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
Cohere has released Embed 5, a pair of embedding models for finding information in enterprise documents, images and other data. Its practical twist is that teams can index with the higher-quality Pro tier and query with either Pro or Fast without rebuilding that index.
Read the report & discussion ↗ASML Watch
ASML reportedly made no sales in Europe in the first half of 2026, despite being one of the continent’s most strategically important technology companies. The figure sharpens a debate about whether Europe can turn its chipmaking strengths into a wider industrial base for AI.
DeepSeek Watch
DeepSeek and Huawei have released open-source programming tools for Huawei’s Ascend AI chips, including a language called TileLang. The move gives developers a new route to build for Ascend hardware, part of a broader effort to make China’s AI software stack less dependent on Nvidia’s CUDA.
Watch Desk
Cloudflare has released two open-weight decision models, Clef and Clef-flash, which handle structured choices and can also process images and video, according to a report carried by news.lavx.hu citing The Register. The launch gives developers another option beyond hosted decision APIs, although running the models locally comes with…
AMD Watch
AMD’s ROCm AITER v0.1.24.post1 release adds tuned GPU kernels and attention improvements for AI workloads, with the release notes reporting up to 1.69× faster performance on a long-context indexing test. For teams serving models on AMD accelerators, the changes could improve throughput without changing the model itself.
DeepSeek Watch
DeepSeek and Huawei are reportedly developing software for Huawei’s Ascend AI processors, a move that could make it easier to build AI workloads for hardware outside the Nvidia ecosystem. The useful detail is that this means adapting tools and kernels, not automatically translating CUDA code.
CoreWeave Watch
CoreWeave says its Vera Rubin NVL72 systems are now available, with Cognition already running production workloads on the new hardware. The company also announced a CPU built for AI agents, betting that the infrastructure around the GPUs deserves a place in the spotlight too.
TSMC Watch
TSMC and imec’s IC-Link are expanding their collaboration to give chip designers a managed cloud route to advanced-node design tools. The notable change is support for mixed-partner EDA flows, so teams can use their preferred tools and licences without first building the whole design environment themselves.
Cerebras Watch
Cerebras says it achieved five times the inference capacity in early tests by separating prompt processing from token generation and using partner hardware for the first stage. The approach could let operators scale and tune each stage independently, rather than simply adding more of the same accelerator hardware.
NVIDIA Watch
NVIDIA’s Ian Buck says the economics of AI infrastructure are increasingly about how much useful inference a whole data centre can deliver, not simply how many GPUs it contains. His account links that shift to power limits, faster responses for time-sensitive work and continual updates to deployed models.
Watch Desk
Europe’s AI debate is not short of plans; it is short of comparable capital and computing scale. A Diari ARA opinion piece argues that private investment, rather than publicly selected gigafactories, should do more of the heavy lifting.
MLCommons Watch
MLCommons has introduced DLRMv4, a new MLPerf Training benchmark for AI recommendation systems built around users’ sequences of past actions. It replaces an older benchmark design that flattened behaviour into aggregated features, giving hardware comparisons a workload closer to how large recommendation models are trained today.
Watch Desk
A proposed 40,000-acre, nine-gigawatt data-centre campus in Utah was derailed by fierce local opposition after most residents and county officials learned of it late in the process. The Verge’s investigation, discussed on its Decoder podcast, shows how a project pitched as an AI infrastructure landmark became a fight…
Watch Desk
A bipartisan Senate permitting bill would prevent utility ratepayers from bearing the costs of new power and grid upgrades needed for AI data centres, Axios reports. The proposal also reaches far beyond AI, changing how infrastructure projects and environmental reviews are handled.
Watch Desk
Eight vision-language models, one consumer GPU and a useful warning for anyone shopping by benchmark: a model’s tokens-per-second figure may tell you more about its serving setup than its underlying speed. A Hugging Face community article by Aakash Gupta compares ranking accuracy and throughput for video-question-answering workloads on…
NVIDIA Watch
NVIDIA’s data-centre strategy combines cooling and power-management software with partner-led construction, including a planned Australian buildout of up to 2GW by 2027. The useful shift is from selling chips to shaping the facilities and infrastructure needed to run them at scale.
Anthropic Watch
Broadcom has agreed to lend Anthropic up to $42 billion to help finance infrastructure spending, Reuters reports, citing Anthropic’s IPO prospectus. The arrangement ties a major chip supplier more closely to Anthropic’s ability to secure the computing capacity its models need, and the filing flags potential conflicts of…
