What AI Is Made Of

Run 04 equipment mass balance

Run identity: R4-RECONSTRUCTION-01. Quantity snapshot: SNAP-RUN03. Configuration: CFG-GB300-2026Q4.

All quantities are metric tonnes per canonical 1 GW of commissioned IT. The base design covers incremental B1 compute, B2 facility equipment, and modeled B3 enabling equipment. B1–B3 sums those increments once. Low, base, and high are coherent input constructions with fixed equipment counts. Other feasible designs could fall outside their range. Transfer from retained product and geometry proxies to the target campus is assumption-dependent. The results are scenario estimates with partial campus coverage.

Elemental lithium is a nested view of host compounds. Aggregate ferrous includes disjoint ferrous classes. Summing every material label would count some stock twice.

Cumulative modeled equipment stocks (B1–B3)

MaterialLow (t)Base (t)High (t)
Copper-grade stock8,718.116,921.430,102.3
Aluminum stock and named alloys1,565.32,478.13,862.6
Disjoint ferrous, linked total32,715.744,281.559,874.2
Elemental lithium, nested view22.624.025.9

Largest equipment contributors in the base case

Component stocks are summed per equipment population. Ferrous combines the disjoint ferrous classes.

StockBoundaryEquipment populationBase (t)
CopperB3R4-FEEDER-GENERATOR4,152.7
CopperB2R4-FEEDER-IT-OUTPUT1,741.0
CopperB2R4-FEEDER-IT-RECTIFIER1,417.1
CopperB2R4-FEEDER-IT-XFMR1,384.2
CopperB2R15-BOM-AIR-COOLED-CHILLER1,041.4
AluminumB2R15-BOM-AIR-COOLED-CHILLER1,154.6
AluminumB3R15-BOM-STANDBY-GENERATOR346.7
AluminumB2R15-BOM-WARM-WATER-CRAH346.6
AluminumB2R15-BOM-IT-UPS265.8
AluminumB1R4-PSU-MODULE127.5
FerrousB3R15-BOM-STANDBY-GENERATOR8,240.0
FerrousB2R15-BOM-IT-XFMR5,468.2
FerrousB2R15-BOM-AIR-COOLED-CHILLER5,451.2
FerrousB2R15-BOM-PDU5,342.3
FerrousB2R15-BOM-PRIMARY-PIPE-RUN4,617.5

Historical reconciliation

The comparison table presents previously reported quantities alongside the current model results. Model development coefficients and ledger quantities are evaluated independently of that table. Agreement here uses historical display precision. The archived report's marginal material endpoints and the modeled joint cases have different uncertainty semantics.

Across low, base, and high, 19 endpoints agree at historical display precision, 128 differ, and 6 lack a value on one or both sides.

Delta is modeled minus historical. Missing estimates are shown as —. The CSV and JSON include all endpoints and historical display tolerances.

BoundaryMaterialModeled base (t)Historical base (t)Delta (t)
B1aluminum213.26290-76.74
B1carbon_steel2,377.452,410-32.55
B1copper1,194.931,420-225.07
B1ferrous2,575.312,690-114.69
B1stainless_steel197.86280-82.14
B2aluminum1,910.93,620-1,709.1
B2carbon_steel13,520.4824,920-11,399.52
B2cast_iron1,725.72,410-684.3
B2copper10,379.8617,320-6,940.14
B2electrical_steel11,946.578,0603,886.57
B2electrolyte_solvents252.01271.8-19.79
B2ethylene_glycol5,462.26,180-717.8
B2ferrous29,881.439,710-9,828.6
B2graphite239.63258.5-18.87
B2lfp507.02546.9-39.88
B2lipf637.6940.7-3.01
B2lithium24.0325.9-1.87
B2nickel22.1323.9-1.77
B2other_ferrous441.02730-288.98
B2polyethylene81.94——
B2r513a180.83180.80.03
B2stainless_steel2,247.643,590-1,342.36
B2transformer_oil1,396.352,470-1,073.65
B2water13,550.2815,350-1,799.72
B3aluminum353.9731043.97
B3carbon_steel4,008.89,090-5,081.2
B3cast_iron4,067.714,720-652.29
B3copper5,346.635,720-373.37
B3electrical_steel3,601.314,600-998.69
B3ferrous11,824.8318,480-6,655.17
B3other_ferrous1478067
B3transformer_oil647.28730-82.72
B1-B3aluminum2,478.134,220-1,741.87
B1-B3carbon_steel19,906.7336,420-16,513.27
B1-B3cast_iron5,793.417,130-1,336.59
B1-B3copper16,921.4224,470-7,548.58
B1-B3electrical_steel15,547.8812,6602,887.88
B1-B3electrolyte_solvents252.01271.8-19.79
B1-B3ethylene_glycol5,462.26,180-717.8
B1-B3ferrous44,281.5460,880-16,598.46
B1-B3graphite239.63258.5-18.87
B1-B3lfp507.02546.9-39.88
B1-B3lipf637.6940.7-3.01
B1-B3lithium24.0325.9-1.87
B1-B3nickel22.1323.9-1.77
B1-B3other_ferrous588.02800-211.98
B1-B3polyethylene81.94——
B1-B3r513a180.83180.80.03
B1-B3stainless_steel2,445.53,870-1,424.5
B1-B3transformer_oil2,043.633,200-1,156.37
B1-B3water13,550.2815,350-1,799.72

The table compares labels and published scope. Differences include effects of component coverage, assumed dimensions, product transfer, and case definitions. The archived material envelopes do not supply the calculation inputs needed to attribute a difference to any one coefficient.

Replacement designs and gaps

Replacement cases in the modeled balance apply to the named assemblies. The historical alternative deltas are carried in the reconciliation as reported claims with their own scope.

ReplacementB1–B3 copper Δ (t)B1–B3 aluminum Δ (t)B1–B3 ferrous Δ (t)
AL-MV-LV-WINDINGS-589.89295.610
AL800-BUSWAY-589.15214.720
AL-PHASE-FEEDERS-10,613.895,318.880
PIPE-HEAVY-WALL002,625.65

The coverage register lists 50 modeled scopes and 11 gap records. Gaps remain outside the quantified sum. See coverage.json for ledger links, physical per-unit masses, budget residuals, missing inputs, and scope notes.

Read budget residuals as accounting constraints and keep them out of the stock totals. The low and high battery module and cabinet budget endpoints are marginal envelopes. Adding those endpoints does not yield a joint 590 kg partition. Modeled battery stock passes the shared 590 kg gross-budget check.

Installed conductor routes and sections, equipment-specific transformer drawings and masses, and selected cooling and battery equipment bills of materials would constrain high-volume metal and chemical-stock estimates. Building shells, foundations, roads, earthworks, and unallocated offsite infrastructure are outside this equipment scope.

Run python scripts/reproduce_run04.py to regenerate these findings and the supporting tables offline. The manifest records SHA-256 hashes of inputs and outputs. The preservation audit compares frozen file bytes with the Run 03 Git tree.

Source details

Run 04 research. Equipment material estimates with documented assumptions and scope. Web prose uses current site terminology; downloadable records preserve original wording. Hashes identify the downloadable records.

Source: build/run04/findings.md
SHA-256: 530ecae0bf37514e35c1f93060f83e7ee534c6582a79e41c7ef70cfc68778607
Download research file

Optional: tell Sid you’re reading
The document is open; sharing your name/email is optional.