AI is reshaping the conversation around data center performance, sparking debate everywhere from boardroom budget reviews to trade press to a recent invitation-only working session[1] at Trellis Impact 26. Across these spaces, a common question resounds: how can operators get more out of the infrastructure they're building in a sector growing too fast for yesterday's assumptions?
The scale helps explain the heightened attention. Capital spending on AI infrastructure is on pace to top $1 trillion by 2027[2], making it one of the largest U.S. infrastructure buildouts in history and surpassing annual investment in oil and gas. At the same time, Berkeley Lab estimates data centers could consume 11.8% of U.S. electricity by 2030[3], even as EPRI suggests the number could go as high as 17%[4]. It’s no wonder regulators are taking note, as with proposals like New Jersey's Data Center Fair Share Act[5].
For owners and operators, these converging trends are increasingly exposing a practical challenge: keeping pace with fast-moving technology demands alongside long-horizon facility planning.
The velocity mismatch: Managing a tale of two lifecycles
The software and silicon technology driving data center compute are evolving on a hyper-accelerating, compounding curve. Examples abound, from Nvidia’s well-known target to release a new GPU every year[6], to the trouble-riddled Stargate project[7], where the timelines for constructing the data center and its systems fell badly out of sync with the GPUs meant to occupy the space.
This pace collides directly with facility infrastructure, which requires multi-year planning and capital commitments that can’t pivot at the speed of a software update. Core computing equipment — CPUs, GPUs, and localized power loops — has a drastically short shelf life, while the outer building envelope is built to last 30 years or more. Servers, cooling strategies, power densities, and operational requirements can each evolve several times during the same 30-year span.
Today, the gap is widening further as workloads diverge by compute type. Traditional cloud facilities run at lower power densities on standard CPU tasks, getting by just fine on basic airflow cooling. AI-focused data centers[8], by contrast, pack dense, power-hungry GPU clusters that generate far more heat than air-based cooling can manage, pushing owners toward complex liquid cooling systems like oil submersion or direct-to-chip liquid cooling systems. That hardware adds real weight, with newer high-density AI racks weighing up to 1.5 metric tons[9] — a load that assumes a building engineered for this generation of compute.
Most buildings weren’t. Roughly 70%[10] of data center capacity sits within pre-existing facilities, including older data centers built for earlier generations of hardware and non-data center spaces, such as commercial offices, being converted within the past 25 years. That means a significant share of the infrastructure supporting the next generation of compute was built around very different assumptions about power density, cooling, and equipment.
Strategy starts with the site
The right approach depends on the physical and operational starting point of each site.
A greenfield project has the most room to plan ahead. Electrical architecture, redundant systems, cooling strategies, and equipment selection can all be shaped before construction begins, setting the trajectory for operating costs and flexibility over the building's life.
An existing modern facility is often more of a tuning opportunity. Performance monitoring can identify underused capacity, while targeted upgrades can prepare existing infrastructure for higher-density workloads.
Older facilities and retrofit candidates generally require a more comprehensive assessment before higher-density workloads can be introduced. The key consideration is how much of the existing infrastructure can support the intended workload and where targeted upgrades may be needed.
Finding the highest-value efficiency gains
Across those starting points, a consistent set of factors helps determine where a facility can support additional performance. Electrical capacity, roof loading, cooling limits, asset condition, operational data, and the economics of potential upgrades all shape where investment can deliver the greatest return.
This begins with a few key questions:
- Does available electrical capacity align with expected AI-driven growth?
- Where are cooling systems approaching operational limits?
- Which assets are best optimized, upgraded, or replaced?
- What operational insights are needed to benchmark performance and prioritize investments?
- Which efficiency improvements offer the strongest return based on the facility’s current needs?
The answers can reveal opportunities at very different levels of investment. For example, airflow optimization and containment can often unlock additional capacity within a facility’s existing footprint. Building management systems and continuous performance monitoring help catch equipment drift early, before it affects uptime or energy consumption, and inefficient or end-of-life mechanical systems can typically be modernized in phases to support higher-density workloads as they’re needed.
The highest-ROI opportunities often extend beyond individual equipment upgrades. Comprehensive multi-measure combinations provide utility assistance or financing incentives when multiple cost savings measures are installed during a single project. Preventive maintenance, strategically deferred capital projects, high-value design, warranty capture, and construction quality assurance can each improve performance and protect investment when applied according to a facility’s specific needs.
Efficiency that keeps pace
Data center efficiency increasingly depends on planning for two timelines at once — the decades a facility is built to last and the much shorter cycle of change inside it. The strongest strategies account for both, using today’s infrastructure more effectively while preserving the flexibility to support the next generation of compute.
Learn more by starting a conversation with Mantis Innovation today.
Key Takeaways
- Data center operators face a critical velocity mismatch between facility buildouts and tech cycles. Balancing 30-year physical building lifespans with hardware evolving annually requires planning for two timelines at once.
- AI workloads are overwhelming legacy infrastructure with intense power and thermal demands. Operating high-density GPU racks in older facilities requires shifting to specialized liquid cooling and structural reinforcement.
- Targeted airflow optimization and continuous monitoring offer high-ROI early wins. Simple adjustments unlock immediate cooling capacity and prevent equipment drift without massive facility overhauls.
- Tailored site strategies ensure capital is invested for maximum return. Whether managing greenfield builds, modernizing existing sites, or retrofitting legacy spaces, solutions must match physical site readiness.
- Phased upgrades and multi-measure incentive packages protect long-term flexibility. Combining mechanical upgrades with utility incentives helps facilities scale efficiently while managing capital costs.
FAQs
Q: How are AI workloads impacting legacy data center infrastructure?
A: High-density GPU racks generate far higher power and thermal loads than standard servers. Since most facilities were built for earlier hardware, owners and operators must retrofit legacy spaces with advanced liquid cooling and structural reinforcement.
Q: How do fast hardware updates affect data center capacity planning?
A: Rapid 1- to 3-year GPU release cycles collide with multi-year facility buildouts and 30-year building lifespans. This forces owners and operators to plan for two timelines at once, designing flexible infrastructure that can accommodate multiple generations of compute over time.
Q: How can operators increase data center cooling efficiency?
A: Implementing airflow containment, deploying real-time performance monitoring, and upgrading inefficient mechanical systems in phases can unlock significant cooling capacity within an existing footprint.
Q: How do data center efficiency strategies differ by site type?
A: Greenfield projects allow owners to design custom electrical and cooling architecture from day one. Existing modern facilities benefit from operational tuning, while legacy retrofits require comprehensive multi-measure upgrades paired with utility incentives.
