SemiAnalysis: AWS and Meta Advance Modular Construction, Cutting Data Center Build Time by 36%

Research firm SemiAnalysis reports in its latest analysis that modular construction has become the default solution for rapidly delivering computing power, from hyperscale cloud providers to AI labs. This approach shortens construction timelines by approximately 36% (about 7 to 9 months) and reduces total capital expenditure (CapEx) per megawatt (MW) by around 8%.

Speed Equals Revenue and Labor Shortages

SemiAnalysis argues that for cloud providers, speed directly translates into revenue. Delivering one additional month of IT compute capacity per MW is estimated to generate approximately $500,000 in value for the operator. For a 50MW facility, the modular approach could yield around $200 million in additional revenue compared to traditional on-site construction.

Additionally, structural labor shortages in skilled trades are a major driver. Electrical workers typically account for 30% to 40% of total data center construction labor hours. SemiAnalysis models warn that demand from large-scale facility builds will lead to a visible shortage of electrical workers by 2027.

Modular construction shifts much of the repetitive work to parallel factory production, reducing on-site labor hours by approximately 63% and cutting demand for licensed electrical workers by about 85%.

From Prefabricated Racks to the 'Tent' Strategy

In practice, AWS and Meta offer representative examples. AWS is advancing a modular design codenamed 'SAMDC,' with its internal 'Project Houdini' breaking down server room space (white space) into standardized, factory-built racks. This reduces preparation time before server deployment from 15 weeks to just 2–3 weeks.

Meta, meanwhile, has adopted a more aggressive strategy at its Prometheus campus in Ohio, deploying temporary fabric-structured 'tent' facilities with aluminum frames. Satellite imagery shows Meta completed eight such structures in under a year, whereas constructing five permanent buildings at the same campus previously took 2 to 3 years.

Three Integration Paths and Market Strategies

The current modular market has evolved into three dominant models:

1. Operator-led Model: Tech giants like AWS and Meta, with strong in-house engineering, design and procure equipment themselves, then outsource integration to contractors.

2. EPC or System Integrator-led Model: Operators define performance requirements, while engineering, procurement, and construction (EPC) firms (e.g., Comfort Systems) handle procurement and construction. This allows operators to retain proprietary designs while benefiting from factory prefabrication efficiency.

3. OEM-led Model: Equipment manufacturers like Vertiv (Vertiv) integrate power, cooling, and infrastructure into complete platforms (e.g., OneCore) for sale. While this increases vendor value, current modular solutions face lead times exceeding 12 months.

Supplier Data and Quality Risks

Despite clear advantages, SemiAnalysis cautions that supplier-reported acceleration figures often reflect narrow scopes rather than end-to-end timelines. For example, some vendors’ claimed 85% acceleration applies only to specific components like overhead busbars.

In terms of cost, fully modular solutions cost about $13.5 million per MW, lower than the traditional $14.6 million, but layered supplier margins remain a key cost friction point.

Moreover, some operators have reported reliability issues with certain modular solutions. SemiAnalysis emphasizes that quality defects could not only erase time savings but also expose expensive hardware assets to significant risk.