Datacenter supply scale

Play the simulator first. The live page is https://tcreations3888.com/eif/. This paper is the largest picture in the set, and it is still a picture. “Datacenter supply scale” means the load names start to look like a computing hall’s bus: racks as discrete blocks, a large share of air handling, a hotel floor of auxiliaries that never goes to zero, and a spike when a row of machines is switched on. It does not mean a plant that supplies a computing hall, a module rating, or a price.

Real computing campuses are often discussed in the megawatt class of *demand*. That sentence is about loads people already build. It is not an output assigned to this concept. No megawatt figure below is a promise of supply.

What this scale is for

The Spike preset closes every breaker:

Switched power is 200 kW. At demand factor 100% the balance is 12 + 200 = 212 kW. The rack alone is larger than Lab day’s entire switched bank. That is the teaching point of the scale. Computing load arrives in blocks. You do not dim a hall into a spike. You close a breaker, and the target jumps, and the field follows on a lag.

The page marks HIGH DRAW when P_load is above 180 kW. Spike qualifies. The badge does not open a trip, shed the rack, or call an operator. It is a visual flag that the toy bus is in the upper part of its play range.

Spike, worked as a picture of a spiky bus

The relations are the ones in WP-01. For P_load = 212 kW the ratio to the 20 kW reference is 10.6.

From Lab day to Spike the bus grows by about 2.8×. Current grows by about 1.7×, because of the square root. The column narrows only from 6.54 mm to 5.61 mm, because radius depends on load to the power 0.15. The azimuthal field roughly doubles. Fusion intensity, being the square of the load ratio, jumps from about 15× to about 112×. τ stretches from 2.81 s to 6.86 s: the heaviest bus in the toy is also the slowest flywheel.

The 1563 V figure is the resonant-tank voltage in the same 60 Hz story as the other papers. It is not a substation specification and not a feeder design. Direct AC coupling, at this scale as at the others, means the tap in the picture is already alternating. The page still does not model a turbine.

The number that should make you suspicious

P_AC,toy at Spike is about 3335 kW against a 212 kW bus. That is the cubic from WP-03, turned up:

PAC,toy = 0.00035 × Pload³

At 212 kW the cube is large, and 0.00035 times that cube is thousands of toy kilowatts. If this were a plant balance, the figure would be a scandal or a breakthrough. It is neither. It is what the published toy does when every breaker is closed. The gauge exists so you can see load-following run away as arithmetic, and so the caption can do its job: toy coupled output, not Q, not net plant power.

A supply-scale thought experiment that hid this number, or dressed it as efficiency, would be a prospectus. This one prints it beside the formula.

Four beats, on a bus that spikes

1. Load → current. Closing the 80 kW rack is the scale’s characteristic gesture. The target for I_z steps up with the square root of the new bus, including the 12 kW hotel floor that was never optional. Opening the rack later eases the current down. It does not fall through the floor.
2. Current → B_θ. At Spike the toy field is about 11.6 T and the drawn radius is about 5.6 mm. Stronger current, tighter column, still a continuous drawing rather than a discontinuity. The rings are the azimuthal field. They are not a magnet specification.
3. Inertia. Nearly seven seconds of τ means a spike you can watch. The breakers are already closed; the gauges are still on their way. That is the entire content of “electromagnetic inertia” on this page. It is not a claim about grid frequency response, reserve markets, or how fast a real hall can ramp.
4. Direct AC. 60 Hz locked, voltage following the square root of load, no turbine. A computing hall in the real world cares about power quality, ride-through, and conversion stages. None of that is modeled. The wave on the gauge is a teaching lock.

What this scale refuses

Earlier drafts of this paper carried module ratings in the tens of megawatts, a commercial architecture, and a path through a demonstration article to a product. Those drafts read like an offer of power. They are withdrawn.

This paper does not claim Q>1, net plant power, a capacity factor, a construction duration, or a site beside a campus. It does not specify a fuel, a blanket, a first wall, or switchgear. It does not authorize construction. Orders of magnitude stay in their lane: campus *demand* is a megawatt-class subject in the outside world; the toy’s own upper play is a 212 kW bus and a HIGH DRAW badge.

The useful residue is the load shape. A computing hall’s draw is lumpy, it has a floor, and a machine that claimed to follow it would have to show its lag. The simulator lets you feel that sentence. It does not build the machine.

In this set

• WP-01 Tabletop proof-of-concept — one switch against the hotel floor.
• WP-02 Garage scale — demand factor as the second knob.
• WP-03 University lab prototype — Lab day’s first paint, and the cubic that is not Q.
• WP-04, this paper — datacenter supply scale. Spike, printed in full so the large number stays a toy.
• Live simulator: https://tcreations3888.com/eif/
• Index of the four papers: https://tcreations3888.com/eif/#whitepapers

Honesty

Numbers are scaled for play. This is an educational demonstration under the MIT License, in the same spirit as the page footer. Datacenter supply scale is a thought experiment about a spiky bus. It is not a power plant, not an offer of supply, and not a construction guide. P_AC,toy is not electrical gain.