Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Steve Altizer, President & CEO, Compu Dynamics
If 2025 was the year the industry woke up to the physical reality of AI, 2026 will be the year operators stop trying to fit it into yesterday’s generic boxes.
Over the last two years, AI has quietly but ruthlessly rewritten the rules of infrastructure. What used to be a predictable environment designed around steady-state cloud workloads, conventional airflow, and familiar rack densities has given way to something far more volatile. Companies are no longer just building “data centers” in the traditional sense; they are constructing specialized industrial plants designed to manufacture tokens and insights.
As the industry looks toward 2026, the “big empty box” approach to design is rapidly running out of runway. Operators who continue to build generic shells hoping to adapt them for AI later will find themselves holding assets that are technically obsolete before the concrete cures. Instead, the industry is entering a new architectural era defined by density, fluid dynamics, and purpose-built facilities.
Here is how this shift is playing out in 2026.
1. The Compute Floor Becomes a “Data Fab”
For the last decade, the prevailing design philosophy was straightforward: build a flexible, low-cost shell and fill it with as much IT space as possible. Roughly 80% of the floor area was white space, while 20% was dedicated to support systems. That model worked beautifully when global average rack densities hovered around 8kW, according to investment management company JLL.
In 2026, that ratio will flip. In the high-performance compute environments some companies are building today, the IT footprint is becoming the minority. The supporting plant – power distribution, switchgear, cooling manifolds, and the structural backbone – is consuming the majority of the site’s footprint.
Consider the dramatic shift in space efficiency. A 5 MW facility designed around traditional low-density racks might require 25,000 to 50,000 square feet of white space. That same 5 MW, when optimized for modern 150-200kW AI racks, can be executed with as little as 4,000 to 6,000 square feet of white space. In modular configurations, the IT modules themselves occupy roughly 1,500 square feet. The rest of the site starts to look less like an occupied office building and more like an industrial plant: pipes and conduits everywhere, structured utility spines, and chiller modules lined up with industrial purpose.
2. Liquid Cooling Moves from “Pilot” to “First-Order Constraint”
In 2025, liquid cooling was the industry’s favorite talking point. Everyone was piloting it, but few had integrated it at scale. In 2026, liquid cooling will stop being an experiment and start being the primary design constraint.
Three distinct paths are solidifying. In-rack cooling distribution units (CDUs) offer simplified deployment, room- or row-based distribution systems with stainless steel piping enable massive scale, and liquid-to-air heat exchangers bridge the gap for legacy facilities. The most significant shift in 2026 will be the realization that high-density liquid cooling cannot be retrofitted at the 11th hour. It must be baked into the facility’s DNA, especially as roadmaps from major chip manufacturers like NVIDIA point toward rack densities approaching 600kW by 2027, Introl reported earlier this quarter.
The biggest mistake operators can make is assuming that any competent mechanical contractor is ready for this work. The tolerances are tight, the fluid quality requirements are stringent, and the failure modes are unforgiving. The industry will see a “flight to quality,” where specialized firms that understand the industrial nature of these systems pull away from generalist contractors.
3. Modular Goes Strategic, Not Just Tactical
Historically, modular infrastructure was a stopgap – a way to add capacity quickly at the edge. That changes now. With compute generations arriving every six to 12 months, traditional stick-built construction, which often takes 18 to 24 months, is simply too slow. By the time a conventional building is permitted and built, the hardware it was designed for is two generations old.
In 2026, modular infrastructure will become a strategic imperative for hyperscalers. Entire campuses will be built where the “building” is minimal and capacity is delivered via pre-engineered IT, power, and cooling modules fabricated off site. This isn’t just about speed; it’s about density optimization and cost. According to Cushman & Wakefield’s “2025 Data Center Development Cost Guide,” traditional data center construction costs in major markets often range from $12 million to $15 million per megawatt, well-executed modular campuses can land closer to $7 million to $8 million per megawatt because every element is optimized for density rather than aesthetic uniformity.
4. The Emergence of the “Blended Footprint”
The industry isn’t going to stop building traditional data centers entirely. Instead, 2026 will be the year of the blended footprint.
Single campuses will combine stick-built, multi-tenant halls for general cloud workloads alongside highly specialized, purpose-built AI blocks and dense arrays of modular infrastructure. These different environments will share common utility spines and operational playbooks, but their physical forms will be radically different.
Existing facilities will also find new life in this ecosystem. They won’t be the primary training grounds for massive LLMs, but they will become inference nodes and hybrid environments. The smart money will be on creatively integrating modular and liquid solutions into these legacy sites to unlock stranded capacity without rebuilding the plant.
5. Engineered Infrastructure as Competitive Advantage
Historically, white space planning revolved around airflow and capacity. In the AI era, that mindset is insufficient. White space must be treated as an engineered, integrated performance layer where electrical, cooling, network, and computing interact in real time.
An effective engineered performance layer exhibits several characteristics: adaptive electrical distribution that can serve both conventional and high-density zones; zoned thermal strategies that allow air, hybrid, and liquid cooling to coexist; and off-site fabrication to compress schedules.
Viewed this way, modern AI facilities have more in common with a petrochemical plant than with the early cloud facilities the industry grew up building. That is not a metaphor – it is where the design language is actually heading.
Stay Ahead of the Compute Curve
AI is forcing the industry to rethink nearly every assumption that guided facility design over the last two decades. Generic templates and “good enough” flexibility will not survive the next wave of demand.
The operators who thrive will be those who accept that infrastructure must now adapt to the pace of compute, not the other way around. Advantages will be ushered in from those with decades of experience designing, building, and servicing facilities from 20kW labs to 100-plus MW campuses. Today, teams are deploying first-of-their-kind liquid cooling systems in hyperscale facilities and designing modular AI campuses measured in gigawatts.
The AI curve is steep and getting steeper. The challenge for the industry is not to flatten it, but to stay ahead of it.
##
ABOUT THE AUTHOR
Steve Altizer – President & CEO, Compu Dynamics
Steve Altizer has nearly four decades of experience building some of the world’s most sophisticated government and commercial facilities. In 2002, Altizer founded the Andrew Browning Group (now Compu Dynamics). Prior to that, he served as a senior executive with several nationally ranked general and mechanical contractors.
Throughout his career, he has been a student of and thought leader in the technology and science behind today’s mission critical environments. This interest has naturally led to an affinity for clients whose requirements drive them toward facilities that are smart, clean, safe, reliable, and secure. His focus for the last 25 years has been exclusively on data centers ranging in size from 20kW to over 100MW.
Altizer earned a BS in Mechanical Engineering and an MBA, both from the University of Virginia.
