Real-time developments tracked by our autonomous agents
PJM announced a plan aimed at managing expected AI-driven electricity demand growth, outlining steps to maintain grid reliability as power needs rise.
Microsoft says it will contribute to grid upgrades and energy costs as it expands data centres, aiming to ease political and community backlash over AI-driven power demand. The company faces scrutiny as soaring electricity use from AI becomes a hot-button issue for Big Tech and the Trump administration.
Meta CEO Mark Zuckerberg announced a new top-level Meta Compute initiative to build tens of gigawatts of AI infrastructure this decade, potentially hundreds over time, positioning infrastructure scale as a long-term strategic advantage. The effort will be led by infrastructure chief Santosh Janardhan and AI executive Daniel Gross, with Dina Powell McCormick focused on government and sovereign partnerships and financing as Meta ramps up large-scale data center investments amid growing local scrutiny.
Bloomberg reports Meta has signed multi-gigawatt nuclear power agreements intended to supply electricity for its AI data centers, reflecting the scale of power demand for new AI infrastructure. The deals highlight hyperscalers turning to nuclear generation for reliable, carbon-free baseload as they expand data-center capacity.
The Washington Post reports growing local opposition to data center developments, highlighting community concerns over land use, power and water demand, and quality-of-life impacts that are slowing or complicating new projects.
Get unlimited access to our complete intelligence feed, custom alerts, and predictive insights. Request early access to see how our autonomous agents can give you 30-90 days advantage.
Goldman Sachs commodities research warns nearly every US power grid could run short on reserve capacity by 2030 as AI data center demand surges, with transmission bottlenecks risking slower buildouts and potentially giving China an edge in the AI race.
Brookfield Asset Management is creating a Radiant cloud unit to lease AI chips directly to developers, using capacity from its own data center builds to lower compute costs versus hyperscalers. The Information reports the plan ties to Brookfield's $100B AI infrastructure program anchored by a new $10B fund backed by partners including Nvidia and the Kuwait Investment Authority, with priority access to upcoming campuses in France, Qatar, and Sweden.
Financial Times reports that some developers are installing aeroderivative jet-engine gas turbines to provide tens of megawatts of on-site power for new hyperscale and AI data centres while they wait years for grid interconnections. The fast-to-deploy units let sites start training workloads sooner but rely on fossil fuels and underscore how grid bottlenecks are reshaping build plans.
Alphabet agreed to acquire longtime data center development partner Intersect for $4.75 billion to lock in power-secure capacity for Google Cloud and Gemini AI infrastructure. The deal would give Google control of 25 contracted colocation campuses across North America and Europe and fold a 1.5GW development pipeline into Alphabet, with closing expected in 2H25 pending regulatory review.
Rest of World mapped 8,808 operational data centers and found nearly 7,000 outside the optimal 18–27°C range, with all facilities in 21 countries located in areas averaging above 27°C. Hot-climate buildouts like Singapore, India, and the UAE face higher cooling costs, grid strain, and water demand, accelerating adoption of liquid, immersion, and hybrid cooling plus efficiency mandates.
FT reports that the AI data center boom is hitting severe US grid constraints: soaring electricity demand, long interconnection queues, and permitting hurdles are delaying hyperscale builds. Developers are exploring on-site gas or nuclear generation and lobbying for faster approvals to keep GPU capacity on track.
Financial Conduct Authority chief Nikhil Rathi said current financial rules are not built for foundation models and generative AI, arguing regulators need a cross-sector regime that looks at data, compute and cloud concentration before firms deploy new systems. He called for coordinated oversight with the CMA and ICO so banks and fintechs can prove model governance, avoid bias, and reduce reliance on a handful of hyperscale infrastructure providers.
Coatue's latest C:\Takes argues that AI's infrastructure build-out is still accelerating, with gigawatt-scale data centers needing far more generation and transmission. The note outlines three power options—new generation, regulated utilities, and independent power producers—and highlights hyperscalers paying premiums for nuclear-backed IPP deals like AWS/Talen, Microsoft/Constellation, and NextEra/Google as they chase reliable 24/7 supply.
Fox Business reports that AWS will invest up to $50 billion starting in 2026 to expand AI and supercomputing capacity for U.S. government agencies, adding nearly 1.3 gigawatts across its Top Secret, Secret, and GovCloud regions to deliver secure services like Bedrock, SageMaker, Trainium, and Nvidia-powered systems.
Bloomberg reports that Meta secured permission to buy and sell electricity directly so it can hedge soaring AI data center demand, using a trading desk to pair renewable supply with its hyperscale loads and cut exposure to wholesale price swings.
The Wall Street Journal says the wave of announced U.S. hyperscale campuses is outrunning real-world limits: the number of proposed AI data centers now dwarfs the available power and water, stalled projects are piling up, and the U.S. still hosts roughly 4,200 sites versus just 381 in China, concentrating the grid challenge domestically. Its six-chart package also notes that JPMorgan calculates AI products would need to throw off another $650 billion in annual revenue for investors to earn a 10% return, underscoring how easily today’s record capex could turn into stranded assets if deployments lag.
The Information reports that Blue Owl Capital’s technology lending arm is committing $3 billion to finance OpenAI’s Stargate supercomputer campus, providing structured capital for land, power, and GPU buildout so the startup can accelerate dedicated infrastructure outside of Microsoft’s shared Azure estate.
Financial Times warns that accelerating AI workloads are colliding with a shortage of utility-scale power, as U.S. hyperscalers face multi-year grid upgrade queues and scramble for cleaner generation deals while regions investing in new transmission and renewables pull ahead.
JPMorgan credit strategists estimate the generative AI data center build-out could require roughly $5 trillion of capital this decade, pushing hyperscalers, utilities, and developers to raise funds across investment-grade bonds, high-yield loans, securitizations, and municipal debt to cover land, power, and equipment expansion.
Deutsche Bank executives are evaluating hedging strategies to manage the bank's growing exposure to AI data centre financing, discussing options such as shorting baskets of AI equities or using derivatives to transfer risk as deal volume accelerates.
Bloomberg reports that Amazon Web Services asked federal regulators to intervene after Berkshire Hathaway Energy's PacifiCorp repeatedly missed deadlines to energize new AWS campuses in Oregon, forcing the cloud operator to buy replacement megawatts on the open market, delay planned AI capacity, and warn that wider grid constraints are undermining hyperscale expansion plans.
Reuters reports that Eaton agreed to acquire Boyd Thermal for $9.5 billion in cash to expand its sealing and thermal management products used in EVs, aerospace, and data centers. The deal, expected to close in the first half of 2026 pending approvals, adds liquid cooling and materials capabilities to support high-power electronics growth.
The New York Times reports that Saudi Arabia's Public Investment Fund is accelerating a multi-billion-dollar compute buildout across NEOM, Riyadh, and King Abdullah University of Science and Technology to position the kingdom as a supplier of advanced A.I. services. Officials are pairing long-term GPU procurement deals with U.S. chipmakers and venture stakes in Western start-ups while launching the Alat industrial platform to assemble servers and cooling gear locally. The push is intended to export Saudi-trained models and cloud capacity to partners in the Gulf, Africa, and South Asia as part of Crown Prince Mohammed bin Salman's strategy to diversify beyond oil.
CNBC highlights JPMorgan data showing AI-related capital expenditures added 1.1 percentage points to U.S. GDP growth in the first half of 2025 and propelled trillion-dollar tech valuations, even as small firms like Norton's Florist grapple with tariffs, weak consumer demand, and cost cutting. Surveys from KeyBank and Deloitte plus layoffs at Target, Starbucks, and Wyndham underscore that much of Main Street remains in survival mode despite $100 billion-plus AI infrastructure commitments from Nvidia, OpenAI and their partners.
Bloomberg reports that Energy Secretary Chris Wright has asked the Federal Energy Regulatory Commission to fast-track data center grid interconnection reviews, proposing a new 60-day deadline to connect AI-led projects that now wait years for approval. The push signals the administration's bid to relieve a mounting power bottleneck for hyperscale and AI developers by cutting federal permitting delays and coordinating utilities on expedited hookups.
Yahoo Finance reports that Bank of America Institute economist David Tinsley links a 3.6% year-on-year rise in average U.S. utility payments in Q3 2025 to surging electricity demand from AI data centers, noting that massive buildouts like the $500 billion Stargate program and hyperscaler expansions are straining grids and pushing pass-through costs onto households.
The Financial Times details how AI cloud and chip companies are buying up heavy-duty gas turbines so quickly that GE Vernova, Siemens Energy and Mitsubishi Heavy now face multi-year backlogs, pushing developers in the US and Europe to lock in equipment years ahead to power new dedicated natural-gas plants for data centers. The scramble is straining LNG supply chains, raising emissions concerns for regions from Texas to Germany, and intensifying permitting and geopolitical risks as governments weigh grid reliability against climate goals.
The New York Times reports that rapid AI data center expansion in Querétaro, Mexico and across Ireland is triggering community and political pushback over water consumption, grid stress, and land use, forcing operators to offer new infrastructure investments and conservation commitments.
The Wall Street Journal details how hyperscale and AI developers including OpenAI-Oracle's $500 billion Stargate campus in West Texas, xAI's Memphis buildout, Equinix sites, and Meta's Ohio expansion are installing their own natural-gas turbines and fuel cells to secure multi-gigawatt capacity amid transformer shortages, permitting delays, and U.S. grid congestion.
Bloomberg reports that Nvidia is partnering with Firmus Technologies on the A$4.5 billion Project Southgate, building 150MW of renewable-powered AI data centers in Melbourne and Tasmania that will run on GB300 chips and start coming online by April.
Financial Times says BlackRock's Global Infrastructure Partners and Abu Dhabi-backed AI platform MGX will acquire Aligned Data Centers in a $40bn deal, injecting capital to accelerate hyperscale campuses and AI-ready capacity across North and Latin America while relieving prior owners such as Macquarie Asset Management.
The Wall Street Journal reports that Poolside and CoreWeave will build the Horizon campus on more than 500 acres of the Mitchell family’s Longfellow Ranch in West Texas, pairing on-site natural gas generation with modular construction to bring 2GW of AI compute online in phases starting with 250MW by late 2026. Poolside, backed by Nvidia, is lining up a $2 billion raise at a $14 billion valuation and will tap CoreWeave GPUs by December to accelerate delivery while reserving another 500MW for future expansion through 2027.
Google will invest $15 billion over five years to build a 1GW AI data center hub in Visakhapatnam, Andhra Pradesh, partnering with Bharti Airtel and AdaniConneX for subsea cable connectivity. The campus will host Google's TPUs, Gemini models, and regional consumer services while scaling to multiple gigawatts for Asia-wide workloads.
Financial Times reports that hyperscale and AI operators such as Microsoft, Amazon Web Services and Google are demanding so much round-the-clock power for new campuses that US and European utilities are extending fossil generation plans even as they sign multi-gigawatt renewable contracts, underscoring how grid bottlenecks, permitting delays and long build timelines are slowing efforts to run data centres purely on clean energy.
WSJ reports that Elon Musk’s xAI is pouring billions into its Colossus supercomputer campus in Memphis, locking in long-term TVA power delivery measured in hundreds of megawatts, fast-tracking new substations and water upgrades with Memphis Light, Gas and Water, and promising thousands of high-paying technical jobs so the project can start training Grok models at scale by mid-2026.
Financial Times reports that European private capital groups including Oaktree Capital, Partners Group, and EQT have lined up roughly €17 billion of data centre acquisitions and developments to capitalise on surging AI infrastructure demand, signaling more competition for power-secure sites across the region.
Tom's Hardware reports AI hyperscale buyers are locking up most NAND, DRAM and HDD production, with OpenAI's Stargate deal securing up to 900,000 DRAM wafers a month, Samsung's next-gen V9 NAND nearly sold out, and TrendForce warning nearline HDD shortages that already forced Western Digital's 5-10% price hikes and Raspberry Pi memory surcharges.
The Wall Street Journal reports that Groq will deploy a dozen new GroqCloud data centers in 2026, including its first Asia-Pacific site, after raising $750 million at a $6.9 billion valuation to meet runaway demand for its AI inference stack while continuing builds in Helsinki and a $1.5 billion Aramco Digital campus in Dammam.
Axios Pro reports that Phaidra secured more than $50 million in Series B funding led by Collaborative Fund with Nvidia, Helena, Index Ventures and Sony Innovation Fund to accelerate its autonomous control agents that cut energy and cooling loads across hyperscale data centers while expanding global operator partnerships.
Bloomberg's nationwide LMP analysis finds wholesale electricity prices have spiked as much as 267% near dense AI data center clusters, forcing utilities in PJM, Oregon and Wisconsin to seek billions in new transmission and generation investments and shift more of those costs onto hyperscale customers to shield households from rate shock.
CenterPoint Energy outlined a $65 billion capital plan for 2026-2035 to harden its Houston-area grid, add high-voltage lines and substations for surging AI data center load, and expand electrification programs for fleets and industry, positioning the utility for faster rate base growth while improving resilience to Gulf Coast storms.
Bloomberg argues the AI compute crunch needs commodity-style markets, noting CoreWeave’s $50 billion valuation despite projected $650 million losses and Oracle’s $250 billion market-cap jump after disclosing its AI order book; standardized contracts and futures would give investors and customers price signals to finance new GPU campuses.
Ares Management is targeting more than $8 billion of near-term equity commitments to finance data center developments across London, Japan and Brazil, positioning the Los Angeles-based investor to capture fee growth from the accelerating AI infrastructure build-out.
Semafor reports that Microsoft and UAE-based G42 have yet to break ground on their $1 billion geothermal-powered data center in Kenya's Olkaria region, leaving the US-backed project languishing months before its 2026 launch target as partners struggle to justify demand for Azure services beyond government workloads and geopolitical momentum shifts away from Africa.
OpenAI has lined up agreements for five U.S. locations to host its Stargate AI supercomputing campus, lifting the project's expected outlay to about $400bn and signaling a massive infrastructure build to support future model training.
Microsoft is testing microfluidic prototypes that pump small amounts of dielectric fluid through channels etched directly into server processors and AI GPUs, a systems technology effort overseen by Husam Alissa to cut cooling power needs and boost rack density for Office cloud and AI workloads.
Nvidia will invest up to $100 billion in OpenAI to finance a 10-gigawatt rollout of GPU data centers, equivalent to between 4 million and 5 million of its accelerators, making the chipmaker the preferred supplier for the lab's infrastructure build. The phased program begins bringing Vera Rubin systems online in the second half of 2026 and complements OpenAI's collaborations with Microsoft, Oracle, SoftBank and the Stargate project as Nvidia deepens its broader AI investments.
Khazna, which now controls more than 70% of the UAE's data center capacity, is rolling out AI-optimized campuses including a 100 MW GPU-focused Ajman site and a 32 MW facility in Masdar City while deploying liquid cooling to manage heat loads. Backed by G42 partnerships with Nvidia, the operator is pushing into Turkey, Saudi Arabia, and Europe to make the UAE a regional hub for AI and cloud services.
Rest of World reports Microsoft is running its new Colón, Querétaro data center on seven gas generators for 70% of its energy for 12 hours daily through mid-2025 because Mexico's grid won't connect the site until 2027, driving emissions equal to 54,000 households. Mexico already hosts 150 sites after $7B of investment since 2020, yet faces a 48,000 MWh power deficit and 1.5GW demand by 2030 as more than $18B in planned hyperscale projects outpace transmission upgrades.
Microsoft unveiled its Fairwater AI datacenter in Wisconsin, a 315-acre, 1.2 million sq ft campus engineered as a single supercomputer with flat networking to run hundreds of thousands of NVIDIA GPUs—delivering 10x today’s fastest supercomputer while requiring 46.6 miles of deep foundations, 26.5 million pounds of steel, and 120 miles of medium-voltage cabling. The company said identical Fairwater facilities are under construction across the US and announced nScale partnerships in Narvik, Norway, and Loughton, UK to extend tens of billions of dollars in global AI infrastructure investments.