Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

How to build AI factories faster

As demand for AI capacity surges, construction has become the bottleneck.

[Photo: Getty Images]

Profile photo of Aviad Almagor

Vice President of Technology and Innovation · Trimble

AI is becoming part of nearly every product, workflow, and business process. As organizations move beyond experimentation and deploy AI at scale, demand for computing capacity is behind one of the largest infrastructure investment cycles in history. Goldman Sachs expects a combined $5.3 trillion in capital spending from 2025 through 2030 by the large technology companies that are leading the buildout.

Meeting this demand will require more than expanding traditional data centers. Traditional data centers are digital filing cabinets, built primarily to store and move information, but AI factories are emerging as the manufacturing plants of the intelligence economy. Unlike general-purpose data centers that support a wide range of computing workloads, AI factories are optimized for the high-density, compute-intensive workloads required to train and run AI at scale. These purpose-built facilities integrate computing, networking, power, cooling, and software into one system, transforming energy and computing capacity into the tokens that power AI models, agents, and applications.

The challenge for technology companies investing in AI infrastructure is no longer simply developing more capable AI, but building the physical infrastructure required to put that intelligence to work.    

Construction is the bottleneck

The race to build AI infrastructure has collided with a physical reality: some of the world’s most advanced technology still depends on fragmented and manual construction processes.

Building an AI factory requires thousands of interconnected decisions across owners, designers, engineers, contractors, utilities, equipment suppliers, and field teams. These decisions cover everything from site preparation and structural systems to electrical distribution, liquid cooling, networking, and commissioning.

These projects often have highly compressed build schedules. Disconnected information and manual handoffs, such as transferring design changes, schedules, and project updates between teams and systems, can slow decision-making and create risk. A design change that does not reach the field, a delayed transformer, or an incorrectly installed cooling component can prevent valuable computing capacity from coming online—and the financial consequences can be substantial.

Kaizen Institute estimates that a one-month delay on a 60-megawatt data center can cost more than $14 million in lost lease revenue, additional overhead, and contractual penalties. The amount varies by project, but the broader lesson is clear: capital alone is not enough to bring an AI factory online. Speed to operational capacity has become a competitive advantage.

Beyond the build

Construction is not the only constraint these AI factories face. Power availability, grid connections, permitting, equipment supply, and community acceptance all determine whether a project moves forward.

Once a project is approved, the ability to design, construct, and commission it becomes critical. The construction industry already knows how to deliver highly complex facilities. The challenge is doing so with greater speed, consistency, and certainty.

Here are four changes technology companies should prioritize with their construction partners to bring AI factories online more quickly.

1. Standardize delivery

    Technology companies cannot afford to reinvent how every AI factory is designed and delivered. Common reference architectures, repeatable components, and validated workflows allow teams to build on proven approaches.

    For instance, NVIDIA created NVIDIA DSX, a playbook for creating AI factories. It provides a reference architecture that connects computing, networking, storage, power, cooling, and facility infrastructure. Standardization does not mean making every facility identical but creating a reliable foundation that can be adapted without starting from zero each time.

    However, standardization cannot stop with the technology stack. It must extend across planning, site preparation, design, procurement, construction, and commissioning because decisions made at each stage affect the cost, schedule, and performance of the completed facility.

    2. Create a continuous digital thread

    A continuous digital thread benefits any construction project, but it matters most in an AI factory. AI hardware evolves rapidly so facilities must constantly adapt to new processors, higher rack densities and changing cooling demands.

    A digital thread does more than just track building progress. It carries data through construction, commissioning, and operations, and it creates a living digital twin that links the physical infrastructure directly to compute performance. That gives facility managers a reliable baseline for swapping out server racks and rebalancing power loads as needs change. This digital representation reduces manual errors during the initial build and ensures the facility enters operations as a highly adaptable machine ready to support the next generation of AI workloads.

    3. Simulate before building

    Simulation must become part of every major project decision. This is especially critical for AI factories, where the ultimate goal is optimizing token generation. Simulating thermal behavior is essential for creating optimal mechanical, electrical, and plumbing designs.

     An AI factory produces fewer tokens and operates less effectively when the thermal environment isn’t optimized, resulting in lost revenue. By simulating cooling performance and resolving conflicts in digital environments before construction begins, teams avoid expensive rework and ensure the facility is primed for maximum output. After construction is complete, simulations help owners assess maintenance, upgrades, and future expansions.

    4. Industrialize construction

    Industrialization applies manufacturing principles to construction. Part of this is prefabrication: building components offsite in controlled factory conditions, and then shipping them to the site for assembly. But the bigger gain comes from pairing prefabrication and physical AI, meaning robotics, autonomous heavy equipment, and intelligent machine control can transform chaotic construction sites into orchestrated, predictable production systems.

    Factory-built components arrive ready to install, and machines place and connect them with a precision that manual methods can’t consistently match. The goal isn’t just to bridge the labor gap, but to bring machine precision and performance into physical builds.

    Next steps

    Construction efficiency is no longer just a real estate concern. It determines how quickly new computing capacity comes online and how much it costs to run.

    For technology leaders, this means AI infrastructure should not be treated as a conventional construction program. AI factory owners and developers should require repeatable delivery architectures, a continuous digital thread, and performance-based commissioning from the beginning of every project.

    Faster and more predictable delivery can help expand the supply of computing capacity and make AI more accessible. As businesses build products, workflows, and operating models around AI, affordable access to compute will become as strategically important as access to electricity and communications infrastructure.

    Factories powered the Industrial Revolution by transforming raw materials into physical goods. AI factories will power the intelligence economy by transforming computing capacity into useful intelligence. Success will depend not only on advances in AI models, processors, and energy infrastructure, but also on technology companies working with the construction industry to turn those advances into reliable, operational facilities.

    Aviad Almagor is vice president of technology and innovation at Trimble.