ParadisLabs flags AI supply chokepoints

ParadisLabs published a thread on X on May 17 arguing that investors and operators still understate the physical bottlenecks behind artificial intelligence build-outs. The post focused on three constraints — power, cooling and minerals — rather than chips or model software. It cited GE Vernova as an energy-linked example and Daikin as a cooling supplier exposed to data-center demand. The thread comes as governments, utilities and equipment makers warn that AI growth is colliding with slower-moving industrial supply chains. The International Energy Agency said in its 2025 “Energy and AI” report that “there is no AI without energy,” framing data centers as a rising source of electricity demand. The U.S. Department of Energy said in a 2024 report that U.S. data-center electricity use rose to 176 terawatt-hours in 2023 from 58 terawatt-hours in 2014, and could reach 325 to 580 terawatt-hours by 2028. ### Why did ParadisLabs focus on power before chips? The IEA said global electricity generation needed to supply data centers is projected to rise from 460 terawatt-hours in 2024 to more than 1,000 terawatt-hours in 2030 in its base case. That makes grid access, transformers, switchgear and generation capacity central constraints on new AI capacity, not just semiconductor availability. GE Vernova says it supplies electrification and power-generation equipment across grid and industrial systems. (iea.org) ParadisLabs pointed to GE Vernova because transformer and grid-equipment lead times have become a recurring concern in AI-related power discussions, according to the account’s earlier published ideas tracked by Buzzberg. The U.S. Department of Energy’s July 2024 recommendations on powering AI and data centers said connection requests for hyperscale facilities of 300 megawatts to 1,000 megawatts with lead times of one to three years were stretching local grid capacity. (iea.org) That report described power delivery as a current and medium-term constraint for AI applications. (gevernova.com) ### Why is cooling turning into its own supply-chain issue? Daikin said on January 28 that it would showcase advanced data-center and commercial cooling technologies at the 2026 AHR Expo. On April 10, Daikin Holdings Singapore said it signed a memorandum of understanding with Delta Electronics to collaborate on coolant distribution unit solutions for AI and high-performance computing data centers across the ASEAN-Oceania region. (energy.gov) Lawrence Berkeley National Laboratory said in a June 2025 report that cooling supply chains face vulnerabilities tied to high-efficiency equipment, critical mineral inputs and energy-security risks. The report said diversification is needed for both equipment manufacturing and mineral sourcing. A U.S. Energy Department presentation posted in 2025 said air cooling still dominates about 90% of existing non-AI hyperscale and colocation facilities, but higher-density compute is pushing more systems toward liquid cooling at the chip level. (daikin.com) That shift broadens the list of suppliers that matter when AI capacity expands. (eta-publications.lbl.gov) ### Which minerals sit underneath the power-and-cooling story? The IEA said China announced export controls on gallium- and germanium-related items effective August 1, 2023. In its 2025 critical minerals outlook, the agency said China later restricted exports of gallium, germanium and antimony to the United States in December 2024, followed by further announcements in early 2025 covering tungsten, tellurium, bismuth, indium, molybdenum and seven heavy rare earth elements. (betterbuildingssolutioncenter.energy.gov) The IEA said rare earth supply chains remain among the least geographically diversified in critical minerals. It also said new export controls on strategic materials create risks for semiconductors, energy systems and other industrial sectors that depend on concentrated processing capacity. Those mineral constraints reach beyond chip fabrication. (iea.org) Lawrence Berkeley National Laboratory said cooling equipment supply chains also depend on critical mineral inputs, linking HVAC availability to the same upstream concentration risks affecting other parts of the AI build-out. ### Why would this matter to resorts and other physical operators? (iea.org) The Department of Energy said rising data-center demand is part of a broader increase in electricity consumption that is testing U.S. infrastructure planning. Resorts, campuses and other property operators adding AI-heavy systems, digital controls or on-site compute face the same practical dependencies on reliable power delivery and thermal management, even if they are not building hyperscale facilities. (eta-publications.lbl.gov) The IEA said electricity demand growth is being supported by AI, data centers and wider electrification across the economy. That means operators competing for grid upgrades, backup equipment and cooling systems are entering supply chains already under pressure from larger industrial buyers. May 12 marked GE Vernova Chief Executive Scott Strazik’s scheduled appearance at Bernstein’s Strategic Decisions Conference, according to the company’s website, while Daikin on May 12 published results for the fiscal year ended March 31, 2026. (energy.gov) Those company updates, along with future IEA and DOE releases on power demand and critical minerals, are the next public markers for whether the bottlenecks highlighted by ParadisLabs are easing or deepening. (gevernova.com) (iea.org)