The race to deliver AI compute is forcing data center developers to rethink the slow, sequential nature of traditional builds. A new partnership between JLL and InfraPartners is built around a simple premise: overlap more of the work, standardize more of the process, and get GPUs producing “first token” faster.
InfraPartners, an ODM manufacturer of prefabricated data centers, has formalized a collaboration with JLL’s data centers and critical environments team to pair prefab manufacturing with global project management, site work and operations support.
“Nothing else matters, right? Those GPUs need to be deployed quickly, and they need to start generate revenue and outputting computing and intelligence,” said Michalis Grigoratos, CEO and co-founder of InfraPartners.
Nothing else matters…those GPUs need to be deployed quickly.
Grigoratos said the combined approach can remove months from the critical path by running parallel tracks: JLL advances pre-development and site work while InfraPartners manufactures major components off-site. “We’re able to remove the sequential phase from a traditional project,” he said, describing a model where site diligence, surveys and validation can happen while the asset is being built in a factory. Based on projects InfraPartners has already been involved in, he said the model can “take away basically anywhere between 13 and 22 months on typical hyperscale development,” which he characterized as “about 48% acceleration.”
Matt Landeck, division president for data centers and critical environments at JLL, framed the partnership as a complementary fit between a product company and a service provider. “We’re a service provider. We’re never getting into the prefab business, and Michaelis is in prefab and really not looking to go into being a service provider,” Landeck said. “When you eliminate the friction around potentially where we may be competing, that makes it quite easy…to work together.”
22 Months
Both executives pointed to repeatability as a key advantage. Landeck said the goal is knowledge retention around a stable “basis of design,” enabling a “rinse and repeat approach” that can compress timelines further over multiple deployments. Grigoratos emphasized the operational upside of not having to retrain teams on every project. “We operate from a blueprint, and we don’t have to explain the same thing over and over again,” he said. “It makes me more profitable… and it makes JLL more profitable because they don’t have to learn every single project, every single time.”
We’re able to remove the sequential phase from a traditional project.
The conversation comes as grid constraints and land availability push AI-grade builds into places that would have been unlikely candidates a decade ago. “From a JLL perspective, really it’s power and land is where we’re at,” Landeck said, adding that the challenge for emerging markets is building an ecosystem that can absorb construction surges and support long-term operations staffing.
Labor remains a looming constraint, particularly in remote power-rich areas such as West Texas. Grigoratos described a scenario where “a gigawatt AI factory” could require “anything between five and 7000 people on site” at peak. InfraPartners’ strategy is to “centraliz[e] the majority, 80% of the work in a factory environment,” he said, while leaning on JLL’s scale to forecast staffing needs earlier: “Say, hey, in three months, I need 200 people in this location.”
Capital is still flowing into the sector, Landeck said, even as higher rates change refinancing math. But both flagged a newer pressure point: AI hardware and power densities are evolving so fast that real estate and infrastructure may become technologically obsolete before they are physically depreciated. That mismatch, they said, is likely to become a bigger board-level issue as AI deployments scale.
