Indian philosopher and writer Sarvesh Jain observed that the scarcity of anything teaches us how to survive under extreme circumstances, which certainly speaks to the data center deployment challenges faced by IT departments in a supply chain-constrained world.

Since the onset of COVID-19 in 2020, supply chain disruptions have significantly impacted many industries worldwide. For months at a time, dozens of cargo ships anchored at sea, waiting to unload inventory in major ports, while hundreds of trucks lined up to load containers at the docks. Manufacturing and shipping delays, especially for IT hardware and data center critical systems equipment, have prevented or delayed enterprise IT departments in executing their data center deployment strategies.

The hardware delivery delays continue. Enterprises ordering new hardware can expect delivery timelines of four to 12 months for servers, six to eight months for routers, and eight to 10 months for in-rack PDUs. Specific delivery timelines vary with the hardware ordered but are even longer for components containing specialty semiconductors.

Enterprises seeking to expand, renovate, or upgrade their on-premises data centers also face significant delays in receiving critical systems equipment. Customers can expect delivery timelines after order placement of 10 to 24 months for backup generators, 10 to 18 months for electrical switchgears, and eight to 12 months for UPS systems. Step-down transformers required for large utility power increases now have lead times ranging from 24 to 36 months.

Abundant versus scare paradigms

To best understand the extent to which supply chain shortages have affected enterprise IT strategies, it’s helpful to evaluate the supply chain delays through two paradigms: one of abundance and another of scarcity.

Under the abundant paradigm, items in demand are widely available in different sizes, qualities, quantities, and prices from different manufacturers, on varying delivery schedules. To illustrate, let's consider eggs.

At our local grocery store, we expect to find many different options when buying eggs. We can select from medium, large, jumbo, and extra jumbo eggs; we can buy them in boxes of six, a dozen, sixty, or even a single egg sometimes. We can find white, brown, organic, Omega-3, cage-free, and free-range eggs. Eggs are available at every grocery store, most major discount stores, and neighborhood convenience stores. Because of this abundance, consumers don’t really need to strategize meticulously on egg procurement or budgeting — we simply buy eggs as needed at the store.

Compare that with the scarce paradigm, as it could be applied to the semiconductor industry. Since the onset of COVID, semiconductor availability has diminished significantly, especially for highly specialized and cutting-edge chips. Similarly, servers, storage, routers, smart PDUs, and other hardware required for data centers has been sharply curtailed or delayed.

Precision planning is essential

Because desired IT equipment is not as immediately deliverable as it historically has been, enterprises have been forced to plan project deployments under the scarce paradigm. The need for precision planning is underscored by the high costs and criticality of data center hardware when compared with other products, like cabling or cleaning supplies.

Large enterprise IT departments that previously planned deployments on a quarterly basis for the coming year are now forecasting hardware installations monthly with related capital expenditures and program phases two or three years out to avoid surprises and sequential deployment interruptions if a specific program task is delayed.

However, improving and pushing the planning process further in the future is difficult, and reaching internal agreement on usage assumptions in an uncertain economy where recession may be right around the corner can be challenging.

Colo and cloud solutions

Colocation and public cloud providers offer numerous solutions to enterprise IT departments challenged with these and other related supply chain challenges.

Hardware on demand — Leading colocation providers can rapidly deliver bare metal as a service (“BMaaS”) for enterprise customer use. Colo providers maintain in-stock inventories of servers, storage, and routers for rapid installation at their largest data centers and can ship warehoused inventory to other locations as needed.

Contract commitments for bare metal from colo providers are usually two to three years, but these services can be included in a “spend shift” bucket of services. Spend-shift is a feature of colocation contracts whereby the customer may change the products and services it procures from the provider, such as locations and quantities of colocation occupancy, bare metal, and managed services. This feature improves flexibility to the customer while preserving contract revenue to the colo provider.

More compute capacity — Enterprises lacking hardware can also rapidly procure computing and storage services from the public cloud providers who have been proactive in developing data centers worldwide. Particularly in major data center markets, the public cloud providers offer vast scalability on short notice plus the flexibility to reduce quantities in the future if costs are too high or legacy applications don’t perform well on a cloud platform. However, the public cloud contracts may also levy egress fees or other unbudgeted costs that are far higher than originally planned by the customer.

Public cloud and colocation campuses have mushroomed in scale in the last five years and now total more than 100 MW when, a few years ago, a 40-MW data center was considered exceptionally large. The major public cloud providers are also deploying local zones worldwide, but without offering all services (and also not in the largest quantities) in some expansion markets. Telecom connectivity and latency from the public cloud campuses back to enterprise data centers varies widely by location and provider.

Private cloud — Colocation providers also offer private cloud from their largest data centers, which is similar in many ways to public cloud. Colo private cloud offers similar compute, storage, database, security, and AI options as public cloud but usually with narrower feature sets. Private cloud is an excellent on-ramp solution for enterprises, which are planning future public cloud deployments but are also unsure about cloud application performance or costs. Private cloud contract durations are typically one year, a longer contract than required by the public cloud providers.

Private cloud offers one other valuable but underpublicized benefit to large enterprise customers, especially those operating in heavily regulated industries. Private cloud can be delivered within a specific data center premises like an access-secured private cage adjacent to the customer’s suite. This offers significantly better audit, compliance, and regulatory features than those available from public cloud providers, where location and access control is opaque.

AI supply chain challenges — Enterprises expanding their AI and ML platforms face additional challenges exacerbated by supply chain challenges. AI and ML tend to work best in tight hardware clusters with low-latency network routers, leading to much higher cooling requirements per square foot than in traditional data centers.

Many legacy on-premises data centers were not designed for the higher cooling densities often sought for AI optimization, and the older facilities can’t be inexpensively or quickly upgraded. End users have, therefore, turned to colocation data centers to host AI deployments, even for those enterprises with plenty of unoccupied low-density data center space.

Leading colocation providers offer high electrical and cooling densities in their newest data centers plus the scale to allow rapid expansion of AI clusters. Especially in the major data center markets worldwide, modern colo data centers can easily cool 25 kW per cabinet across an entire cabinet cluster using hot- or cold-aisle containment. Some cutting-edge colo providers use fan wall technology to cool 50 kW per cabinet, while others offer liquid refrigerant and immersion cooling to reject 50 kW to 100 kW per cabinet equivalent.

Strategies for AI hosting

Enterprises evaluating AI applications can adopt specific colocation procurement strategies to enhance AI effectiveness, by seeking:

  • written representations from the colo provider on the heat rejection capacities for the suite being evaluated;
  • a computational fluid dynamics (CFD) model using intended hardware deployment to identify any cooling system design weaknesses;
  • service level agreement terms that match the colo provider’s cooling representations;
  • contractual requirements by the colo provider to supply additional cooling for future density increases; and
  • installation of hot- or cold-aisle containment from the first day of occupancy.

Cloud AI “as a service” — Each of the major public cloud providers now offers an advanced suite of AI and ML services. The public cloud providers have also been leaders in deploying thousands of power-hungry, high-capacity processor servers most suitable for AI applications usage. Many AI vendors are also offering services within their suites inside new colocation data centers on a contract basis.

I’d like to expand, but — Supply chain interruptions have caused significant delays in self-operated data center expansions, as large enterprises can’t get new critical power or cooling equipment quickly. Delays in delivery of large critical systems components have caused delays in the retirement of older data center phases, risking overall reliability by extending some compute functions past their useful lives.

Colocation capacity is available — Enterprises facing on-premises expansion delays have turned to colocation space for accelerated deployments, and those users can carefully select facilities to maximize flexibility and minimize costs. While overall occupancy rates have increased in major colocation markets worldwide since 2020, space, power, and cooling are still available for enterprise customers in most locations.

Astute users engage expert consultants to assist in colocation planning and procurement, and intelligently plan their needed capacities with more granular quantities than in prior years, seeking:

  • a ramp of increased capacity over time;
  • early termination and contraction options to maximize contract flexibility;
  • “spend shift” rights to other facilities and services;
  • shorter contract terms coupled with fixed-rate renewal options; and
  • bidirectional inflation adjustments if a colocation provider requires rate adjustment triggers.

We don’t know yet what changes are coming across the supply chain disruption continuum, although signs are pointing toward improvements in the next year. Meanwhile, the colocation and public cloud providers have provided a vital role in helping enterprises meet evolving computing and storage needs challenged by supply chain disruptions worldwide.