Data Center Noise Levels and Hearing Conservation: What EHS Directors Need to Know [2026]

The global data center market reached an estimated $384 billion in 2025 and is projected to more than double by 2033, fueled by AI infrastructure buildouts, cloud migration, and hyperscale expansion. That growth means more facilities, more workers, and a noise exposure problem that most EHS programs have not caught up with.

Soundtrace unifies audiometric testing, noise monitoring, HPD fit testing, and automated recordkeeping in a single platform built for multi-site data center operations.

The CDC estimates that 22 million U.S. workers are exposed to hazardous noise levels at work each year. Data center technicians, operators, and maintenance staff are increasingly part of that count, yet the industry has been slow to recognize that server room noise exposure carries the same OSHA obligations as a manufacturing floor.

Why Data Centers Are Louder Than Most People Think

Data centers are not quiet offices with servers humming in the background. The noise environment inside an operating data center is driven by multiple overlapping sources that create sustained, broadband noise exposure throughout the shift.

Primary Noise Sources

Server cooling fans are the dominant noise generator. Each rack contains dozens of high-speed fans spinning at thousands of RPM to move air through densely packed components. As rack density increases to support AI and GPU workloads, fan speeds increase proportionally. A single high-density rack can generate 75–80 dBA on its own; rows of racks in an enclosed aisle push aggregate levels into the 85–96 dBA range.

HVAC and cooling systems add a constant broadband noise floor. Computer Room Air Conditioning (CRAC) units, chiller systems, and air handling equipment operate continuously. Cooling towers associated with data center operations can generate noise levels up to 85 dBA, and rooftop air handling units may generate 85–100 dBA depending on size and configuration.

Backup generators represent an intermittent but extreme exposure risk. Small diesel generators operate at approximately 85 dBA, while larger generators run closer to 100 dBA. Monthly testing is standard practice, and data centers typically run multiple generators simultaneously during testing or actual power events, compounding the exposure.

Uninterruptible Power Supplies (UPS) and power distribution units contribute additional noise, particularly in older facilities with less efficient designs.

The AI-Driven Density Problem

The shift toward GPU-accelerated computing is making data centers louder. AI training clusters can draw 50–100 kW per rack, compared to 5–10 kW for traditional server racks. Higher power density means more heat, which means more aggressive cooling, which means more noise.

For the data center maintenance technician troubleshooting a network issue in a hot aisle between rows of GPU racks, the noise exposure is not abstract. It is the constant, high-pitched whine of hundreds of fans at maximum speed, the background roar of cooling systems operating under load, and the gradual, irreversible damage to the hair cells in the inner ear that will not become apparent until years after the exposure occurred.

AI/GPU Rack Density: What 30–100 kW Does to Your dBA

Rack power density is the single biggest variable in a modern data center’s noise profile, and it has moved fast. A general-purpose enterprise rack ran 5–10 kW for most of the last decade. Cloud and high-density compute pushed the typical envelope to 15–20 kW. AI training clusters now routinely land at 30 kW, with NVIDIA HGX and GB200-class deployments specified at 50–100+ kW per rack. The thermal load scales linearly with power; the fan speed required to remove that heat does not.

Server, switch, and PDU fans are variable-speed. Below a thermal threshold they sit at 30–50% duty cycle and contribute modestly to the room. As inlet temperature or workload climbs, controllers ramp fans toward 100% in seconds. At full speed, blade tip speed roughly doubles and acoustic output rises by 8–15 dBA per rack. A hot aisle that measured 87 dBA TWA behind a general-purpose row can measure 94–98 dBA behind an adjacent GPU pod running training jobs — without any change to the building, the CRAHs, or the staffing model.

Practically, this means three things for EHS programs:

• Treat every new GPU pod as a "change in process" under 1910.95(d)(3). Workers who were below the action level last quarter can be above the PEL this quarter if a tenant or owner stands up 50 kW racks in their work zone. See: OSHA Noise Re-Monitoring Requirements and Noise Monitoring in Data Centers.
• Measure during workload, not idle. A dosimetry sample taken during cold-aisle racking on a Saturday morning will under-report exposure for a technician who works during peak training cycles. Coordinate measurement windows with the application owner.
• Map noise by row, not by hall. A 200,000-square-foot data hall can contain rows that are 84 dBA and rows that are 96 dBA. Treat each row as its own zone in the facility noise map.

Liquid and Immersion Cooling as Engineering Controls

The same density that drives noise up is also pushing operators to liquid and immersion cooling — and those technologies are legitimate engineering controls under the OSHA hierarchy of controls. When the heat leaves the rack in a coolant loop instead of an air stream, the in-rack fans either disappear (full immersion) or drop to a fraction of their air-cooled duty cycle (direct-to-chip). Hot-aisle dBA typically drops 8–15 dB in well-implemented direct-to-chip rows, and 15–25 dB in single-phase immersion tanks.

The catch is what stays loud. Heat does not disappear; it moves. Residual noise sources after a liquid rollout include:

• Coolant Distribution Units (CDUs) — pumps, internal fans, and heat exchangers, typically 70–85 dBA at the unit face.
• Pumps and manifolds — usually within mechanical rooms, but accessible to maintenance staff during service.
• Dry coolers and adiabatic units on the roof or yard — large fan arrays that run 75–95 dBA at close range.
• Any remaining air-cooled hardware in the same hall — switches, storage shelves, and legacy compute that share the room with the liquid racks.

Document the rollout the way OSHA expects engineering controls to be documented: pre-control TWA per affected role, the control implemented, post-control TWA verified by dosimetry, and a written feasibility assessment for any residual exposure above the PEL. If post-rollout TWAs drop below 85 dBA for a given role, the HCP enrollment decision can be revisited — but only after the dosimetry confirms it, not based on the rack-level dBA alone.

Hyperscale vs. Colocation vs. Edge/MTDC

"Data center" covers three operating models with very different hearing conservation realities.

Hyperscale. Single-operator campuses (Google, Meta, AWS, Microsoft, Oracle) with standardized rack designs, in-house mechanical staff, and one EHS organization owning the entire site. Noise programs are typically uniform across halls and roles. The biggest exposure drivers are AI buildouts and generator/load-bank testing.

Colocation (colo) and multi-tenant data centers. The host operates the building, power, and cooling. Tenants own racks and the equipment inside them. Multiple employers share the data hall: host facilities staff, tenant deployment teams, network-provider technicians, OEM field engineers, and short-duration vendors. Under OSHA’s multi-employer worksite doctrine, the host typically controls the noise environment while each employer remains responsible for its own workers’ HCP. See: Contractors & Subcontractors Hearing Conservation and Who Is Responsible for Vendor Employee Hearing Conservation. The host should publish current room dBA in the tenant onboarding packet so tenants can make defensible HPD and enrollment decisions for their staff.

Edge / MTDC. Small, often unstaffed facilities at carrier hotels, telecom huts, retail/industrial sites, and the bottom of cell towers. Permanent workforce is zero or minimal, but visiting technicians and contractors enter frequently for break/fix work. Noise is often as high or higher per rack than in larger sites because thermal headroom is tighter. Exposure is intermittent but real: a network tech who spends 90 minutes inside a 92 dBA edge cabinet, three days a week, can still meet the action level on a TWA basis. Use task-based dosimetry for these roles rather than full-shift area surveys.

Per-Role Exposure Profile

Not every data center worker belongs in the hearing conservation program, and not every NOC analyst should be issued earplugs. The table below shows typical TWA ranges and HCP-scope decisions for the roles that actually staff a modern data center. Treat it as a starting point for your own dosimetry, not a substitute.

HPD selection and 24/7 fit-testing logistics for these roles are covered in detail in Hearing Protection for Data Center Workers. Re-monitoring triggers (GPU pods, generator changeouts, liquid cooling rollouts, containment retrofits) are covered in Noise Monitoring in Data Centers.

OSHA Requirements for Data Center Noise Exposure

OSHA’s hearing conservation standard (29 CFR 1910.95) applies to every general industry employer where workers are exposed at or above 85 dBA TWA, including data center operators. There is no industry exemption.

The Two Thresholds That Matter

85 dBA TWA (Action Level): Triggers the full hearing conservation program requirement. This includes noise monitoring, audiometric testing (baseline and annual), hearing protection provided at no cost, annual training, and recordkeeping. Data center server rooms routinely exceed this threshold.

90 dBA TWA (Permissible Exposure Limit): Triggers mandatory HPD use and engineering or administrative control requirements where feasible. Server rack-level noise (96 dBA) puts workers above the PEL within approximately 3.5 hours of exposure under OSHA’s 5 dB exchange rate.

What Compliance Looks Like in a Data Center

A complete hearing conservation program for data center operations includes six elements:

1. Noise exposure monitoring to establish TWA for each job role, including technicians, maintenance staff, security personnel, and construction workers during buildout phases. See: OSHA Noise Monitoring Requirements
2. Baseline and annual audiometric testing for all employees exposed at or above 85 dBA TWA. Baselines must be established within 6 months of first exposure (or 12 months with HPD use while waiting).
3. Hearing protection devices provided at no cost in multiple styles and sizes. Fit testing verifies actual attenuation for each worker. See: HPD Fit Testing for OSHA Compliance
4. Annual training covering noise hazards, HPD use, and the purpose of audiometric testing.
5. Recordkeeping with noise exposure records retained 2+ years and audiometric records retained for duration of employment.
6. Employee notification of monitoring results and audiometric test outcomes.

Common Compliance Gaps in Data Center Programs

Data center operators often treat noise as an IT infrastructure problem rather than an occupational health exposure. The most common gaps include:

No noise monitoring on record. Many data centers have never conducted formal dosimetry for maintenance and operations roles. Area noise surveys for equipment procurement decisions do not substitute for personal noise monitoring under 1910.95.

No audiometric testing program. Data center workers are frequently excluded from hearing conservation programs because the industry is not traditionally classified as “high-noise.” OSHA citations do not require traditional industry classification. They require evidence of noise exposure at or above the action level.

Inconsistent HPD use. Workers may have earplugs available but receive no training on proper insertion, no fit testing, and no documentation of HPD selection.

No program for construction-phase workers. Data center buildouts involve concrete cutting, steel fabrication, heavy equipment, and power tool use that can produce exposures well above 100 dBA. Construction contractors are subject to 29 CFR 1926.52 noise requirements.

Building a Data Center Hearing Conservation Program

Step 1: Conduct Baseline Noise Monitoring

Start with area noise surveys using calibrated sound level meters to map noise levels across the facility. Identify zones at or near 85 dBA and perform personal dosimetry on representative workers in those zones. This establishes the TWA data needed for HCP enrollment decisions.

Data centers should monitor separately for:

• Server room technicians (typically highest exposure)
• Network operations staff
• Maintenance and HVAC workers
• Security personnel who patrol server areas
• Construction and commissioning workers during buildout

Step 2: Establish Audiometric Testing

All workers with TWA at or above 85 dBA must receive a baseline audiogram and annual follow-up testing. Data center operations present unique scheduling challenges because of 24/7 shift coverage, geographically distributed sites, and rapid workforce growth.

In-house audiometric testing using automated microprocessor audiometers eliminates the scheduling bottleneck of mobile van testing. Data center facilities with multiple shifts can run testing on any shift, any day, and retests can happen immediately when a Standard Threshold Shift is identified, rather than waiting months for the next van visit.

Step 3: Implement Hearing Protection with Fit Testing

Not all hearing protection performs equally, and the data center environment creates specific challenges. Communication headsets used for troubleshooting must be integrated with hearing protection rather than replacing it. REAT-based fit testing verifies that each worker’s HPD is providing adequate attenuation in their actual ear canal, closing the gap between the manufacturer’s NRR and real-world protection.

Step 4: Train and Document

Annual training should be relevant to data center roles specifically. Workers need to understand the noise sources in their facility, why 85 dBA does not “sound loud” but still causes cumulative damage, how to properly insert hearing protection, and what their audiometric results mean.

The Business Case for Proactive Compliance

Data center employment is expanding rapidly. Industry projections indicate the sector could reach 650,000 jobs by 2026, a 30% increase from 501,000 in 2023. Each new facility and each expansion adds noise-exposed workers who require hearing conservation program enrollment.

The cost of non-compliance is not theoretical. OSHA serious violations carry penalties up to $16,550 per violation, and willful or repeat violations can reach $165,514. Workers’ compensation claims for occupational hearing loss can range from $25,000 to over $1 million depending on severity and jurisdiction. The average cost of a hearing test is approximately $300 per employee through third-party services.

A well-run in-house program, by contrast, reduces per-test costs by 40–60%, eliminates scheduling gaps that lead to compliance lapses, and creates the documented record that protects the employer in both regulatory inspections and litigation.

Soundtrace for Data Center Hearing Conservation

Soundtrace unifies audiometric testing, noise monitoring, HPD fit testing, and automated recordkeeping in a single platform built for multi-site operations. For data center operators managing hearing conservation across dozens of facilities with 24/7 shift coverage, this means:

• In-house audiometric testing on any shift without scheduling a mobile van or sound booth
• Real-time noise monitoring data linked to each worker’s profile
• REAT-based HPD fit testing that documents actual attenuation for each employee
• Centralized compliance dashboards with per-site visibility
• Automated STS detection and audiologist review workflow
• HIPAA-compliant, SOC 2 certified data storage with 30+ year retention

See how Soundtrace works →

Frequently Asked Questions