From the archive
AI hardware provides the computing power, memory and networks used to train and run artificial intelligence models. Follow news on GPUs, AI accelerators, data centres, chip design and energy demand. Read how changes in infrastructure affect performance, availability and cost, and distinguish measured results from vendor claims about what the next generation of computing will deliver.
Lead story
Hugging Face has published a detailed preview of tokenizers v1, a major rewrite aimed at stopping text preparation from leaving increasingly fast AI systems waiting around. The project reports large gains in decoding throughput, 76% scaling across eight native threads on an Apple M4 Max, and identical token…
Read the report & discussion ↗DeepSeek Watch
DeepSeek is planning to use more domestic chips to train its AI models, with Huawei expected to begin supplying training processors as early as the fourth quarter, according to The Information. The move would deepen the partnership between two Chinese technology companies trying to reduce their exposure to…
ASML Watch
ASML is working with TSMC, Nvidia, Intel, Samsung and SK Hynix on High NA EUV tools that could print larger AI processors, with a pilot line targeted for 2031. The important point is that the next chip race may depend on making the canvas bigger, not merely drawing…
ASML Watch
The next constraint on AI expansion may not be money, models or demand. VanEck product manager Nick Frasse says the industry’s pace is being set by the slow, stubborn business of building chip factories, data centres and the machinery inside them. ASML, TSMC and memory suppliers sit at…
Unsloth Watch
Unsloth’s 0.1.811-beta release expands Studio from a single-user local AI workspace into a more capable shared tool, adding isolated accounts, AMD support and a preview operating-system sandbox. The practical shift is less glamorous than a new model, but considerably more useful: fewer people having to share one login…
NVIDIA Watch
NVIDIA says demand for its AI infrastructure could exceed available production capacity by roughly $100bn next year, while the company is also using guarantees and other financial support to help customers buy more of its hardware. The result is a revealing picture of an AI boom constrained not…
Watch Desk
HiDream.ai has moved from image and video generation into interactive world models, while building an enterprise software strategy around the technology. The Chinese start-up’s progress matters because it is attempting to turn capable models into repeatable commercial workflows, rather than leaving customers to admire the demo and assemble…
Mistral AI Watch
Samsung Electronics is deploying Mistral AI’s services on-premises across its semiconductor operations, targeting chip design, defect detection and manufacturing yields. The deal matters because it moves generative AI from boardroom strategy into one of the world’s most sensitive industrial environments, where keeping operational data inside the fab is…
Anthropic Watch
Anthropic says Claude has optimised more than 30 open-source biology models in under four weeks, with average speed-ups of roughly four times and a new low-memory mode for much larger molecular systems. The practical significance is clear: some protein-structure and design work could require far less computing, although…
Watch Desk
Oakland is preparing to consider a temporary moratorium on new data centres after more than 70 people addressed the city council in opposition to proposed development. The immediate issue is a planned facility linked to an artificial intelligence and innovation campus, but the larger question is whether city…
DeepSeek Watch
A DeepSeek engineer says AI-written GPU kernels could match or exceed his own work within six to 12 months, while a separate NVIDIA experiment shows a narrower version of that workflow already producing numerically correct code. The important shift is from AI assisting with low-level optimisation to agents…
NVIDIA Watch
CoreWeave has placed seven NVIDIA Vera Rubin NVL72 racks into production across two regions, totalling 504 GPUs. The milestone matters because multi-rack AI systems turn networking, cooling, storage and failure recovery into part of the computer, rather than mere facilities plumbing.
AWS AI Watch
AWS has published a technical walkthrough for making large PyTorch training jobs on Amazon EKS recover from GPU, network and worker failures. The practical change is that teams can combine asynchronous checkpointing with in-process and in-job restarts, rather than losing hours of expensive training whenever one worker misbehaves.
Microsoft AI Watch
Microsoft has open-sourced TauGrid, a Kubernetes platform for scheduling, monitoring and reproducing AI workloads on GPU clusters. The practical promise is simple: researchers can submit training, fine-tuning and inference jobs without becoming part-time Kubernetes administrators.
Watch Desk
Arcee AI, the US developer of open-weight models, has raised at least $150m in a Series B at a reported pre-money valuation of $1bn. The round is a useful signal that investors still see room for smaller, specialist AI companies, even as the biggest laboratories hoover up headlines,…
Intel AI Watch
AAEON has announced the MIX-PTLWV1, a Mini-ITX motherboard built around Intel Core Ultra Series 3 processors and aimed at edge-AI work such as machine vision, robotic automation and industrial inspection. The important bit is not another processor family name in a press release. It is the attempt to…
Watch Desk
Positron AI has raised $875 million to develop Asimov, an inference chip designed to tackle one of AI’s least glamorous but most expensive problems: moving enough memory quickly enough to keep models fed. The startup says its approach could make large-scale AI serving more efficient, but the headline…
Anthropic Watch
Anthropic has signed its first Australian data-centre lease, securing planned capacity of 2.16 gigawatts near Brisbane. Reuters says the facility is expected to begin coming online in 2027 and will handle inference, the rapid processing used to serve AI responses, rather than model training.
