Pod estimating: first-pass sizing without burning the week.

Why the first pass eats the week

Prefab pod, e-house and power-skid builders live on bids where the RFQ says something like “1 MW, 40 racks, Tier-appropriate redundancy” and the customer wants a number this week. Getting to that number means a circular sizing exercise:

• kW per rack → PDUs and feeds, and everything upstream of them.
• Redundancy → UPS: N, N+1 or 2N changes the module count, battery strings and step-down transformers per side — not by a percentage, by topology.
• Load → generators and switchgear, with interrupting ratings checked at the voltage, not assumed.
• Density → cooling: CRAH units, coolant distribution, rear-door heat exchangers — sized against the heat each rack actually sheds. The classic miss: a 40 kW rack quoted with 25 kW of cooling, discovered after the steel is ordered.
• Plus the parts that get forgotten at pricing time: containment, row busway, commissioning levels 1–3.

Done by hand, that's days of a senior engineer's time — spent on a bid you may not win. The estimating desk becomes the growth constraint not because demand is short, but because first passes are expensive.

The expensive part of a pod bid isn't the pricing. It's the engineering pass you do before you know if the bid is real.

What good pod estimating looks like

• One system model, not six spreadsheets: racks, power train and cooling in a single model where a change in one resizes the others.
• Redundancy as logic, not a multiplier: N+1 and 2N carried through module counts, transformers and switchgear topology.
• Checks that refuse, with reasons: the estimate should flag the rack whose heat exceeds its cooling — before the quote ships, not after.
• A BOM your engineers can defend: every size traceable to a rating or a rule, so the first pass survives contact with the customer's consultant.
• Speed that matches bid economics: cheap first passes mean you can afford to answer more RFQs — and reserve the engineering week for bids that shortlist.

The AI-era version

This is what Nteli's data center line does: say “1 MW, two rows of 20 racks at 25 kW, N+1” to an AI assistant backed by a deterministic rules engine with a nested system model. The AI asks what narrows it; the engine sizes the power train and cooling together, refuses what doesn't fit with the reason shown, and prices the pod as one BOM — UPS frames and modules per side, battery strings, step-down transformers, generators, switchgear, CRAH, CDUs, rear-door heat exchangers, rack PDUs, commissioning. It runs inside Claude, and the 90-second demo shows a real pod priced end to end.

As a planning benchmark: our leak-ledger rates — engagement-derived planning rates, not surveys — put config and discount errors at 3% of quoted value for configure-to-order manufacturers. On pod-sized orders, a single sizing miss dwarfs that number; the cited-rule check is the point.