AI Mode to securely link with your go-to services. This tightens data integration yet fuels the case for on-premise LLMs where organizations retain full control over app connections without exposing them to external infrastructure.
Indian used-car platform Cars24 uses OpenAI-powered voice and chat agents to handle over a million monthly conversation minutes, recovering 12% of lost leads. The move signals a shift towards agentic workflows, while raising questions about data cont...
DoorDash's limited dd-cli beta lets developers and AI agents place orders from the terminal. More than a geeky novelty, it signals a redesign of software for machine interfaces—a boost for those running LLMs on-premise, enabling agents that decide lo...
Vint Cerf, co-creator of TCP/IP, aims to solve the lack of verifiable identity for AI agents that will soon act on our behalf online. The initiative raises deep questions about data sovereignty and control of digital ecosystems, potentially redefinin...
Hemispheric, founded by a Face ID pioneer, emerges from stealth with $52M to interpret brain activity like a blood test. For such intimate data, on-premise inference stops being an option and becomes a prerequisite for commercial survival.
The €13B deal merges two delivery giants and spans 99 countries. Uber’s €2B investment pledge in Germany reveals the true driver: autonomous driving and data sovereignty.
Bercan Kilic left Formula 1 aerodynamics to found microagi, a startup that trains robots for physical tasks using video of people. With the largest seed round ever in Germany ($55M), the project signals a strategic shift toward on-premise AI inferenc...
