California Data Center Intel

Recent California data center news

• Dec 17, 2025

  The next big shifts in AI workloads and hyperscaler strategies

  McKinsey & Company outlines how AI-driven workloads are forcing US hyperscalers to redesign data center strategies, power sourcing, and campus architectures while rapidly scaling capacity.

  • AI demand is expected to expand US data center power capacity from ~30+ GW (2025) to 90+ GW (2030, ~22% CAGR), with inference workloads growing at 35% CAGR to >90 GW and training at 22% CAGR to >60 GW, driving shifts toward high-density, liquid-cooled, AI-ready campuses, modular builds, and tier 2 markets where power, land, and permitting are more accessible and faster.
  • Hyperscalers are restructuring capital and infrastructure models, including JVs, special-purpose vehicles, lease‑to‑own deals, behind‑the‑meter power (e.g., New APR Energy’s 100 MW+ mobile gas turbines), and hydrogen-powered microgrid campuses, while retrofitting existing sites at $4–7M/MW for co‑locators and $20–30M/MW for hyperscalers to support GPU‑intensive AI and consolidating into multifacility campuses projected to represent ~70% of deployments by 2030.
• Dec 16, 2025

  Uptime Institute's Max Smolaks: Power, Racks, and the Economics of the AI Data Center Boom

  Uptime Institute research analyst Max Smolaks discusses infrastructure forces reshaping AI data centers in a Data Center Frontier podcast episode (published December 16, 2025).

  • Main announcement/action: Smolaks outlines that AI capacity is being built where power exists, not traditional data center markets — citing a crypto-to-AI pivot (former crypto-mining sites repurposed for AI), Applied Digital developing roughly half a gigawatt of AI data center capacity in North Dakota, and industry alignment around 800V DC power distribution and rack-level changes (racks moving toward 200–300 kilowatts per rack). The episode date is December 16, 2025; Smolaks spoke in context of OCP Global Summit 2025 (San Jose) and ongoing industry tracking by Uptime Intelligence.
  • Background and supporting details: The conversation covers growing liquid cooling complexity (need for CFD, digital twins, new commissioning/load banks), rack disaggregation into “compute/power/network” groupings, risk and operator resistance to liquid cooling (filtration, corrosion control), energy storage and UPS/rack smoothing approaches (including GPU PowerBurn), and concentration of GPU-driven economics with Nvidia influencing design. Concrete technical details include 200–300 kW per rack, 800V DC distribution, and mention of capacitors integrated into rack/power shelves to manage “spiky” AI loads.
• Dec 16, 2025

  Southfield City Council greenlights divisive data center plan: ‘Too many unknowns’

  Metrobloks received Southfield City Council approval for a 109,683-square-foot data center on 12.19 acres that will require 100 megawatts of power and was approved by a 5-2 vote.

  • Project details and developer commitments: The site plan approval covers a 109,683-square-foot facility on 12.19 acres (east side of Inkster Road between 11 Mile Road and I-696) that will require 100 megawatts of power; the developer says the project will be built in two phases with an estimated total cost of $1.5 billion (the slide deck lists $500 million, a noted discrepancy). Metrobloks said it currently has funding for land and utility upgrades but not full capitalization and will borrow against a tenant contract once a tenant is confirmed; the facility will use a closed-loop cooling system consuming 10–20 gallons of water per day, and the developer expects ~35 permanent positions and estimated $400,000/year in property tax revenue (early estimate). The city placed conditions on approval including compliance with ordinances, minimal tree/soil removal, and inspection rights, and EGLE can enforce state environmental laws.

  • Community response, oversight, and background: The approval followed nearly six hours of public comment with majority opposition asking for a health impact study, community benefits agreement, and transparency; Councilmembers Charles Hicks and Ashanti Bland voted no (citing insufficient fact-based information and public health concerns from AI data centers referenced in the HBR article), while Coretta Houge supported the project saying the city must consider new technologies. Metrobloks (founded 2024) has not yet built a data center but cites executive experience from Amazon and Meta and lists Detroit as a strategic hub; it is targeting Indianapolis, Phoenix, Miami, and Paris for future development. Next Southfield City Council meeting noted: Jan. 5, 2026 at 6 p.m., Southfield City Hall, 26000 Evergreen Road.

• Dec 16, 2025

  Will Google throw gasoline on the AI chip arms race?

  Google released its 7th-generation TPU codenamed “Ironwood” and is reportedly preparing to sell TPUs to third parties; media reported Meta may consider purchasing about 100,000 units.

  • Main announcement: Google introduced Ironwood (7th-gen TPU) optimized for inference with massive memory scale and bandwidth; reports surfaced that Meta is considering a purchase of ~100,000 units and Google may seek additional external customers (reported last month, with follow-up reports a few weeks later).
  • Context and details: Analysts (Jack Gold of J.Gold & Associates, Alvin Nguyen of Forrester Research) say TPUs are targeted at inference workloads and complement Nvidia GPUs used for large language models; Google has previously provided TPUs to some external companies (mainly startups from ex-Googlers or Google-sponsored startups), and selling TPUs broadly would require building support and on-premises infrastructure capabilities that Google has limited experience with.
• Dec 16, 2025

  How Data Centers Became the Hidden Backbone of Our Modern World

  Stepchange Ventures’ co-founders, writing in the MCJ Newsletter, outline how data centers have evolved into critical infrastructure, how AI is driving unprecedented power demand, and why this creates both grid constraints and opportunities for more sustainable, abundant energy and compute.

  • Data centers emerged from early internet hubs like MAE-East and One Wilshire into hyperscale regions such as Ashburn, Virginia, where data-center-zoned land can reach $6M per acre, while overbuilt fiber networks and subsequent advances like virtualization, cloud (EC2 in 2006), containers, and serverless steadily increased hardware utilization and enabled Web 2.0 and hyperscale cloud growth.
  • Power efficiency innovations—including PUE (Power Usage Effectiveness) invented by Christian Belady (typical mid-2000s PUE ≈ 2.5, pushed toward 1.1 by hyperscalers) allowed internet traffic to grow 17x (2010–2020) with relatively flat energy use, but the rise of AI GPUs and 5GW-scale builds now creates 10–100x more power-hungry data centers, intersecting with broader load growth from industrial expansion and electrification, and prompting calls to reengineer chips, grids, and infrastructure for an abundant, sustainable era.
• Dec 16, 2025 · Tech Xplore

  Transparency in AI companies falls to new low

  Stanford, with coauthors at Berkeley, Princeton, and MIT, published the 2025 Foundation Model Transparency Index reporting a marked decline in corporate transparency across major AI developers.

  • Main announcement: The 2025 Foundation Model Transparency Index finds the industry average transparency score fell to 40/100, down from 58/100 in 2024; the Index assessed 13 companies and identifies three performance clusters (top ≈75, middle ≈35, low ≈15). Key score changes include IBM 95/100 (highest in Index history), xAI and Midjourney 14/100 (among the lowest), Meta 60→31, and Mistral 55→18. The 2025 edition adds four companies for the first time: Alibaba, DeepSeek, Midjourney, xAI.
  • Background and details: The Index evaluates companies on 15 areas (training data, risk mitigation, economic impact, etc.) and finds systemic opacity on training data, training compute, model use, and societal impact. 10 companies disclose no environmental-impact data (AI21 Labs, Alibaba, Amazon, Anthropic, DeepSeek, Google, Midjourney, Mistral, OpenAI, xAI). The report notes delays/non-release of documentation by major firms (e.g., Google’s Gemini 2.5 model card delay; Meta did not release a technical report for Llama 4) and cites the Index as an input for policy interventions already under way in California and the European Union.
• Dec 16, 2025

  How Researchers Are Driving Advances for Data Centers

  Lawrence Berkeley National Laboratory outlines multiple initiatives to improve the energy and water efficiency of U.S. data centers that underpin AI and high-performance computing.

  • Berkeley Lab’s programs and tools – including the updated United States Data Center Energy Usage Report (2014–2028), NERSC liquid/direct-to-chip cooling retrofits, CoE tools (DC Pro, Air Management Tool Suite, Electrical Power Chain Tool), and the Modelica Buildings Library and MOSTCOOL – target reduced electricity use, cooling energy, and water consumption, e.g., 42% cut in non-IT power, >2 million kWh and 0.5 million gallons saved annually at NERSC, plus 8% cooling energy reduction and >1 million gallons saved via best-practice retrofits.
  • Industry collaboration and standards efforts span work with top AI companies, utilities, Meta, Carrier, Open Compute Project, Energy Efficiency High-Performance Computing Working Group, and BP Castrol, producing liquid cooling specifications down to chip level, guidelines for transfer fluids, workforce training via the Data Center Energy Practitioner Training Program, and stakeholder input (≈150 attendees at the 2025 OCP Global Summit listening session) to refine models showing data centers may reach up to 12% of U.S. electricity consumption by 2028.
• Dec 16, 2025

  Development Deal Will Provide Hydrogen for California Data Centers

  Vema Hydrogen has entered a hydrogen purchase and sale agreement with Verne to supply Engineered Mineral Hydrogen (EMH) to Verne’s data center customers; operations could begin as soon as 2028.

  • Agreement details: Vema will increase production to more than 36,000 metric tons per year of EMH over the course of a 10-year agreement; the deal establishes a framework to supply reliable, affordable clean power to Verne’s data center customers and operations could start as soon as 2028.
  • Background & partners: Vema’s proprietary EMH technology uses geoscience to produce high-purity sustainable hydrogen and was recognized as a qualified supplier by The First Public Hydrogen Authority (FPH2); Verne is supported by investors and partners including Amazon’s Climate Pledge Fund, Breakthrough Energy Fellows, Caterpillar VC, NextEra Energy Resources, United Airlines Ventures Sustainable Flight Fund, Collaborative Fund, ARPA-E, and the U.S. Army. The companies cite data center energy consumption projections (expected to double by 2030 to roughly 945 terawatt-hours).
• Dec 15, 2025

  Permitting the AI Boom: A New NEPA Landscape for Energy Infrastructure Projects

  The authors (Hunton attorneys) summarize how recent Congressional amendments, a Supreme Court decision, CEQ action, and executive orders have reshaped NEPA to streamline permitting for energy infrastructure and data center projects.

  • Main announcement: Congress enacted NEPA amendments (Fiscal Responsibility Act of 2023 and the One Big Beautiful Bill Act 2025) and the Supreme Court issued Seven County Infrastructure Coalition v. Eagle County (May 29, 2025) establishing substantial deference to agencies and limiting NEPA to reasonably foreseeable environmental effects; the OBBBA allows developers to opt into expedited review by paying 125% of anticipated EA/EIS costs, triggering 180-day deadlines for EAs and one-year deadlines for EISs.
  • Background and implementation details: CEQ rescinded prior NEPA regulations (interim final rule following Feb 2025 D.C. Circuit decision), issued revised guidance (Sept 29, 2025) and directed agencies (DOE, Corps, DOI) to adopt streamlined procedures (EA/EIS deadlines, fewer public comments, expanded categorical exclusions, allowance for applicant-prepared documents); Presidential EOs (Jan 20, 2025 and July 2025) and the AI Action Plan prioritize permitting for fossil fuel and dispatchable baseload resources and call for categorical exclusions and FAST-41 listing; DOE/Secretary directed FERC to initiate rulemaking to expedite interconnection of large loads (data centers) under 42 U.S.C. § 7173.
• Dec 15, 2025

  Environmental Activists Abandon Persuasion For Litigation

  Greenpeace International has filed an “anti-SLAPP” suit in the Netherlands against Energy Transfer after a North Dakota jury found Greenpeace liable and ordered roughly $660 million in damages; separately, California’s attorney general and environmental groups sued ExxonMobil over alleged contributions to plastic pollution in September 2024.

  • Main action: Greenpeace International filed an anti-SLAPP case in the Netherlands against Energy Transfer after a North Dakota jury found Greenpeace liable for defamation, property damage, and civil conspiracy and ordered roughly $660 million in damages; the author frames this as an attempt to escape responsibility following the domestic verdict.
  • Background and additional details: In September 2024California’s attorney general and environmental groups sued ExxonMobil over single-use plastics despite ExxonMobil’s claims it does not manufacture single-use plastics and has processed more than 60 million pounds of plastic waste into raw materials; the piece also references a 2024 YouGov poll showing 73% of voters rank price and availability of energy as top concerns and warns that litigation could hinder energy supply needed for AI-driven data centers.