Batteries (high-C)

Batteries are the one part in this report that wears out, and the one part that, on raw volume, never binds anywhere. A humanoid carries a small pack: about two-and-a-third kilowatt-hours, the size of Tesla Optimus’s Gen2 cells¹. Because the robot is cycled hard every day, that pack lasts only a few years before it is swapped⁷, so we model it as a consumable: the energy the worldwide humanoid fleet needs is its initial fitment plus a growing wedge of replacements⁸. Even with that replacement wedge, the total is tiny, about two hundred and ten gigawatt-hours by 2040. Set that against a world that already makes batteries by the terawatt-hour: global cell output was about one thousand five hundred gigawatt-hours in 2024 and is climbing toward eight or nine thousand a year⁴⁶. The humanoid slice is a rounding error on an electric-vehicle-scale base, so cumulative production never falls behind cumulative need, the global crossover never arrives, and neither does China’s nor even the United States’ alone (all three crossover years are legitimately null²). The bind is a different kind entirely. It is concentration: China makes about seventy-six percent of the world’s cells⁵ and dominates the cathode, anode and refined-material chain beneath them¹¹. The risk is not running out of gigawatt-hours; it is that almost every one of them, today, is Chinese.

The battery is the robot’s energy store, a lithium-ion pack of roughly two-and-a-third kilowatt-hours, the figure published for Tesla Optimus’s Gen2 unit (a 2.3 kilowatt-hour pack at fifty-two volts)¹. In the model that is about 0.0000023 gigawatt-hours of cells per robot, which is why batteries are tracked here in energy (gigawatt-hours), not in unit count. What makes this component unique is that it is the report’s only consumable. Every other part, a roller screw, a magnet, a coreless motor, is built into the robot and lasts its whole service life, so the fleet’s need is a one-time durable stock. A battery is not: a working humanoid runs only one-and-a-half to four hours per charge and is recharged daily, so the pack is cycled relentlessly⁷. High-rate liquid lithium-ion cells deliver only about one thousand to one thousand five hundred usable cycles before they fall below useful capacity, which under daily charging is roughly two to four years of service⁸. We therefore set a four-year mean service life, a fleet average that allows heavy-duty packs to wear out nearer two-to-three years and gentler-duty or lithium-iron-phosphate packs to last five-to-six, all cycle-limited well inside the eight-to-ten-year calendar life. So a battery is not bought once per robot; it is bought again and again across the robot’s life.

For a durable part the fleet’s need is simple: robots multiplied by the per-robot count. For a consumable it has two terms. The first is initial fitment, the worldwide installed fleet times two-and-a-third kilowatt-hours each. Integrating the consensus shipment ramp², that fleet reaches about five hundred and eighty-five thousand robots by 2030, ten million by 2035 and forty-seven million by 2040, so fitment alone is roughly one, twenty-three, and one hundred and eight gigawatt-hours of cells respectively. The second term is the replacement flow: each year about one-quarter of every pack already in service, one over the four-year life, is re-bought as the old one wears out⁸, and those replacements accumulate. The wedge grows fast: by 2040 the worldwide fleet needs about two hundred and eight gigawatt-hours in total, of which roughly one hundred and eight is fitment and about ninety-nine is replacements, nearly half the need is re-buys, because a large standing fleet re-buying its packs every four years starts to rival the trickle of genuinely new robots. In the chart below, the black line is that worldwide cumulative need, fitment plus replacements, the grey band is cumulative global cell production, and there is simply no red wedge, because output runs hundreds of times ahead of need.

Production is the only thing that differs between the Global, China and US views. The worldwide humanoid fleet and the cells it needs stay identical in all three; only the supply line changes, asking whether one bloc’s factories alone could equip the whole global fleet. A caution on the numbers: for batteries, announced capacity and actual output diverge sharply. Global nameplate passed three terawatt-hours in 2024 and four by the end of 2025, and could triple again this decade if every announced plant is built³, but the industry is structurally over-built, running near sixty to seventy percent utilization on a base roughly one-and-a-half times larger than demand⁹. So these waypoints track real output, not nameplate. Global cell output runs from about one thousand five hundred gigawatt-hours a year today⁴ toward roughly four thousand two hundred by 2030 and six thousand eight hundred by 2035, the electric-vehicle-led demand path⁶, then softening to a modeled eight-to-nine-thousand plateau as vehicle penetration matures; the post-2035 figures are a modeled extrapolation, not a reported forecast. China is about seventy-six percent of that, some one thousand one hundred and seventy gigawatt-hours of cells in 2024, led by CATL at roughly thirty-eight percent of the global market and BYD at about seventeen, with nine of the world’s top ten cell-makers Chinese⁵. The United States is the thin line: about eighty-five gigawatt-hours a year today, anchored by Tesla and Panasonic’s Nevada plant with Ford, General Motors and SK lines ramping, roughly five percent of global cells¹⁰, growing under federal incentives toward a few hundred gigawatt-hours by 2030, but staying far below its announced nameplate at low early utilization. Every one of these curves towers over the humanoid need.

It never does, and that absence is the finding, not an error. The crossover, the year cumulative production drops below cumulative need², simply does not occur in any region. By 2040 the worldwide humanoid fleet needs about two hundred and eight gigawatt-hours of cells in total; cumulative global output by then is on the order of eighty-seven thousand, more than four hundred times the need. Even cumulative China-only output (about sixty-four thousand) and even cumulative US-only output (about six thousand, still some thirty times the need) stay comfortably ahead across the whole horizon. So all three crossover years are legitimately null². This holds even though batteries are a consumable: the replacement wedge nearly doubles the fleet’s need by 2040, yet doubling a rounding error is still a rounding error against an electric-vehicle battery base measured in thousands of gigawatt-hours a year⁴. A humanoid pack is small¹, robots are few next to cars, and the world is building cell capacity it cannot fully use⁹. And the replacement model is deliberately conservative: it bills a fixed share of the installed packs as replaced every year, which slightly front-loads replacement demand and overstates the need during the fleet’s growth, and a shorter real-world pack life than four years would only enlarge the wedge, never open a deficit, because volume runs hundreds of times ahead either way. The honest verdict is that volume is not the wall here. The wall, if there is one, is where those gigawatt-hours are made, and that is a concentration problem, examined next.

Unlike the precision-mechanical links, batteries are not rate-limited by a grinder or a winding line, a cell plant is genuinely elastic, and the world has more of them than it needs⁹. The exposure is concentration, and it deepens as you descend the chain. At the cell itself, China makes about seventy-six percent of global output⁵, and the IEA puts it at over three-quarters of all batteries sold³, CATL and BYD alone are more than half the global market, and nine of the top ten makers are Chinese⁵. One level down, the concentration is worse: China dominates the midstream processing of the materials inside every cell, refined lithium, anode graphite and cathode active material, so even a non-Chinese cell line depends on Chinese-refined inputs¹¹. That is the categorical difference from a part that runs short on tonnage. There are plenty of gigawatt-hours; the question is whether a humanoid programme outside China can secure cells and the refined feedstock beneath them without routing through a single country that controls roughly four-fifths of the supply. The United States is building a thin domestic slice, about five percent of cells today, ramping on federal incentives¹⁰, and that slice, though it never falls behind the small humanoid volume, is the strategically meaningful one: it is the only non-Chinese capacity a Western fleet could lean on, and it is small. The bind, then, is not the four-series gap, which never opens. It is that the supply this component depends on is concentrated almost entirely in one bloc, upstream and down. One note on the other embodiments: of the non-humanoids in this report, quadrupeds and drones are the ones that carry a battery. A quadruped’s is small, about three-tenths of a kilowatt-hour, mostly the little packs in untethered consumer units like Unitree’s Go2. Drones are the paradox: they are by far the largest embodiment by units, tens of millions a year, and a drone is defined by its battery, yet the packs are tiny (a fleet-blended ~60 watt-hours, from a ~20-Wh consumer mini to a ~1.5-kWh agricultural sprayer), so the whole drone fleet draws only one-to-two gigawatt-hours a year, well under two-tenths of one percent of the electric-vehicle-dominated ~1,200-GWh lithium market. So the biggest robot fleet in the world is immaterial to the volume verdict here, though it too rides the same China-concentrated cells.