# Lancium, Nvidia Partner on Gigawatt-Scale AI Data Centers > Source: > Published: 2026-08-25 12:33:39+00:00 # Lancium, Nvidia Partner on Gigawatt-Scale AI Data Centers Lancium says it has 4 GW under lease and 15+ GW of powered land as Nvidia partnership advances its grid-responsive AI data center model. Lancium is partnering with Nvidia to deploy the chipmaker’s AI factory technology across a portfolio that the Texas infrastructure developer says includes 4 GW of leased capacity and more than 15 GW of powered land in development. The partnership anchors Nvidia’s role in designing and deploying AI infrastructure across Lancium’s portfolio. Lancium will use Nvidia’s DSX reference designs and power-management technologies at its campuses, including systems intended to increase compute density and adjust AI factory power consumption in response to grid conditions. Nvidia is also making a strategic investment in Lancium, which is backed by Blackstone. The companies did not disclose the investment amount. Lancium said its campuses will serve as deployment sites for Nvidia’s full-stack AI factory platform, including accelerated computing, networking and software. The companies said the arrangement will give cloud providers, infrastructure developers and AI companies in Nvidia’s ecosystem access to power-ready capacity at gigawatt scale. ## 4 GW vs. 15+ GW The scale of Lancium’s portfolio is central to the announcement, but its two headline figures represent different stages of development. Lancium says it has 4 GW of capacity under lease and more than 15 GW of powered land in development. The company did not identify the projects behind those figures or provide timelines for bringing the full portfolio online. Lancium’s publicly announced projects include its [1.2 GW Clean Campus in Abilene](/build-design/crusoe-expands-abilene-ai-campus-with-new-900mw-ai-factory-for-microsoft), where Crusoe is developing AI data center capacity; a 1 GW campus in Childress County, also with Crusoe; and a new campus near Turkey in Hall County, where QTS will design, build and operate the data center buildings. Lancium owns the campuses and is responsible for their electrical and civil infrastructure. At the Hall County campus, Lancium and QTS said they will fund all energy infrastructure improvements, with Lancium planning to bring its own power to the site using battery storage and solar resources. The projects are already attracting billions of dollars in planned investment. QTS and Lancium said the Hall County campus is expected to bring more than $10 billion in capital investment to the region. Lancium also closed a $600 million debt financing package in 2025 to advance its Clean Campus strategy, beginning with the Abilene site. At Abilene, a 2024 joint venture between Crusoe, Blue Owl Capital and Primary Digital Infrastructure was established with $3.4 billion to fund more than 200 MW of build-to-suit data center capacity at the Lancium campus. Those figures are not additive measures of Lancium’s investment. They represent different projects, financing structures and stages of development. But “powered land” does not necessarily represent load that can be delivered to the grid on the same basis as leased capacity, said Neil Osnato, founder of Persistence Analytics Group. “Four gigawatts described as ‘under lease’ suggests a materially stronger commercial commitment than 15+ GW of ‘powered land in development,’” Osnato said. For the larger figure, the important questions are how much capacity has an executed interconnection path, what infrastructure has been studied and is required, when each tranche can energize and how much customer demand is committed behind it, he said. “A large development pipeline should not automatically be read as 15 GW of executable load,” Osnato said. Developers are seeking to secure power for AI campuses years before the facilities are fully built. Land, generation resources and an interconnection path can establish a development position without creating an equivalent amount of load that is ready to energize. ## Grid-Responsive Load Lancium is also pitching power flexibility as part of the value of its campuses. Lancium will use Nvidia DSX MaxLPS to improve how power is allocated to GPUs, potentially allowing more GPUs to operate within the same facility power budget, said Matt Kimball, vice president and principal analyst for datacenter technologies at Moor Insights & Strategy. The “up to 40%” figure represents an ideal-case scenario rather than a guaranteed improvement, Kimball said. The underlying benefit comes from avoiding the need to reserve each GPU’s maximum rated power when its actual workload typically requires less. If a GPU is rated for 1 kW but is using 600 watts, for example, conventional power allocation could leave 400 watts of capacity unused. DSX can allocate power based on actual workload requirements, allowing that capacity to be used elsewhere in the facility, he said. “It’s the ability to better utilize the incoming power” that matters more than the 40% figure, Kimball said. At a 1 GW facility, even a 20% improvement in power utilization would represent 200 MW of capacity that could potentially support additional GPUs, while a 10% improvement would represent 100 MW, Kimball said. Actual results would depend on workloads and operating conditions. Kimball said the approach is directionally significant because it connects compute workloads more closely to the power available to the data center, rather than treating each GPU’s maximum rated consumption as a fixed requirement. For a gigawatt-scale campus, the grid value would depend on what that flexibility can deliver in practice, Osnato said. A useful resource would need to reduce or reshape consumption when the grid needs it, with a response that is fast, measurable, dependable and available during the relevant system conditions. “The question is not whether the software can technically move load; it is how much load can move, for how long, how often, under whose control, and what operating constraints remain,” he said. That could affect how utilities [plan for large AI loads](/energy-power-supply/how-ercot-s-post-crez-bet-is-reshaping-ai-infrastructure). A genuinely flexible gigawatt-scale customer presents a different planning problem from an inflexible one, Osnato said. But utilities should not assume that technically available flexibility can substitute for generation or transmission investment. If a utility relies on a data center’s flexibility to avoid or defer infrastructure, the capability needs to be measurable, available when needed and subject to revalidation as the campus, workload mix and operating economics change. Lancium said its campuses will combine grid interconnections with behind-the-meter generation and energy storage. Its power-management systems are intended to allow the data centers to respond to grid conditions while maintaining the compute density required by AI workloads. ## Nvidia’s Role The Nvidia partnership gives Lancium a technology platform to deploy across its growing data center portfolio, while giving Nvidia customers and infrastructure partners another route to large-scale AI capacity. “AI factories are the essential infrastructure of this new industrial era,” Nico Caprez, Nvidia’s vice president of global AI infrastructure growth, said in the announcement. Michael McNamara, Lancium’s CEO and co-founder, said the company had spent years assembling the power, land and infrastructure expertise needed to develop AI data centers at gigawatt scale. The companies did not disclose which Lancium campuses will use Nvidia technology. For Lancium, the more consequential test will come as those projects move from development into interconnection and operation. “Announced capacity is not executable capacity, and technically flexible load is not the same as dependable grid capacity,” Osnato said. If Lancium can demonstrate both durable load and verifiable flexibility, its model could provide a meaningful grid benefit, he said. The evidence, however, should follow the projects from announcement through interconnection, energization and operation rather than being assumed at the outset.