Blog

The 120kW Rack Challenge: How AI Is Rewriting Data Center Power Rules

For decades, data center managers worked within comfortable boundaries: 8-15 kilowatts per rack, air cooling, predictable power distribution. That era is over. In 2026, AI workloads are forcing infrastructure teams to support 60-120kW per rack, with some deployments exceeding 300kW. This isn’t a gradual evolution; it’s a fundamental disruption that’s exposing the limitations of traditional data center design.

What This Means for Infrastructure Teams

Power distribution is becoming the bottleneck. Traditional 208V circuits can’t deliver enough amperage. It’s now normal to see three, four, or even six rack PDUs in each rack, with multiple 400V 60A power drops becoming standard in AI data halls. NVIDIA is preparing for this with an 800 VDC power architecture designed to support 1MW IT racks starting in 2027.

Cooling can no longer rely on air. Once rack densities exceed 30-40kW, air cooling becomes impractical. Physics dictates a shift to liquid cooling, direct-to-chip systems, immersion cooling, or rear-door heat exchangers. The liquid cooling market reflects this urgency: projected to grow from $6.6 billion in 2026 to $38.4 billion by 2033, driven almost entirely by AI adoption.

The Retrofit vs. New Build Decision

Many organizations face a choice: retrofit existing facilities or build greenfield AI-optimized data halls. Retrofitting is possible but expensive. It requires electrical upgrades, cooling infrastructure replacement, and often structural modifications to handle coolant distribution units (CDUs) and liquid loops. Budget 18-24 months and significant capital expenditure.

New builds offer flexibility to design for 100kW+ racks from day one: 800 VDC power, integrated liquid cooling, and modular scalability. But they also require upfront investment and longer timelines.

The Smart Hands Implication

High-density AI infrastructure changes operational requirements. Liquid cooling systems need regular monitoring: flow rates, pressure differentials, coolant chemistry, leak detection. Rack PDUs must be verified and balanced. Thermal imaging becomes routine maintenance, not troubleshooting.

If you’re planning AI deployments, start with these steps:

  1. Audit your power capacity. Understand your limits today.
  2. Evaluate cooling options. Air cooling won’t work for 60kW+ racks.
  3. Engage specialized partners early. AI infrastructure deployment requires expertise in liquid cooling, high-density power, and GPU hardware.
  4. Plan for lifecycle management. AI hardware refresh cycles are rapid.

The 120kW rack isn’t a future scenario; it’s happening now. Infrastructure teams that prepare for this reality will have a competitive advantage.

Smart Hands Done Right: 8 SLA Essentials Every IT Manager Should Demand

Your data center runs 24/7. Equipment fails at 2 AM. Deployments need hands-on-site tomorrow. This is why smart hands services exist: qualified technicians acting as an extension of your team.

Here are the eight SLA essentials you should demand:

  1. Guaranteed Response Times, With Teeth
    Demand response time commitments with financial penalties for failures.

  2. Clear Scope Definition
    Define what’s included in base service vs. billable add-ons.

  3. Geographic Coverage That Matches Your Footprint
    Confirm technician availability in every location.

  4. Qualified, Vetted Technicians
    Verify technical qualifications and security clearances.

  5. Proactive Maintenance, Not Just Break-Fix
    Include scheduled inspections and preventive maintenance.

  6. Detailed Documentation and Reporting
    Maintain records for all operations.

  7. Integration with Your ITSM Platform
    Ensure compatibility with your ticketing and workflow systems.

  8. Performance Metrics and Regular Reviews
    Track average response time and first-time fix rate.

Red Flags to Watch For

Avoid providers who resist adding SLA penalties for missed response times, can’t provide references, or don’t have processes for emergency escalations.

IT Infrastructure Planning for Enterprises: What Needs to Be Defined Before You Buy

Step 1: Define Business and Operational Requirements

  • Performance: Workloads the infrastructure will support.
  • Capacity: Storage, compute, and network bandwidth requirements.
  • Availability: Uptime requirements.
  • Compliance and Security: Regulatory obligations.

Step 2: Identify Stakeholders

Engage IT architects, security teams, and finance leaders early to ensure alignment.

Step 3: Establish Budgets and Timelines

Avoid overspending on unnecessary features and ensure procurement lead times are accounted for.

The Enterprise IT Infrastructure Lifecycle: From Planning to Decommissioning

The IT infrastructure lifecycle includes planning, procurement, deployment, operations, refresh cycles, and decommissioning. Each phase influences the next and requires careful consideration to minimize risks.

  1. Planning & Requirements Definition
    Define critical workloads, performance requirements, and budget constraints.

  2. Procurement & Vendor Selection
    Evaluate vendors based not only on price but also on compatibility and operational risk.

  3. Deployment & Implementation
    Manage testing and documentation carefully to avoid challenges.

  4. Operations & Ongoing Support
    Monitor and manage performance continually.

  5. Refresh & Upgrade Cycles
    Plan for proactive refreshes to avoid risks associated with outdated infrastructure.

  6. Decommissioning & End-of-Life Management
    Ensure secure data erasure and compliance during decommissioning.

The Hidden Cost of Undertrained Data Center Technicians: Why Skills Investment Matters

The skills crisis in UK data centers impacts service quality and reliability. Training investment is crucial for developing skilled professionals to prevent downtime and improve operational efficiency.

The Skills Crisis

  • High Turnover Rates: Young technicians often leave after one year.
  • Limited Technical Depth: A focus on basic tasks hampers growth.
  • Service Inconsistency: Multivendor environments can lead to more downtime issues.

DACPROS: Investing in Tomorrow’s Infrastructure Specialists

At DACPROS, we prioritize professional development, ensuring high-quality service and proactive problem-solving. This approach extends to advanced training in areas such as compliance standards and risk management to mitigate human error.

The Changing Landscape in Data Centers

London’s data center market is experiencing unprecedented growth and transformation driven by operational capacity, evolving infrastructure needs, and AI workloads. Understanding these changes is crucial for maintaining a competitive advantage and successful infrastructure support in an increasingly complex environment.