AI Infrastructure & LLMs
Large language models, AI assistants, data centres, AI chips, cooling, optical networks and the power systems behind modern artificial intelligence.
-
Raystech’s 500 MW AI Campus Plan Starts With a 130 MW Proposal
Raystech has outlined an early-stage South Australian campus, with a first phase around 130 MW and possible growth to 500 MW. Connection and planning work remain ahead.
-
CERN’s Mistral Pact Turns Scientific AI Into a Data-Infrastructure Test
CERN and Mistral plan a foundation model trained on simulated particle collisions. The partnership will test whether shared models can serve scientific tasks at an immense data scale.
-
NVIDIA and Cisco’s New Optical Plug Targets the Space Between AI Machines
USFP aims to carry more data through a familiar-sized optical module. The announcement separates an initial 6.4 Tb/s design from a path to 12.8 Tb/s, with deployment evidence still ahead.
-
Tasmania’s AI Community Fund Moves the Data-Centre Debate Beyond Jobs
Tasmania expects major AI and data-centre developers to contribute to a community fund. The policy’s practical value will depend on contribution rules, transparency and local outcomes.
-
Adani and Jabil’s AI Hardware Pact Puts Rack Manufacturing Ahead of the Build
A proposed 50:50 venture would assemble AI racks, networking, power and cooling equipment in India. The non-binding agreement has not yet become an operating factory.
-
Scaleway’s Nordic AI Expansion Tests What Sovereign Compute Really Means
Scaleway announced AMD-powered capacity in Sweden and Nvidia systems in Copenhagen. EU hosting strengthens jurisdictional control without creating a European-only supply chain.
-
Microsoft’s Maia 300 Turns AI Inference Into a Power-Efficiency Contest
Microsoft says its 2027 Maia 300 system will deliver 40% more tokens per megawatt than leading GPU systems. The figure is a vendor forecast, not an independent benchmark.
-
Upscale’s Token Fabric Tries to Make Rival AI Chips Share One Network
Token Fabric combines scale-up links, Ethernet scale-out and network software for heterogeneous AI systems. The commercial launch still needs independent deployment evidence.
-
The 800-Volt DC Push Is Rewiring the Roadmap for Megawatt AI Racks
OCP and NVIDIA are advancing 800-volt direct-current distribution for megawatt AI racks. The roadmap could reduce copper and conversion losses, but deployment starts later.
-
A Data Centre’s Tier Rating Stops Short of an AI Service Guarantee
Tier ratings describe facility infrastructure. AI-service reliability also depends on operations, software and dependencies that need their own evidence.
-
Four-Bit AI Shrinks the Model, but Accuracy Still Has to Survive
Low-bit quantisation can reduce AI memory requirements. The useful result depends on which numbers are compressed, the runtime and the quality retained.
-
Data-Centre Noise Needs Better Evidence Than a Single Decibel Reading
Continuous hum can become a community concern, but scattered readings are not a comparable exposure dataset. A new review sets out the measurement gap.