Here at OCOLO, our sales team, Sean Murray and Karmann Chan, spend their days bringing both sides of the colocation solutions transaction process together – service providers with space available to lease and buyers looking for the ideal solutions to meet their exact needs. In the process, our team learns a lot about the unique and shared pain points on both sides of a matchmaking process that’s only grown more challenging with the rise of AI.
Recently, they’ve been hearing more and more about a Chicken or Egg quandary buyers and sellers are grappling with as a result of the swift evolution and growth of AI.
It goes like this: AI workloads need more power and efficiency to support them. While traditional servers consume 500W to 1.5kW per rack unit, AI workloads consume anywhere from 5kW to 100kW+. For traditional data centers, which typically have 208V power systems, to be prepared to take on AI workloads, they must upgrade to 415V power systems.
Beyond broadening their audience to an important, growing audience, there are significant benefits to upgrading: 415V power systems increase efficiency lower copper and cabling costs by requiring thinner wiring; increase rack density and reduce heat generation and cooling costs. Plus, 415V power systems are already the standard in Europe and Asia and meet the requirements of the world’s largest AI hardware manufacturers.
There are also meaningful hurdles to clear. Converting a data center from 208V to 415V power involves a complex set of changes across infrastructure, equipment, and operational systems – all of which cost money.
The total cost depends on multiple factors, starting with the size of the facility itself and extending from there to the extent of the necessary physical upgrades, downtime and compliance requirements. Here’s a breakdown of the key cost drivers:
1. Electrical Infrastructure Upgrades
- Replacing or reconfiguring step-down transformers, power distribution units (PDUs), and automatic transfer switches (ATS).
Estimated cost Impact: $25K–$100K+ per unit.
- Upgrading to higher-voltage switchgear and circuit breakers
Estimated cost Impact: Tens of thousands per switchgear unit (for N+1 or 2N setups).
2. Cabling and Conduit Modifications
- Rewiring conductors and connectors to the smaller gauge wires that 415V systems use that draw less current
Estimated cost impact: Variable – this reduces copper costs over time, but creates significant near-term labor and downtime expenses.
3. IT Equipment Compatibility & Reconfiguration
- Power Supplies – while most modern servers, switches and routers support 100–240V, some legacy or proprietary gear may require upgrades.
Estimated cost Impact: Minimal if equipment is compatible; significant if large hardware replacement is needed.
4. Rack-Level Power Distribution
- Upgrading metered or switched rack PDUs to match the new voltage and plug type (e.g., C19/C20 vs. C13/C14).
Estimated cost Impact: $500–$2,000+ per rack, depending on features (metered, monitored, remote switching, etc.)
5. Design, Engineering & Permitting
- Electrical Engineering & Compliance – involves electrical redesigns, load balancing and code compliance (e.g., NEC, IEC), and potentially permits, inspections and coordination with utilities
Estimated cost Impact: $20K–$100K+, depending on size and jurisdiction.
6. Cooling Systems Impact
- Accommodating the heat load shifts that can come from AI’s higher power density – for example, rebalancing or upgrading HVAC and cooling systems in high-density AI clusters.
Estimated cost Impact: Variable – can exceed $100K+ if CRAC or liquid cooling is added.
7. Downtime & Labor Costs
- In live facilities, the conversion may require Business Continuity Planning (BCP) efforts such as staged outages, risk management and planned overtime shifts.
Estimated cost Impact: Variable – measured in lost productivity, SLA penalties or the need to implement temporary backup systems.
So what kind of bill does all this effort add up to? Talking to operators, Sean Murray and Karmann Chan have heard estimates ranging from $500,000 to $1 million per MW of capacity, with the main factors driving the expense for them the needed upgrades to UPS distribution switchgear and the lengths of new feeders to the new 415v PDUs or transformers, as outlined above.
Interestingly, they’ve also heard that, even as these data center operators acknowledge they’re missing out on AI workload leasing opportunities because they haven’t upgraded, they still don’t plan to at the moment.
Why not? Therein lies the chicken and egg conundrum. Data center operators don’t want to incur the expense of upgrading facilities until they have the contracts in hand, and buyers don’t want to sign on the dotted line unless they’re confident the AI workload-ready space with an upgraded power system is in place as of Day One.
So, even a march of progress as relentless as AI has its roadblocks along the way.
What are you hearing from the field? Is the need to upgrade power systems slowing down your business or that of your clients? Is there a better solution out there or a way to manage costs and timing of conversion? Let us know in the comments!
At OCOLO we truly believe in #coloforeveryone and want to be part of the solution for all the data center industry’s growing pains. Let’s all work together to help shape a brighter future for the sector and the world!
