Anthropic's reported $518 billion in infrastructure commitments and AMD's $8.2 billion deal for World Labs added to the day's spending news. Local opposition surfaced in Weld County and Jefferson Davis County, while the US Senate is set to vote on AI data center regulation on Wednesday.
Samsung committed $1 billion to Helix, the AI infrastructure firm. The Information reported the $1 billion investment in the KKR-backed company. The Next Web said Samsung and five affiliates put in the money, and qz.com described Helix as backed by KKR and Nvidia.
Anthropic has locked in $518 billion of AI infrastructure commitments. TechTarget carried the figure, citing a report.
AMD agreed to acquire World Labs for $8.2 billion. proactiveinvestors.com framed the deal as part of an AI infrastructure push, and Digital Watch Observatory identified World Labs as Fei-Fei Li's start-up.
Bain says the AI build-out needs $6 trillion in annual revenue to justify it. The National reported the figure. Futu's coverage of the same research cited a $4.2 trillion gap.
Capital markets are opening to data-center builders. Accelevation and its shareholders raised $540 million in a US IPO ahead of a Wednesday listing, according to Benzinga and CXO Digitalpulse. Bloomberg reported that Brookfield-backed developer 5C is eyeing an IPO.
Data-center load growth is raising risk for grid operators. TD World reported that 75% of utilities face higher operational risk as AI data center load grows. On the supply side, Bloom Energy shares rose nearly 11%, TradingKey reported, and NAI500 said the stock had risen 16-fold in three years.
Google Research, Meta AI and DeepMind's KV-Kaizen reports a 32× smaller decode cache with accuracy preserved, and Tsinghua and Zhipu AI's STEPQuant cuts serving memory by 68.7%. The largest benchmark gains were +13.5 points on ProgramBench for meta-reasoning (Meta AI, CMU, Amsterdam, Apple) and +11.4% on BrowseComp-Plus for Context Language Models. Hugging Face and ETH Zurich released weights and code for a fully synthetic single-stage training recipe, STEPQuant released code only, and the other six papers released nothing yet.
2 charts built from 3 stated figures across 11 stories, the first planned by the composer from the record. Also had the data for when the energy stories landed, pressure and relief by topic. Held back: power committed in deals that state a figure (no deal in this edition states the capacity it commits); power disclosed per edition · last 7 (only 1 of the last 7 editions put a number on it); what today covered · topic by region (today’s energy stories fall in one topic or one region — a grid needs two of each).