Racks Above 100 kW: Why AI Data Centers Need Standards, Not Just Rules

Rack densities exceeding 100 kW, direct liquid cooling as the new default and fiber counts that few facilities carried only a few years ago — these are the figures now driving AI data centers from low-profile infrastructure projects into the center of public policy debates about electricity, water and land use.

That is the argument advanced by Melissa Newman, senior vice president of government affairs at the Telecommunications Industry Association (TIA), in an Expert Opinion column published exclusively by Broadband Breakfast. Newman, who spent more than 20 years in telecom including a stint as deputy division chief of the FCC's Policy Division in the Common Carrier Bureau, makes the case that legislation alone cannot deliver responsible AI infrastructure.

"Public policy defines the outcomes society expects. Consensus standards provide the technical, quality and security framework for achieving those outcomes consistently, transparently, and at scale," she writes.

What is changing in the data center buildout?

Generative AI's compute demands are accelerating data center construction worldwide while fundamentally altering what facilities require. Newman points to rack densities above 100 kW, widespread direct liquid cooling adoption, expanding fiber connectivity and increasing electrical capacity as evidence that complexity has reached levels uncommon even a few years ago.

New facilities now influence utility investments, transmission planning, water availability, permitting processes and workforce development. Policymakers have responded, focusing on energy consumption, water use, environmental impacts, ratepayer protections and economic competitiveness.

That focus is appropriate, Newman concedes. But it covers only part of the picture. Once a facility is approved, its long-term performance depends on engineering practices, quality systems and operational disciplines — and, as she notes, those practices are rarely defined in legislation or regulation.

Why can't regulation do the job?

Technology evolves too rapidly for regulations to specify design methods, equipment architectures or operational procedures, Newman argues — and they were never intended to. Instead, consensus standards translate broad policy objectives into measurable technical requirements covering design, interoperability, quality, resilience, security and operational accountability.

Rather than drafting unique technical requirements for every jurisdiction, she writes, policymakers can reference recognized standards in permitting, procurement, certification and oversight — an approach she says encourages innovation while giving regulators, operators, suppliers, investors and communities consistency and transparency.

A caveat worth weighing: the standards Newman champions are developed by TIA itself, the industry body she represents. Her column is an argument for standards' importance, not an independent evaluation of whether they are being adopted at scale.

Which standards are on the table?

Newman highlights three frameworks:

• ANSI/TIA-942 — covers data center design, telecommunications infrastructure, power distribution, cooling, cabling, redundancy and operational reliability. TIA is developing an AI Addendum to address high-density power distribution, advanced liquid cooling architectures and high-fiber-count cabling for AI workloads.
• DCE 9000 — TIA's Data Center Excellence Quality Management System Standard, built on ISO 9001, targets the organizations that design, manufacture, install, integrate, commission and maintain data center infrastructure, emphasizing process consistency, supplier accountability, risk management and continual improvement.
• SCS 9001 — a Supply Chain Security Management System Standard addressing security risks across the lifecycle of network-connected products, covering supplier management, secure development practices and product integrity. Modern AI data centers source networking, power, cooling and control components from numerous manufacturers across multiple countries, Newman notes.

The TIA-942 AI Addendum is the only development with a stated timeline in the column — it is in progress, with no publication date given.

What is left unsaid?

Newman's framework rests on voluntary adoption. Nothing in the column specifies how many jurisdictions currently reference these standards in permitting, or whether AI data center operators have certified against DCE 9000 or SCS 9001 in meaningful numbers. The claim that standards make responsible development "repeatable, measurable, and independently verifiable" is accurate in principle; verification of actual uptake is another matter.

The column also stops short of addressing the harder policy fights — grid interconnection queues, water rights disputes and community opposition — where standards offer little leverage.

What comes next?

The debate, Newman writes, will continue to center on electricity, water, environmental impacts, community planning and job creation, and those issues will remain essential to responsible development. Her closing pitch to governments shaping the next generation of AI infrastructure policy is direct: "Public policy helps define the destination. Standards provide the technical roadmap for getting there."