As rack power rises toward the megawatt range, the physical footprint of power-delivery equipment becomes an explicit design constraint alongside the space required for compute hardware. As rack power climbs, electrical conversion increasingly competes with processors, memory, networking, and cooling hardware for the same physical envelope. A published engineering projection places rack-level power requirements at roughly 1.5 MW by 2028, illustrating how quickly electrical delivery can become a spatial design constraint rather than a background utility function. However, the relevant commercial question is not simply how many racks fit inside a hall, but how much productive compute each allocated portion of the building can support. That shifts planning toward measures such as compute throughput per usable area, power delivered per occupied row, and ultimately tokens produced from a defined electrical and spatial investment.

The change matters because a rack can contain equipment that contributes directly to computation and equipment that merely enables computation, yet both consume physical capacity. Traditional layouts often accept power shelves, conversion stages, backup components, and distribution hardware as necessary occupants within or beside the compute envelope. As rack power approaches the megawatt range, however, rack-level power-delivery equipment can consume substantial physical capacity that could otherwise accommodate compute hardware. At higher densities, every additional cabinet, shelf, or service clearance competes with an increasingly valuable compute position. A more useful planning equation therefore starts with usable compute area rather than gross building area, then relates that area to delivered megawatts and measured workload throughput. This approach does not claim that square footage alone determines performance, but it highlights the physical tradeoff created when power-conversion equipment occupies space within or immediately alongside the compute environment.

What Actually Leaves the Floor When Conversion Shrinks

The physical change starts with the power-conversion chain that previously followed electricity deep into the compute environment. Conventional architectures can place power shelves, rectification stages, intermediate conversion equipment, high-current busbars, backup units, and associated distribution hardware close to the IT load. Raising distribution voltage substantially reduces the current required for the same power, which in turn allows smaller conductors and less bulky distribution hardware. One engineering analysis illustrates the difference with a 1 MW load: a 50 V distribution bus would require about 20,000 amps, while an 800 VDC bus reduces that requirement to roughly 1,250 amps. That reduction does not automatically translate into an equal reduction in building area, but it changes the physical scale of conductors, busbars, connection hardware, and conversion equipment that designers must accommodate. 

The larger opportunity appears when conversion stops occupying positions immediately adjacent to the compute racks. Moving AC-to-DC conversion to a facility-level power area can remove rack-mounted AC-to-DC conversion hardware from the IT rack and shift its associated electrical equipment into a dedicated power environment. The remaining rack architecture can receive higher-voltage DC and perform only the conversion stages that must remain close to the processor load. This distinction matters financially because removing rack-level AC-to-DC conversion hardware can return previously occupied rack space to compute equipment or allow a more compact rack-level power arrangement. The same architectural progression can shift conversion into a utility room or centralized electrical area while retaining the distribution advantages of high-voltage DC. The value of that reclaimed space ultimately depends on whether the resulting architecture increases deployable compute capacity, improves utilization, or reduces other infrastructure requirements. 

The Side Rack Tax on White Space

The side-rack architecture solves an immediate engineering problem by separating conversion equipment from the main IT cabinet without requiring a complete redesign of the surrounding electrical system. That makes it attractive for deployments where existing AC distribution remains useful and high-density racks need a practical path toward 800VDC. Yet the side rack introduces a different form of spatial cost because the conversion equipment remains physically attached to the compute deployment. Instead of occupying rack units, the sidecar places the power-conversion equipment alongside the IT rack, creating a separate physical footprint that must be incorporated into the surrounding layout. A rack-capacity calculation alone therefore does not capture the additional floor area occupied by a sidecar, which must be included when comparing overall facility layouts.

As the number of high-power racks increases, repeating the sidecar architecture also repeats its dedicated power footprint across the deployment. Each side unit requires a defined location, electrical connection, protection strategy, service access, and physical relationship with the rack it supports. The architecture can therefore preserve much of the existing upstream AC infrastructure while shifting part of the space requirement from rack-mounted power equipment to the area immediately alongside the compute rack. Because the sidecar adds a separate physical footprint while returning rack units to compute, an economic comparison should account for both the recovered rack capacity and the additional floor area occupied by the power center. The right calculation consequently compares the added side-rack footprint with the compute capacity gained inside each associated rack.

How Power Centers Buy Back Your Row

Centralized power centers change the calculation by moving power conversion out of individual compute racks and into a shared facility-level electrical architecture. A single conversion zone can serve multiple racks through an 800VDC distribution network, allowing the high-power electronics to occupy a location designed specifically for electrical equipment. This arrangement can remove repeated rack-mounted conversion equipment from the compute row while concentrating the primary AC-to-DC conversion function within the facility-level power architecture. Therefore, row planning can focus more directly on the placement of compute cabinets, cooling interfaces, networking, and distribution pathways instead of reserving recurring positions for conversion equipment. The gain comes from aggregation: one shared electrical footprint can support multiple compute positions when the architecture and protection scheme allow sufficient scalability and redundancy. 

The economic consequence reaches beyond the simple number of cabinets that fit into a room. Consolidation can change row lengths, aisle planning, equipment adjacency, maintenance routes, and the amount of electrical infrastructure that must repeat from rack to rack. It can influence how designers allocate building area between electrical infrastructure and compute space, particularly as higher rack power increases the physical requirements of the power-delivery system. A centralized approach can therefore make the power center a strategic capacity node rather than an unavoidable collection of cabinets scattered through the IT space. The calculation should include conversion equipment cost, distribution equipment, redundancy requirements, cabling, cooling implications, service access, and the value of compute capacity enabled by the reclaimed area. A lower equipment count alone does not guarantee lower capital expenditure, but a better ratio between infrastructure footprint and deployable compute can materially improve the economic case. 

Floor Space Is Now a Compute Design Decision

The significance of 800VDC extends beyond lower current, smaller conductors, and fewer conversion stages because moving AC-to-DC conversion away from the rack can also return physical rack capacity to compute hardware. When conversion equipment moves away from the rack, the designer gains freedom to treat the compute row as a higher-value production area rather than a mixed zone for IT and electrical hardware. That does not mean every deployment should centralize conversion, since topology, redundancy, retrofit constraints, protection requirements, and operating strategy can produce different outcomes. It does mean that power architecture should enter the facility economics model at the same time as rack density, cooling capacity, and usable hall area. Ultimately, the important measurement becomes the amount of useful computational work produced from the combined investment in power equipment, building area, and electrical capacity. 

A practical capital model can therefore move beyond dollars per installed megawatt and examine dollars per deployable compute unit, dollars per unit of delivered throughput, or capital invested per sustained token output under a defined workload. Such measures force electrical decisions to compete on the same economic basis as compute hardware rather than receiving separate treatment as fixed facility overhead. The strongest architecture will not necessarily have the smallest converter, the shortest cable, or the highest nominal electrical efficiency in isolation. Its advantage should therefore be assessed by how effectively the architecture allocates scarce physical and electrical resources while maintaining the required resilience and serviceability. This perspective makes row geometry, conversion placement, distribution voltage, and power-center location part of the compute architecture itself rather than later-stage facilities optimization.