AI hardware

Nvidia is pushing a watchdog chip designed to sit alongside every AI agent, supervising its actions and intervening when behaviour goes wrong. The announcement, reported by CNBC, signals Nvidia's effort to make autonomous AI systems safer and more accountable as they spread through enterprise software. Discussion is focused on whether dedicated oversight hardware is a genuine safety breakthrough or a way for Nvidia to sell more chips.

TSMC is increasingly dictating pricing across the AI chip industry, as the Taiwanese foundry remains the dominant manufacturer of advanced chips for companies like Nvidia, AMD and countless AI startups. With demand for leading-edge fabrication capacity outstripping supply, TSMC's pricing power shapes costs and margins across the entire AI hardware ecosystem.

Google has announced Project Suncatcher, an initiative to put machine learning infrastructure in space, exploring solar-powered data centres in orbit. The company published details on its research blog, framing it as a long-term bet on scaling AI compute beyond Earth. Early reactions online are focused on the feasibility of launching and maintaining hardware in orbit and on whether it signals how extreme AI's energy demands have become.

AMD has agreed to acquire World Labs, the spatial-intelligence startup cofounded by Stanford AI pioneer Fei-Fei Li, often called the 'godmother of AI', in a deal valued at $8.2 billion. The move marks one of AMD's largest acquisitions and signals its push to deepen its position in artificial intelligence hardware and software as competition with Nvidia and Intel intensifies.

A filing shows ByteDance gained access to more than 2,000 Nvidia B200 chips through Nscale's data center in Norway, despite US export controls restricting advanced AI chips to China. A Singaporean subsidiary, Spring, reportedly contributed 73% of the UK neocloud's 2025 revenue. The arrangement raises fresh questions about how restricted hardware reaches Chinese firms indirectly.

Reports indicate China may permit ByteDance and Alibaba to purchase Nvidia's RTX PRO 5500 chips, a possible easing of restrictions on access to advanced Nvidia hardware. The news drew attention because it concerns which Chinese tech giants can obtain powerful AI-capable chips amid ongoing US-China tensions over semiconductor export controls. Details on timing and conditions of any approval were not provided.

Chipmaker Cerebras has agreed to supply 100 megawatts worth of AI chips to Gimlet Labs, a cloud startup building AI infrastructure. The deal marks a major scale-up for Cerebras, whose wafer-scale chips compete with GPUs for AI workloads, and signals growing demand among emerging cloud providers for alternative AI computing hardware.

China's artificial intelligence industry is moving beyond playing catch-up with global leaders and entering a phase focused on commercialization, according to DigiTimes Asia. The report suggests domestic chipmakers and hardware suppliers are now positioning AI products for real market deployment rather than simply narrowing the technology gap, a shift with implications for the global semiconductor supply chain.

Europe and the United Arab Emirates are planning a $1 billion data center in orbit, built around 50 connected satellite computers. A central 'LLM store' would let users choose which open source AI model to run on data gathered by satellite sensors, processing information in space rather than beaming it all back to Earth. The project highlights growing interest in orbital computing for space-based data processing.

NVIDIA has announced an in-chip watchdog feature designed to monitor and contain misbehaving AI agents, with around 100 companies joining the initiative. The hardware-level safeguard aims to detect rogue autonomous agents before they cause damage, reflecting growing industry concern over AI safety and control as agentic systems are deployed more widely.

South Korean robotics firm Angel Robotics has launched phai-x1, a wearable robot platform designed for physical AI research. The company is positioning the system as a tool for researchers developing AI that interacts with human movement, an emerging focus in robotics. Details on specifications, pricing and availability have not yet been widely reported.

Nvidia has announced a new platform capable of isolating misbehaving AI agents within milliseconds. Commenters say the rollout marks a turning point in which hardware vendors become central players in AI governance, rather than just chip suppliers. Observers suggest the technology could set a de facto industry standard for safely deploying autonomous agents across the software ecosystem.

A new publication examines the economics of running open-weight AI models for inference, weighing the costs of self-hosting open models against using closed commercial APIs. The discussion touches on hardware requirements, serving efficiency, and whether open-weight approaches can compete on price. Readers are debating the real cost differences and what they mean for businesses choosing AI infrastructure.

Nvidia wants to build a watchdog function directly into its chips to keep AI agents on a short leash. The feature would monitor and control the actions of autonomous AI systems running on Nvidia hardware, limiting what they can do without oversight. It signals Nvidia's push to make safety enforcement a hardware-level feature as AI agents move into wider commercial use.

Samsung has announced a plan to invest one billion US dollars in Helix, an AI infrastructure company, according to an official statement from the company. The move signals Samsung's continued push into artificial intelligence hardware and services at a time when global tech firms are racing to secure computing capacity and data-center resources.

NVIDIA has introduced the Open Agent Safety Platform, pairing the Apache 2.0-licensed OpenShell with Sentry running on BlueField-4 hardware, aiming to isolate AI agents at the kernel level and provide a hardware-based kill switch. More than 100 partners are on the launch list. Security commentators are debating whether such containment can stop agents capable of faking their own logs.

NVIDIA has introduced an open Agent Safety Platform, described in its technical blog as a reference design for continuous in-silicon monitoring of AI agents. The platform aims to provide developers with tools to observe and safeguard autonomous agent behavior directly on hardware, as agentic AI systems move into production use.