Humanoid Working-Capacity Workbook (Release Asset #11)

Scope statement — read this first

This workbook is an illustrative embodied-capacity extension of the whitepaper’s core chain (Energy → hardware → compute → models → tokens → AI working capacity → digital work → outcomes → value → agency), extended to: Capital + energy + maintenance + utilisation + orchestration → humanoid physical working capacity.

It is not a general claim about robotics economics, a product recommendation, a forecast of any real robot’s field performance, or investment/procurement guidance. It models one narrow question — what does an hour of a humanoid robot’s uptime cost to own, under stated assumptions — using (a) one clearly-labelled scenario assumption (the author’s original €25,000 illustrative robot) and (b) a real market price range assembled from four named, currently sold or announced platforms. All figures below are either cited primary specs/prices, attributed secondary claims, transparent arithmetic, or explicitly labelled scenario assumptions — never a forecast. This is educational research and scenario analysis, not investment, procurement, tax, or policy advice. This boundary applies to every section below, not only here.

Evidence class legend (used throughout)

Part 1 — Two starting points: illustrative scenario vs. real market range

Explicitly excluded from this workbook’s numeric model because pricing is unconfirmed or non-existent as of 2026-08-12: Tesla Optimus (Musk has repeated a $20,000–$30,000 long-term target, ATTRIBUTED STATEMENT, no commercial orders or confirmed retail price exist) and Figure AI’s Figure 03 (no official pricing published at all; a ~$20,000 figure circulates only via third-party aggregators). Both would sit inside or near the MID band if their aspirational figures ever became real prices — flagged here as INTERPRETATION, not fact.

INTERPRETATION: the €25,000 (~$27,000) illustrative case sits between the real-world MID consumer band ($20,000) and would be roughly 2x the observed LOW price and roughly 1/9th the ATTRIBUTED enterprise HIGH price. It is a plausible illustrative point within the real spread, not an outlier — but it is still not a verified market price for any named product.

Part 2 — Required economic layer separation

Per this project’s modelling rules, the layers below are kept separate and never collapsed into one number: capital cost → financing cost → electricity (operating) cost → maintenance reserve → utilisation → cost-per-productive-hour. (The further layers — token production cost, workload/AI-working-capacity cost, outcome/value — are out of scope for a hardware-hours model and are addressed only in the Part 6 interpretation, explicitly labelled as such.)

2.1 Capital layer (all four price points)

The 20% down payment is a SCENARIO ASSUMPTION controlling how much capital is financed vs. paid up front; it is editable (see Part 3.2 sensitivity).

Part 3 — Financing layer

DERIVED CALCULATION. Standard amortizing-loan formula:

M = P × i / (1 − (1+i)^-n) where P = amount financed, i = monthly rate (annual rate ÷ 12), n = number of monthly payments (years × 12).

Base case: 5-year term, 8% annual rate (both SCENARIO ASSUMPTIONS — editable; 8% is illustrative, not a quoted lender rate). Down payment is amortized straight-line over the same term and added back so total capital+financing cost is comparable across cases.

Total annual capital+financing cost = (M × 12) + (Down payment ÷ financing term years)

3.1 Base case (5yr @ 8%, 20% down)

3.2 Financing-term sensitivity (8% rate, 20% down held constant) — total capital+financing cost/yr

(Illustrative column recomputed at the $27,000 USD-equivalent scale for internal consistency; native-EUR values scale proportionally.)

3.3 Down-payment sensitivity (5yr @ 8% held constant) — total capital+financing cost/yr

INTERPRETATION: term length matters far more than down-payment size in this model — stretching from 3 to 7 years cuts the annual capital+financing burden by roughly half, while doubling the down payment only trims it by a few percent. Both are editable levers, not market facts.

Part 4 — Electricity layer

Two parallel methods, both SCENARIO ASSUMPTIONS unless noted:

(a) Flat operating-hour assumption (as in the original illustrative case): $0.11/operating hour (≈€0.10/h at the assumed FX rate) — a simple placeholder, not tied to any specific robot’s power draw.

(b) Spec-derived rate, using 1X NEO’s OBSERVED battery spec (842 Wh runtime over 4 hours → average continuous draw ≈ 0.2105 kW) at an illustrative $0.15/kWh electricity price (SCENARIO ASSUMPTION — a plausible commercial/industrial US-average rate, not a quoted utility tariff):

Electricity $/hour = average power draw (kW) × electricity price ($/kWh) = 0.2105 kW × $0.15/kWh = $0.0316/hour

Source for the underlying spec: https://www.1x.tech/neo (OBSERVED FACT — 842 Wh runtime, 4-hour battery life), checked 2026-08-12.

INTERPRETATION: the spec-derived figure is an order of magnitude lower than the flat $0.11/h placeholder — humanoid robots at ~200W average draw are electrically cheap to run relative to almost any other capital cost in this model; electricity is not the binding cost driver here (unlike in data-center/token-production economics elsewhere in the whitepaper). This spec applies to 1X NEO specifically and is used as an illustrative order-of-magnitude anchor for the other three price points too — it is not verified for Unitree G1, Agility Digit, or the illustrative case.

Part 5 — Maintenance layer

SCENARIO ASSUMPTION: maintenance/repair reserve = 10% of purchase price per year (unchanged from the original illustrative case; not a manufacturer-quoted service contract for any of the four platforms — none of the four sources in the register publish a maintenance-cost figure).

Part 6 — Utilization sensitivity and cost-per-hour ranges

DERIVED CALCULATION.

Cost/hour (financing+electricity) = (Total capital+financing cost/yr + electricity $/hr × annual hours) ÷ annual hours Cost/hour (financing+electricity+maintenance) = (Total capital+financing cost/yr + electricity $/hr × annual hours + maintenance/yr) ÷ annual hours

Shown at both electricity assumptions (flat $0.11/h vs. spec-derived $0.0316/h) and four productive-utilization levels (2,000 / 4,000 / 6,000 / 8,000 hours/year — themselves SCENARIO ASSUMPTIONS bracketing single-shift to near-continuous operation).

6.1 Illustrative base case (~$27,000 / €25,000) — USD, native-EUR cross-check in parentheses

(EUR values shown are the exact recomputation in native EUR at €0.10/h flat electricity — they match the original worked example, confirming internal consistency.)

6.2 Real-range LOW ($13,500, Unitree G1)

6.3 Real-range MID ($20,000, 1X NEO early-access price)

6.4 Real-range HIGH ($250,000, Agility Digit)

6.5 Summary — cost-per-productive-hour range across the whole spread

INTERPRETATION: utilization dominates the cost-per-hour result more than any other single variable in this model — moving from 2,000 to 8,000 hours/year cuts the per-hour cost by roughly 3.5–4x at every price point, because capital and financing costs are fixed while the denominator (hours) grows. The illustrative €25,000/~$27,000 case tracks closely with the real MID consumer band ($20,000, 1X NEO) once financing/maintenance assumptions are held constant — it is not an outlier relative to real currently-purchasable platforms, but it sits far below the ATTRIBUTED enterprise-tier HIGH case, whose cost-per-hour is dominated by a purchase price that itself rests only on secondary-source corroboration, not a primary Agility Robotics price sheet.

Part 7 — What this model explicitly excludes

Per the original scope (unchanged; SCENARIO ASSUMPTION boundary, not a completeness claim): supervision/human-oversight labor cost, robot software/subscription fees, insurance, workspace modification, consumables, and downtime/unplanned-repair time beyond the flat maintenance reserve. None of the four cited platforms publish data allowing these to be estimated from a primary source as of 2026-08-12; adding them would only push cost-per-hour higher, so all figures above should be read as a floor, not a full commercial cost.

Part 8 — Where this connects to the wider chain (interpretation only)

INTERPRETATION: this workbook extends the whitepaper’s core chain — Capital + energy + maintenance + utilisation + orchestration → humanoid physical working capacity — but stops at “capacity,” deliberately not continuing into “work → outcome → value.” A robot costing $1/hour to own tells you nothing about whether the work it performs is worth $1, $100, or nothing at all; per the whitepaper’s value framework, working capacity is not value, and more hours of operation are not automatically better. Any use of these cost-per-hour figures to argue a specific business case, investment decision, or procurement choice would be going beyond what this workbook supports — this is not investment, procurement, tax, legal, or policy advice, and the figures above are illustrative arithmetic on stated, editable assumptions, not a forecast of any real deployment’s economics.

Full input register for this workbook (for editing)

This workbook (Part 1 real-price sourcing, Parts 2–6 arithmetic, Parts 7–8 scope/interpretation) is a standalone numeric asset for the whitepaper “Why Are They Spending Trillions on AI?” (CC BY 4.0, Valto Loikkanen).