Modders have lifted a restriction that stops some GeForce RTX cards from using their full AI compute power, doubling theoretical throughput on an RTX 4090. Real-world gains are smaller, and the tweak carries risks.
NextDC, Australia's largest datacentre operator, suffered a setback in its plan to double its M3 site in west Melbourne. Powering the expansion meant replacing timber poles on suburban streets with taller concrete ones for higher-voltage lines, which residents opposed.
Nearly all of Amazon Web Services' capacity for 2027 is already reserved, along with much of what is planned for 2028. Despite $220 billion of investment planned this year, demand for AI infrastructure still outstrips what the group can supply.
Meta has announced a two-part fundraising worth $95bn to pay for its AI spending, much of which goes into data centres. The move comes as Elon Musk warned that money for smaller AI players is drying up.
Eight years ago, chipmaker Marvell Technology chose to jump ahead to more advanced process nodes. SiliconANGLE reports that the bet has kept the company on a rising path as demand for AI chips grows.
At the 2026 Open Compute Project summit, Meta detailed the networking hardware behind its Muse Wonder and Muse Spark models and its gigawatt-scale Prometheus cluster. Highlights include IcePack and IceCube, its first liquid-cooled network rack, with 102.4T switch blades in Broadcom Tomahawk 6C and NVIDIA Spectrum-6 versions.