Tech & Sourcing @ Morgan Lewis

As AI continues to shape the technology industry, computing equipment housed in data centers is evolving to support a broad range of workloads. While traditional general-purpose servers remain essential for everyday business operations, AI workloads may run on conventional servers, accelerator-equipped servers, or more integrated computing systems. The resulting differences in equipment architecture and deployment can provide useful context for understanding data center operation and colocation arrangements.

Traditional Servers and Modern AI Computing Equipment

Traditional servers typically rely on central processing units (CPUs) to perform general-purpose computing tasks, including hosting websites, managing databases, and running business applications. These servers may be installed individually or in modular or integrated configurations and may operate independently or as part of a broader network.

AI workloads may run on CPUs, graphics processing units (GPUs), tensor processing units (TPUs), and other specialized accelerators, including application-specific integrated circuits (ASICs). Different processors are suited to different tasks, including AI model training and inference, and may be deployed in individual servers, clusters, or integrated rack-scale systems. Not all AI workloads may require specialized accelerators.

Operational and colocation agreements may cover a variety of commercial and technical arrangements, depending on the services and infrastructure involved. In some circumstances, a provider supplies physical space for a customer's computing equipment, along with supporting services such as power, cooling, and connectivity. Other arrangements may involve different levels of infrastructure, equipment, and operational support.

While the scope of these agreements varies, the equipment and deployment configuration can influence the technical requirements, services involved, and certain contractual obligations. More broadly, these equipment and architectural differences can affect data center planning, facility operations, capacity management, and coordination among operators, customers, and service providers, whether or not a colocation arrangement is involved.

• Space and Equipment Configuration: Computing equipment may be installed in standard racks, modular assemblies, or integrated multi-server or rack-scale configurations. Some AI and high-performance computing deployments may have specialized space, weight, structural-loading, cabling, or installation requirements.
• Power Requirements: Electrical capacity and rack-level power density depend on the equipment, processor mix, utilization, and deployment configuration. Certain accelerator-intensive AI deployments may require substantially greater capacity than many conventional enterprise deployments, with requirements varying widely depending on the technical environments. Power distribution, redundancy, and consumption may therefore differ between arrangements.
• Heat Generation and Cooling: Cooling needs vary with equipment design, heat output, rack density, and facility conditions. Both general-purpose and AI computing equipment may use air cooling, direct-to-chip liquid cooling, immersion cooling, or hybrid approaches, where supported by the equipment and facility. Higher-density deployments may call for additional infrastructure, such as coolant distribution units, piping, or heat-rejection capacity. Appropriate temperature and humidity controls remain relevant across deployments. Cooling choices may affect facility modifications, monitoring, maintenance, and the allocation of operational responsibilities.
• Connectivity and Maintenance: Both general-purpose and AI computing systems may rely on shared networks and interconnected components. Some distributed AI workloads use high-bandwidth, low-latency interconnections across servers or racks, while others operate on less tightly coupled infrastructure. Equipment architecture and serviceability may affect installation, maintenance access, upgrades, and operational coordination.

  Contractual and Operational Considerations

  These deployment-specific technical characteristics may need to be reflected in contractual obligations and risk allocation. Operational considerations may include facility readiness, monitoring, planned maintenance, capacity constraints, and procedures for responding to infrastructure incidents.

• Contractual Obligations: Agreements may specify available and committed power capacity, cooling capabilities, connectivity, and responsibilities for equipment installation, facility modifications, infrastructure upgrades, monitoring, and maintenance.
• Risk Allocation: Variations in equipment and infrastructure configurations may influence how parties allocate responsibility for power interruptions, cooling failures, leaks, and equipment damage. For example, interruptions affecting shared infrastructure could impact multiple interconnected components, raising questions about service levels, remediation, and responsibility for resulting disruptions.

As computing technology evolves, changing processor architectures and deployment configurations may introduce new considerations for colocation arrangements. Understanding the relevant equipment and facility requirements, rather than relying on a fixed distinction between traditional and AI servers, can help parties assess how data center infrastructure and related agreements may evolve. These developments may also affect data center design, day-to-day operations, infrastructure investment, and the management of increasingly varied computing environments.

Morgan Lewis advises data center developers, operators, cloud providers, AI companies, investors, and other participants on the legal and commercial issues arising throughout the data center lifecycle. Our multidisciplinary team can assist with data center development and operations, colocation and other service agreements, compute capacity arrangements, power procurement and interconnection, equipment and technology transactions, and the allocation of operational and infrastructure-related risks. We work with clients to align contractual commitments with the particular technical and commercial requirements of their facilities and computing deployments.

Through our Data Center Strategic Initiative, Morgan Lewis brings together lawyers across technology transactions, energy, real estate, finance, regulatory, and other practices to address interconnected data center challenges. This coordinated approach can help clients evaluate infrastructure and capacity needs, negotiate arrangements with operators and service providers, and plan for evolving AI and other computing workloads. For more information about our capabilities and insights, visit the Morgan Lewis Data Centers initiative.