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
Meta’s Muse may be the AI agent in the spotlight today, but investors are looking past the product leaderboard and betting that agent growth will boost demand for chips, memory and power. That infrastructure thesis is gaining ground, even as the winning AI brand remains anyone’s guess.
Read the report & discussion ↗AWS AI Watch
AWS has published a reference architecture for letting multiple teams share one SageMaker HyperPod cluster while keeping their workloads and access separate. The practical appeal is clear: share expensive GPU infrastructure without leaving every team to fend for itself on permissions, quotas and cost tracking.
AWS AI Watch
Amazon’s Mississippi data-centre expansion faced a stark warning from local utility Entergy: during peak demand, the company might have to disconnect Amazon’s facilities to protect other customers. The report makes the AI infrastructure boom’s power constraint unusually tangible: grid access cannot simply be assumed.
Watch Desk
Behavox has committed $8 million to secure dedicated GPU capacity from UK cloud provider Civo, for processing regulated customers’ data in the UK. The deal gives financial institutions another option for keeping sensitive data within the country while running AI models, though it is not evidence of a…
Astera Labs Watch
Astera Labs has announced a broad PCIe 7 signal-conditioning portfolio for AI infrastructure, spanning copper and optical connections. Its central pitch is flexibility: customers can choose the link technology to suit the reach, power and density they need, with the product family expected to begin sampling to selected…
TSMC Watch
TSMC has agreed a five-year, $2bn manufacturing deal with GlobalFoundries for silicon interposers used in its CoWoS advanced-packaging ecosystem. The planned Malta, New York production would give TSMC a US-based source for a component increasingly important to AI and high-performance computing chips.
Watch Desk
A Senate energy bill would make large data centres cover more of the electricity-grid costs linked to their growth, while speeding up energy infrastructure approvals. The proposal could reshape who pays for the AI build-out, but it is not yet law.
MIT CSAIL Watch
MIT researcher Christina Delimitrou is applying machine learning to make data centres more efficient, reliable and secure. Her work targets a practical problem: getting more useful computing from existing equipment could reduce wasted energy and pressure to build more capacity.
Watch Desk
At a recent Modal conference, AI leaders made the case for software agents that work towards outcomes, while developers and company executives described the gap between AI doing much of a task and taking it off people’s hands. The question is no longer just what models can do,…
AMD Watch
AMD’s ROCm 10.1 release adds faster storage-to-GPU data paths, NUMA-aware host memory and a broader set of tools for deploying and profiling AI workloads. For teams running models on AMD GPUs, the changes target the unglamorous work of getting data to the hardware and understanding where a workload…
Watch Desk
Arista has announced an Ethernet rack-scale portfolio for AI infrastructure, led by a design intended to connect up to 144 accelerators in a single-hop domain. The pitch is an open alternative aimed at heterogeneous systems beyond Nvidia’s ecosystem, where networking becomes part of the rack’s core architecture rather…
OpenAI Watch
Broadcom is arranging more than $50 billion in financing for a custom AI chip it is developing with OpenAI, Bloomberg reports, citing the Wall Street Journal. The scale points to the financing challenge behind the race to build AI computing capacity, not just the chip itself.
Watch Desk
ASUS has announced ProArt laptops powered by NVIDIA’s RTX Spark platform, aimed at running AI models and creative workflows locally. The P16 and P14 are listed for US pre-orders from $2,599.99, with the first units due on 16 October; the announcement also outlines a compact desktop system for…
Applied Digital Watch
Applied Digital reported $341.9 million in fiscal first-quarter revenue, up 322% year on year, while setting out a growing pipeline of AI data-centre projects. The company’s results show the scale of demand it is building for, alongside the cost and execution challenge of turning that pipeline into operating…
DeepSeek Watch
DeepSeek has published two new open-source repositories targeting Huawei’s Ascend accelerators, for matrix multiplication and mixture-of-experts communication. Their familiar API shapes could make experimenting beyond Nvidia’s CUDA ecosystem less of a rewrite, although the supported hardware is limited.
Watch Desk
AI’s climate promise runs into an awkward question: can it help develop cleaner technology while its expanding energy demand props up fossil fuels? A new Jacobin essay argues that political and economic incentives matter as much as what AI might make possible.
Watch Desk
An open-source project called openTPU uses AI agents in the design of an AI accelerator and has run several language models on a physical FPGA card. Its reported results offer a concrete, inspectable experiment in AI-designed hardware, though memory access is currently the bottleneck.
Microsoft AI Watch
Microsoft Research Asia has released Agent Lightning v1.0, an open-source framework for training AI agents with reinforcement learning while keeping the agent harness used in deployment in the loop. Its coding-agent example reports a 14.6 percentage-point improvement on SWE-bench Verified, with about 6,000 training samples.
Watch Desk
San Francisco has passed a temporary ban on new data centres within city limits, according to Bloomberg. The decision puts a practical question beside the AI boom: how much room should a city make for the infrastructure that powers it?
