AI hardware

Nvidia has announced plans for a watchdog chip designed to run alongside every AI agent, monitoring its behavior and intervening when needed. The proposal is drawing debate among technologists over whether dedicated hardware can meaningfully police autonomous AI systems, and whether Nvidia is positioning itself as the gatekeeper for agent safety.

Meta is signalling plans to build its Muse AI technology into a wider range of consumer gadgets, according to TechCrunch. The move suggests the company wants Muse to become a core layer in upcoming devices beyond its existing headsets and smart glasses. Details on which products or partners are involved have not been laid out yet, leaving observers guessing about Meta's next hardware push.

Meta reportedly wants people to build their own Muse gadgets, extending its hardware ambitions beyond official first-party devices. The move would let developers and tinkerers create their own versions of the wearable technology, though details on how the programme would work remain scarce. The news, reported by Engadget, adds to speculation about Meta's broader strategy around AI-powered wearables and open hardware ecosystems.

Amazon is reportedly looking to sell roughly $8 billion worth of Nvidia chips to outside investors, according to the Financial Times. The move would shift the cost of AI computing infrastructure away from Amazon's own balance sheet, and it is drawing attention as investors question whether tech companies' massive AI hardware spending can generate adequate returns.

Meta has released the source code needed to build its Muse AI gadgets, letting developers and hobbyists create their own versions of the hardware. The move follows Meta's broader push into open-source AI tooling and could spark a wave of third-party AI wearable and smart device projects. Details on licensing and supported hardware remain limited in early coverage.

Robotics company Figure AI has announced the decommission of its F.02 humanoid robot, detailing the decision in a company news post. The move is being discussed on tech forums among robotics watchers, who see it as a sign of how quickly humanoid hardware generations are being cycled out as the field accelerates development.

Figure AI has announced the decommission of its F.02 humanoid robot, marking the end of the program and pointing to a successor or revised hardware direction. The company has not detailed in the announcement what specifically prompted the retirement or what will replace the robot. Robotics watchers are weighing whether the move signals a design pivot or an accelerated roadmap.

A developer reports enabling multi-token prediction (MTP) on an RTX 3090 graphics card raised local LLM generation throughput, while questioning whether the speedup affects coding quality. The write-up, originally published on a personal blog, has drawn attention from AI and open-source software communities interested in getting more performance from consumer GPUs for running large language models locally.

Industry analysts say the bottlenecks holding back AI chip performance are no longer transistor scaling. Power delivery, memory bandwidth and advanced packaging have become the key constraints, shifting how chipmakers like Nvidia, AMD and TSMC approach next-generation AI hardware design and investment.

Amazon is looking to sell $8bn worth of Nvidia chips to investors, according to the Financial Times. The reported move suggests the company is seeking to recover costs on a large stockpile of AI computing hardware, fuelling debate about whether big tech firms have over-invested in data centre capacity during the artificial intelligence boom.

A new video explores a speculative question gaining attention among tech enthusiasts: what would computing look like if the industry moved away from graphics processing units? Discussion centres on whether alternatives could handle the parallel workloads that GPUs currently dominate in AI, gaming and scientific computing, and what a post-GPU hardware landscape might mean for the future of chip design.

OpenAI is deploying its custom Jalapeño ASICs in rack-scale systems hosted by AMD EPYC 'Turin' CPUs, bypassing Nvidia's Vera platform. The choice signals OpenAI is building its AI infrastructure around AMD server processors rather than Nvidia's latest offerings, a notable shift given the two companies' close relationship. Tech observers are weighing what the pairing means for data centre economics and for Nvidia's grip on AI hardware.

Robotics company Figure AI is reportedly training retired robots to dive into molten steel, according to a Gizmodo report. The story, circulating in robotics and AI discussion circles, highlights how the company handles decommissioned hardware while continuing its work on humanoid machines. Details beyond the headline remain thin, and the purpose of the molten steel dives is not spelled out.

Amazon is reportedly approaching investors to help fund an $8 billion financing structure centered on Nvidia chips, a move that would spread the cost of its massive AI computing buildout. The reported plan highlights how tech companies are turning to creative financing to afford the expensive hardware needed to stay competitive in artificial intelligence.

Amazon is reportedly looking to move roughly $8 billion worth of Nvidia AI chips off its balance sheet, according to an Investopedia report. The move suggests the retail and cloud giant may be rethinking parts of its AI hardware stockpile, a development drawing attention given Nvidia's central role in the AI data-center buildout and the scale of chip spending across hyperscalers.

Federal prosecutors in California have charged a man with smuggling roughly $300 million worth of Nvidia chips to China, in defiance of US export controls on advanced semiconductors. The case highlights ongoing concerns about high-end American AI hardware reaching Chinese buyers despite restrictions, and is drawing attention to enforcement of technology export rules.

Amazon is reportedly arranging to offload roughly $8 billion worth of AI chips in a sale-leaseback arrangement, selling the hardware while continuing to use it under lease. The move would free up capital for the company's vast data center buildout as it races rivals in AI infrastructure. The deal highlights how costly AI chip deployments have become for cloud providers.

Google has unveiled the first prototype satellite for Project Suncatcher, an initiative to test artificial intelligence hardware in orbit. The prototype is designed to evaluate how AI compute systems perform in space, where solar power is more abundant. The announcement has drawn attention from technology observers discussing what it could mean for the future of data centers and space-based computing.

Amazon is reportedly arranging to move about US$8 billion of chips off its balance sheet, according to a Financial Times report picked up by the Taipei Times. The arrangement would shift ownership of the semiconductors it uses, a move likely tied to how the company finances and accounts for the hardware behind its cloud and AI infrastructure. Details of the counterparties and structure have not been disclosed.

A California-based technology executive has been arrested on allegations of smuggling roughly $300 million worth of Nvidia chips into China, circumventing US export controls on advanced semiconductors. The case highlights ongoing enforcement of restrictions blocking high-end AI hardware from reaching Chinese firms, and is drawing attention to how widely such controls may be being evaded.

Federal prosecutors say a California man ran a smuggling scheme that sent roughly $300 million worth of American AI hardware, including advanced chips, to China in violation of export controls. The case highlights growing US efforts to stop sensitive semiconductors from reaching Chinese buyers, and has drawn attention to gaps in enforcement of tech export restrictions.

Amazon is increasing the price of renting out its AI chips by 15% and is reportedly exploring an $8 billion leaseback arrangement involving Nvidia hardware. The moves point to how aggressively cloud providers are trying to monetize scarce AI computing capacity as demand for chips and data center power keeps climbing.

Meta has launched Muse Gadgets, a line of open-source hardware designed to work with its AI assistant. The products are aimed at developers and small businesses, allowing them to build and customise devices around Meta's AI technology. The open-source approach is being read as a move to expand Meta's AI ecosystem beyond its own apps and attract a developer community.

Alphabet has launched its homegrown Tensor Processing Units into orbit aboard a SpaceX rocket, taking its custom AI semiconductors beyond Earth for the first time. The move highlights the growing competition in AI hardware, with Google showcasing its own chip technology alongside SpaceX's launch capabilities. Details on the mission's purpose and duration have not been disclosed.

The Financial Times reports that Tencent has agreed to lease 100,000 AI chips from Oracle in a deal worth around $7 billion. The arrangement would give the Chinese tech giant access to large-scale computing capacity via Oracle's infrastructure, highlighting how companies are securing scarce AI hardware through leasing rather than outright purchases amid tight supply and export restrictions.

Nvidia is pressing Wall Street banks to finance AI chips using structures similar to aircraft financing, where equipment itself serves as collateral for loans. Banks, however, remain hesitant, unconvinced that rapidly depreciating and fast-obsolete GPUs offer the same durable value as planes. The debate highlights growing questions about how to fund the enormous buildout of AI datacenters and who should bear the risk if chip values fall.

Researchers at UC Berkeley, working with chipmaker FuriosaAI, have published work on HBF, a memory approach aimed at high-throughput serving of large language models. The piece, carried by Semiconductor Engineering, focuses on how new memory architectures could ease the bandwidth and cost bottlenecks that limit LLM inference at scale. The work is being followed by readers tracking hardware innovation for AI infrastructure.

Apple is reportedly developing a smart home camera that would not record video footage of users or anyone else in the home. According to a report shared by Engadget, the device would rely on AI to analyze activity in the home without actually filming it. The approach would put privacy at the center of the product's design, in line with Apple's public positioning on user data protection.