US Data Centers News

Our World in Data has published an analytical article on how much electricity data centers and artificial intelligence use, and how much energy individual AI queries consume.

• The article estimates that data centers used around 485 TWh in 2025, with AI-focused data centers accounting for about 155 TWh and 0.5% of global electricity.
• It compares IEA and S&P Global estimates, notes that crypto mining explains part of the gap, and says a typical ChatGPT/Gemini/Claude query uses only a small amount of electricity, while future AI demand remains highly uncertain.

AMAX is promoting its AI factory integration services using NVIDIA networking technologies. - It says organizations should design and deploy scalable AI factories with NVIDIA Quantum InfiniBand and NVIDIA Spectrum-X Ethernet, covering fabric selection, topology design, compute and storage integration, validation, and ongoing production operations.

• The page highlights workload-driven network design considerations such as GPU count, latency, bandwidth, storage throughput, redundancy, and rack power/cooling capacity, and lists supported platforms including RackScale 32 with NVIDIA HGX B300, RTX PRO Blackwell Server Edition, RackScale 72 with NVIDIA GB300 NVL72, and NVIDIA GB200 NVL72.

TaiwanPlus reports that Taiwan’s machine tool sector is recovering as AI-driven demand supports new orders.

• Major manufacturers reported stronger bookings amid global AI hardware expansion and rising demand from semiconductor, aerospace, and AI server cooling industries.
• The story says the sector is recovering from US tariff pressures; it is a news report/commentary rather than a first-time corporate announcement.

NVIDIA has published a blog post explaining agentic AI post-training and how its hardware and software stack are positioned to maximize intelligence per dollar.

• The post says agentic post-training is a continuous loop that uses reinforcement learning, rollouts, rewards and weight updates to keep improving models as environments and tools change.
• It highlights NVIDIA Nemotron 3 Ultra as a 550-billion-parameter MoE model that scored 71.7% on SWE-bench verified, and cites workloads at Prime Intellect, Perplexity and Together AI running on NVIDIA systems.
• The article also says NVIDIA Blackwell lowers cost per run, while Vera Rubin is designed to train the largest models with one-fourth the GPUs of the Blackwell generation.

ETH Zurich has announced that it copied openly available NASA climate and Earth observation datasets to the Swiss National Supercomputing Centre (CSCS) in Lugano to support AI-based weather forecasting and natural-hazard detection.

• Around 100 petabytes of publicly available NASA data were copied over the past few months to CSCS in Lugano; the data includes greenhouse gases, clouds, precipitation and ice sheets and will be continuously updated.
• ETH says the data will be made available for researchers in Switzerland and around the world; it also plans to add NOAA datasets and link the repository to the Alps supercomputer and AI processing workflows.
• The broader strategy includes the new ETH Swiss GeoLab in Lucerne over the next ten years, with an AI model trained on satellite, drone, sensor and monitoring-station data to identify landslide-prone areas.

The European Commission has proposed a regulation to future-proof electricity bills by reducing system costs, improving grid efficiency, and supporting electrification and digitalisation.

• The proposal covers network charges, smart grids, energy data exchange, smart meters, taxation, and grid connection measures; it includes a minimum smart meter trajectory of 50% by end-2030 and 75% by end-2033, plus measures on grid data sharing and connection prioritisation.
• It is a Commission staff working document accompanying COM(2026) 600 final / SWD(2026) 600 final dated 17 July 2026; it is primarily an announcement/proposal and impact explanation, not a report of a completed deal.
• The text cites multiple quantified impacts, including EUR 600 billion in grid investments by 2030, EUR 1.2 trillion by 2040, and examples such as EUR 22.5 billion potential savings in the Netherlands and EUR 7 billion TSO savings in France over 15 years.
• It also states that tariffs could be adjusted for data centres and other energy-intensive users, and that data centres can provide grid flexibility through cooling, batteries, backup generators and workload shifting.

The Council General Secretariat has circulated an annex to the Council Implementing Decision amending Ireland’s recovery and resilience plan.

• The annex sets out reforms and investments across the green transition, digital transformation, social recovery, audit/control, and REPowerEU, with many milestones and targets through 2026.
• It includes specific commitments such as a EUR 1,153,797,007 estimated total plan cost, a EUR 323,803,933 first instalment, and multiple project-level budgets and targets for rail, data centres, schools, housing, health, and biomethane.
• This is a formal EU policy/legal document, not a company announcement, and it mainly records planned milestones, targets, and financing arrangements rather than a new market deal.

The European Commission has proposed a regulation to future-proof electricity bills in the Union by reducing system costs, reforming network charges, and aligning electricity taxation with electrification goals.

• The proposal is a first-time legislative initiative from the Commission, dated 17 July 2026, and it would amend Regulation (EU) 2019/943.
• It includes smart meter rollout targets of 50% by 31 December 2030 and 75% by 31 December 2033, plus a rule that electricity should be taxed no higher than natural gas; it also allows special network charge regimes for energy-intensive industries and data centres.
• The financial statement estimates EUR 0.391 million for ACER in 2027 and EUR 0.255 million for DG JRC in 2027, with 5 Commission FTEs and 2 ACER FTEs required overall.
• The article is an EU Commission proposal/explanatory memorandum, not a report of a completed deal or policy already adopted.