The Fifteen-Point-Six-Terawatt Number Is Real Arithmetic Built on the Wrong Fleet Size

The Real Number Comes First: A Fifth to a Half of the Grid, Not Five Times It

State the finding before the arithmetic behind it, because this is the fourth entry in a five-part audit of the same viral post and the shape of the correction is, by now, a known quantity: a fleet sized to population instead of labor-hours. Run at 2024’s actual global electricity data, the robot economy’s real added demand lands between 21 and 52 percent of today’s average generation, centered near 30 percent — a genuinely large new claim on the world’s grid, and nothing close to the post’s own “505%,” five-times-the-grid framing.

What makes this entry worth reading past its own headline number is not the size of that gap. It is where the error turns out to sit. Unlike this series’ other audited figures, the post’s electricity claim is built on a per-unit wattage that is not invented, rounded, or reverse-engineered to sound alarming — divide the post’s own two numbers into each other and the implied draw per robot is a specific, physically ordinary figure that checks out against real industrial hardware. The 15.6-terawatt total is wrong for the same reason every other number in this audit has been wrong: an unstated, population-scale fleet size multiplying an otherwise defensible per-unit figure. The rest of this entry shows that arithmetic in full, starting with the one part of the post’s own claim that survives scrutiny.

Sixteen Billion Times Nine Hundred Seventy-Five Watts Equals Fifteen-Point-Six Terawatts, Exactly

Among the five headline numbers a widely repeated, unattributed post attaches to a coming robot economy, one names an electricity figure: a fully robotic fleet would draw 15.6 terawatts of new demand, described elsewhere in the same claim as a “505%” jump against the grid’s current output. Like the other four figures in this series’ audit, the post itself cannot be located under any search phrasing tried for this piece — no platform, no account, no date attaches to it in any retrievable form. It is treated here exactly as this series treats its other four numbers: an unverified claim stated in a commissioning brief, not a quote from a named source, and never attributed to one.

Divide the post’s own two numbers into each other before doing anything else to them, and something worth taking seriously happens before anything worth doubting does. Fifteen-point-six terawatts of demand attributed to sixteen billion units implies a continuous average draw of exactly 975 watts per unit: 15,600,000,000,000 W ÷ 16,000,000,000 units = 975 W. That is not a round, invented-looking figure of the kind that shows up when a number has been reverse-engineered to sound alarming. It sits inside the real operating envelope of a machine doing continuous physical work — well above a resting sensor package, nowhere near a car engine, in the general neighborhood of a hand power tool run continuously or a small industrial appliance. Whoever assembled the post’s numbers performed a real division against an input that looks like a real spec, not a number invented to produce a scary terawatt figure and then reverse-engineered into a plausible-sounding wattage.

That is precisely what makes this entry different from a simple debunking, and precisely why it belongs in this series rather than dismissing the post out of hand. The question is not whether 975 watts is a defensible number for a robot to draw — it plausibly is. The question is whether either of the two inputs multiplying up to 975 watts and 15.6 terawatts — the sixteen-billion-unit fleet, or the per-unit draw itself — describes anything a real electrical grid would actually have to supply. This series’ opening entry already found that the post’s fleet size assumes a two-robots-per-human population-replacement ratio rather than a labor-hours-replacement one, and that the labor-hours-corrected fleet lands roughly an order of magnitude lower. This entry runs that same correction through the electricity ledger specifically, using primary 2024 generation and capacity data, and asks what 975 watts per unit actually costs the grid at each fleet size — population-scale, and labor-hours-scale.

To size the claim before doing anything else to it: 15.6 terawatts of continuous new demand is, by itself, more than twelve times the United States’ entire utility-scale generating capacity — 1,280,799 MW, or about 1.28 TW, at the end of 2025, per the U.S. Energy Information Administration ^[2] — and larger than the record 585 gigawatts of new renewable capacity the entire world managed to add in all of 2024 combined, by a factor of about 27 ^[4]. That comparison is a scale check, not this series’ historical-buildout audit, which belongs to a separate entry in this project; it exists here only to make concrete what “15.6 terawatts” is actually asking the physical world to supply before any percentage gets attached to it.

What the World’s Grid Actually Generated in 2024, Measured Two Ways

A terawatt figure needs something real to compare against, and two different real numbers answer two different questions. The first is how much electricity the world actually generated over a full year, expressed as an average continuous rate; the second is how much generating equipment the world actually has installed, expressed as a nameplate ceiling. Both come from institutions with no stake in a viral robot-economy claim, and both were checked directly for this piece rather than carried forward from an earlier pass.

Global electricity demand reached 30,856 terawatt-hours in 2024, up 1,172 TWh — a 4 percent increase — over 2023, crossing 30,000 TWh for the first time on record, per Ember’s Global Electricity Review 2025 ^[3]. The IEA’s own independently tracked figure for the same year agrees closely without matching to the decimal: “global electricity generation grew by over 1,200 TWh in 2024… this annual increase of 4% represents a significant acceleration from the average growth rate of 2.6% seen between 2010 and 2023” ^[1]. Two separate institutional trackers, using different country-coverage methodologies, converge on the same event — a four-percent, roughly-1,200-TWh jump — without converging on the same digit. That small, honest gap between two credible primary sources is worth holding in mind; a similar but far larger gap is exactly what this piece finds later inside the post’s own numbers.

Convert Ember’s 30,856 TWh into a continuous average rate the way the post’s own 15.6-terawatt figure is expressed, and the arithmetic is simple: 30,856 TWh ÷ 8,760 hours in a year = 3,522 gigawatts, or 3.52 terawatts of average continuous global generation. That is the average rate — real demand swings well above and below it hour to hour — but it is the correct like-for-like comparison for a continuously-drawing robot fleet’s own average demand, which is likewise expressed here as a constant rate rather than a peak.

The second number, installed capacity, answers a different question: not how much the world’s plants generated on average, but how much generating equipment exists to draw on at all, including everything that sits idle outside of peak demand. IRENA’s Renewable Capacity Statistics 2025 reports that “by the end of 2024, total renewable capacity had reached 4.4 TW, representing 46% of global installed power capacity” ^[4], with IRENA’s own release of that report giving the precise figure as “4 448 gigawatts (GW)” ^[5]. Dividing the renewable share by its own stated percentage of the whole gives the whole: 4,448 GW ÷ 0.46 = roughly 9,670 GW, or 9.67 terawatts of total installed generating capacity worldwide, across every fuel and technology combined, at the end of 2024. That 9.67 TW ceiling and the 3.52 TW average generation rate imply a global capacity factor — the share of nameplate capacity actually running, averaged across the whole fleet and the whole year — of about 36 percent, a believable blended figure for a mix that still runs substantial coal, gas, and hydro alongside a fast-growing but intermittent renewable share. The two numbers check out against each other; neither is a stray or contradictory figure.

The Post’s Own Number Reproduces Exactly; A Labor-Hours-Corrected Fleet Does Not Reproduce It At All

With 975 watts per unit established and 3.52 TW / 9.67 TW established as the grid’s two real baselines, the post’s own scenario is trivial to confirm: 16,000,000,000 units × 975 W = 15,600,000,000,000 W = 15.6 TW, exactly the figure the post states. This is not a finding so much as a consistency check — it shows only that the post’s 15.6-terawatt figure and its (implied, never stated outright) 16-billion-unit fleet were built to be consistent with each other, using a wattage that happens to be realistic. It says nothing yet about whether the fleet size is realistic.

That fleet size is where this series’ opening entry already did the necessary work, and the correction is inherited here rather than rebuilt from scratch: 16 billion units treats the robot fleet as a straight two-for-one replacement of the roughly 8.2 billion people alive on Earth, not of the smaller population that actually performs paid or productive labor. Rebuilding the fleet size from labor-hours instead of headcount needs three real inputs, checked fresh for this piece: global labor force, average hours a worker actually works in a year, and the fraction of a calendar year a continuously-operated machine can realistically stay in service.

The World Bank’s most recent modeled estimate puts the global labor force at 3,736,625,386 people for 2025 — 3.74 billion — using the standard ILO definition of people either employed or actively seeking work ^[6]; this is the identical figure, at the identical vintage, this series’ opening entry already established, checked again independently for this piece rather than carried forward unverified. Average hours actually worked per worker has no single official global aggregate — the OECD tracks it for its own membership, not for the larger non-OECD majority of the world’s labor force — but the documented national range runs from Germany’s 1,343 hours a year up to Mexico’s 2,207 and Costa Rica’s 2,171, with the OECD’s own 38-country average sitting at 1,683 hours ^[7]. This piece uses the same 1,600-to-2,200-hour bracket this series’ opening entry already established as its own stated modeling choice, not a single published figure, centered above the OECD average because most of the world’s labor force works inside the higher-hours, less-unionized labor markets the OECD’s own membership underrepresents.

A robot’s usable hours depend on a different variable: duty cycle, the fraction of the 8,760-hour year a unit spends actually working rather than charging, undergoing maintenance, or switching tasks — run here across the same 90/70/50 percent grid this series applies throughout, giving 7,884, 6,132, and 4,380 operating hours per unit per year respectively. Solving the labor-hours ledger,

for fleet size $F$ across all nine combinations of the labor-force total $L$, the hours-worked bracket $H_w$, and the duty-cycle-adjusted operating hours $H_r$, produces a range from about 0.76 billion units (90 percent duty cycle, 1,600 hours worked) to about 1.88 billion units (50 percent duty cycle, 2,200 hours worked), with a representative central case — 70 percent duty cycle, 1,750 hours worked — landing at roughly 1.07 billion units. That is the identical nine-scenario grid this series’ opening entry already published, reproduced here from the same primary inputs checked fresh rather than a new range of its own. Every cell in that grid sits at least eight times below the post’s sixteen billion, several cells above twenty times below it — the same order-of-magnitude gap this series’ opening entry already established.

Fifteen-Point-Six Terawatts at Population Scale; Something Much Smaller at Labor-Hours Scale

Apply the identical 975-watt per-unit figure — unchanged, taken directly from the post’s own implied number rather than a new assumption — to that labor-hours-corrected fleet instead of the post’s population-scale one, and the terawatt figure collapses by the same order of magnitude the fleet size did.

Two things about that table are worth stating plainly rather than letting the numbers speak alone. First, the post’s own scenario does not merely strain the grid — at 15.6 TW of continuous added demand against 9.67 TW of total installed capacity across every plant on Earth, the post’s implied load exceeds the entire existing global generating fleet, of every fuel and technology combined, by about 61 percent. That is not a load a grid absorbs by running its existing plants harder; it is a demand for more new continuous generating capacity than currently exists, built and commissioned, on some unstated timeline the post never addresses at all. Second, the labor-hours-corrected range never gets close to that. Even at the corrected grid’s own highest-demand cell — a conservative 50 percent duty cycle paired with the top of the hours-worked bracket — the added load is 52 percent of current average generation and under 19 percent of total installed capacity: a large, real infrastructure commitment, comparable in scale to running a meaningful fraction of today’s global data-center buildout in parallel, but not remotely a fivefold multiplication of the world’s electricity system.

This entry’s own falsifiable threshold, stated before running the grid rather than fitted to it afterward: if the corrected fleet’s added demand fell below 10 percent of current average generation at every duty cycle tested, the “serious load” verdict would fail toward negligible; if it exceeded 100 percent at any cell, it would fail toward “still a crisis even after correction.” Neither happens. The full corrected range — 21.0 to 52.0 percent of current average generation — clears the low bound comfortably and stays well under the high one across every one of the nine underlying scenarios, not merely at the central estimate.

Five Hundred and Five Percent Fits Neither of the Two Honest Ways to Read It

The post’s stated “505%” figure deserves its own scrutiny, separate from the fleet-size question, because “505%” is not actually one number — it is an answer to an ambiguous question, and the post never states which question it is answering. Take the post’s own 15.6 TW of added demand against the 3.52 TW of average current generation this piece just derived, and there are exactly two textbook-correct ways to express that as a single percentage, depending on what “percent” is meant to describe.

Read the added load alone as a percentage of the existing baseline — “how much bigger is the new demand than what already exists” — and the answer is 15.6 ÷ 3.52 = 442.9%. Read the new total demand, existing plus added, as a percentage of the existing baseline instead — “how big does the whole system become, expressed against where it started” — and the answer is (15.6 + 3.52) ÷ 3.52 = 542.9%. Both are correct arithmetic on the same two underlying numbers; they simply answer different questions, and English usage genuinely does not settle which one “a 505% increase” or “505% of the grid” is supposed to mean without more context than the post provides. The post’s own “505%” sits in the gap between them — 62 percentage points above the added-load reading, 38 percentage points below the total-system reading — closer to, but distinctly short of, the total-system figure, and matching neither reading exactly.

This is not a case where this piece can point to the post’s exact arithmetic and show where a specific error was introduced — the post itself remains unlocatable, and nothing here should be read as reverse-engineering its author’s actual calculation. What can be shown, cleanly, is that the single most natural pair of ways to turn this piece’s own real, sourced 15.6-TW-against-3.52-TW comparison into a headline percentage produces two different, both-defensible numbers 100 percentage points apart, and the post’s own figure sits inside that gap rather than matching either endpoint. A round, confident-sounding percentage attached to an unstated definition is exactly the kind of number that survives a decontextualized repost while quietly picking up authority it never earned — precise-sounding without being precise, in a way that “roughly quintupling the grid” would not have concealed nearly as well.

It is worth naming, too, what the post’s number gets structurally right even inside that ambiguity: 15.6 TW added to 3.52 TW is 19.12 TW total, a factor of 5.43 — within about 9 percent of a clean, literal fivefold multiplication of the grid’s current average output. “Roughly quintupling global electricity generation” is a defensible, if extreme, one-sentence gloss on the post’s own population-scale numbers. “A 505% increase,” stated without saying increase-in-what, is not the same claim, and reads as considerably more precise than either honest calculation actually is.

A Robot That Lifts and Carries Draws More Than a Robot That Stands and Waves

None of this settles the question the other direction, and this entry’s own strongest objection to its own “serious but non-catastrophic” verdict deserves the same arithmetic treatment as everything above it rather than a hand-wave. The 975-watt figure this whole comparison inherits from the post comes from dividing two already-stated numbers against each other, not from any disclosed robot’s actual measured power draw. If real task-equivalent robots — machines actually capable of the kind of continuous, load-bearing physical labor a human worker performs, not a light bipedal demonstration platform — draw meaningfully more than 975 watts on average, the labor-hours-corrected scenario could climb back up toward, or even past, the post’s own population-scale figure, even at a much smaller fleet size.

The disclosed numbers available today run lower than 975 watts, not higher, for every general-purpose humanoid platform this piece could confirm a figure for. Tesla’s Optimus carries a 2.3 kWh battery pack that a 2026 industry battery-life review states supports roughly eight hours of runtime ^[9]; separately, Tesla itself stated at its September 2022 AI Day unveiling that Optimus would draw roughly 100 watts sitting and roughly 500 watts walking briskly, a figure reported contemporaneously by Teslarati but never reissued in a later, formally published spec sheet, and flagged here accordingly — implying an average in the low hundreds of watts across a typical mixed workload ^[10]. Figure AI’s second-generation Figure 02 carries a company-disclosed 2.25 kWh battery pack, described in Figure’s own announcement as delivering “50%+ more energy” than its predecessor, though Figure’s release does not state a corresponding average wattage figure directly ^[11]. A standard fixed industrial robot arm, per one manufacturer’s own published figures, draws 0.25 to 0.5 kilowatts depending on reach and payload ^[8] — comparable to or below the post’s own 975-watt implied figure, not above it.

The one disclosed figure that runs meaningfully higher belongs to an older, hydraulically-actuated design rather than a current electric one: the DARPA Robotics Challenge-era Atlas, redesigned by Boston Dynamics around 2015, carried “an onboard 3.7-kWh lithium-ion battery pack” supporting “the potential for one hour of ‘mixed mission’ operation that includes walking, standing, use of tools, and other movements” ^[12] — implying an average draw near 3.7 kilowatts, nearly four times the post’s own 975-watt figure. That number is over a decade old, describes a hydraulic actuation architecture current commercial humanoids have moved away from precisely because it is power-hungry, and should not be read as representative of any robot being manufactured today. It does establish, honestly, that machines built to do continuous heavy physical work at humanoid scale have drawn multi-kilowatt continuous power before, when their actuation architecture demanded it.

Run the arithmetic the other direction and ask how high a per-unit draw would need to be to make the labor-hours-corrected fleet actually match the post’s own 15.6-terawatt figure. At the corrected grid’s smallest fleet, 0.76 billion units, matching 15.6 TW requires 15.6 TW ÷ 0.76 billion = 20,570 watts per unit — nearly 21 kilowatts, continuously, per robot. At the largest corrected fleet, 1.88 billion units, the requirement falls to 15.6 TW ÷ 1.88 billion = 8,310 watts per unit — still 8.3 kilowatts continuous. Even the lower end of that range is more than double the highest disclosed figure this piece could confirm for any humanoid platform, hydraulic or electric, and more than sixteen times the fixed industrial-arm figure. Closing the gap back to population-scale electricity demand through the per-unit-draw channel alone would require every unit in a billion-plus-unit fleet to continuously draw power in the range of a running gasoline lawnmower engine or a residential electric-vehicle fast charger, sustained around the clock — a real possibility for some individual heavy-industrial machines, but not a documented characteristic of any general-purpose robot platform on the public record as of this writing.

None of that rules the counter-case out on principle. A robot genuinely doing continuous heavy field labor — digging, hauling, forging, operating powered tools rather than walking and gesturing — plausibly does draw closer to a power tool’s or a small engine’s continuous rating than to a demonstration platform’s. A circular saw or an angle grinder alone draws 1 to 2 kilowatts while cutting; a robot built to run one continuously, plus the actuation to hold and aim it, plus locomotion, could reasonably stack well past 975 watts long before reaching anywhere near 8 to 21 kilowatts. What the arithmetic here actually forecloses is narrower and more precise than “heavy-duty robots use more power”: it forecloses the specific claim that ordinary task-heavy actuation, of the kind disclosed for every humanoid or industrial platform checked in this piece, closes the entire remaining gap back to population-scale electricity demand. The observable signature that would overturn this entry’s verdict is exactly stated, not hand-waved: a manufacturer disclosing a sustained, task-representative average power draw at or above roughly 8 kilowatts per unit for a general-purpose field robot, not a peak or a marketing figure. No such disclosure exists on the record as of this writing.

The Range That Survives Every Assumption Tested Here

Stated plainly, because a verdict without a stated way to break it is a competing assertion rather than an audit: this entry’s “serious but non-catastrophic” finding fails toward negligible if the labor-hours-corrected fleet’s added demand drops under 10 percent of current average generation at every duty cycle tested, and fails toward “still a crisis” if it exceeds 100 percent at any cell. Across the full nine-scenario grid this piece computed from primary 2024 data, the range runs from 21.0 to 52.0 percent — comfortably inside both bounds, not close to either edge.

What survives contact with real IEA, Ember, and IRENA data is not a single number to replace the post’s 15.6 terawatts and 505 percent, any more than the rest of this series has found single numbers to replace its other four claims. It is a bounded, arithmetic-shown range: a genuinely large new electricity demand, on the order of a fifth to a half of today’s global average generation, driven almost entirely by which of two fleet-size assumptions — population or labor-hours — gets fed into an otherwise-plausible per-unit wattage that the post itself, whatever its other flaws, did not invent out of nothing. The 975 watts checks out. The percentage the post built on top of it does not check out against either honest definition of the word “percent.” And the fleet size underneath both numbers is the one input this entire series keeps finding was never stated, never defended, and never as large as sixteen billion.

The consequence of taking that seriously runs past this one number. A power-grid claim is exactly the kind of figure a reader has no independent way to sanity-check — nobody carries a mental model of how many terawatts the world’s plants produce, the way a reader might sense that “sixteen billion robots” sounds like a lot of robots. That is what makes 975 watts per unit doing real, checkable work matter more than it might first appear: a viral number built on a genuinely plausible per-unit spec, multiplied by an unstated and unrealistic fleet size, is far more durable in circulation than one built on an invented spec would have been, because the one part a skeptical reader might think to check — does 975 watts sound like a robot — comes back looking fine. The fleet size was always the load-bearing error, on every one of this series’ five audited numbers, and it is the one number the post itself never had to state out loud for the terawatt figure to already carry it.