system
A robot's battery doesn't just power it. It decides how long it can work and how much it can carry, which really means deciding what kind of robot it can be.
电池、热管理与电源。能量决定占空比与载荷，也就决定了机器人形态的可行域。
Bottlenecks瓶颈 P15 Battery, thermal, power电池、热管理与电源
What it isHeart is the energy layer: battery, thermal, and power management for mobile and humanoid robots. It doesn't just power the robot. It decides its duty cycle, how long it can run before it needs to recharge, and how much it can carry. More battery means more runtime, but every extra cell is weight and space that could have gone to payload, actuators, or compute instead. That tradeoff sets the feasible physical shape of the robot: what kind of machine is even possible to build for a given job. It's the central bottleneck here (P15, Battery, thermal, power).
是什么“心脏”是能量层：面向移动与人形机器人的电池、热管理与电源管理（P15，电池、热管理与电源）——这一系统决定占空比与载荷，进而决定机器人形态的可行域。
ThesisHeart ranks 12th of 18 on CE-N, in the weaker half of the board, not the worst case but not an attractive one either. The reason is straightforward: the economics here mostly don't belong to robot companies at all. Component economics accrue to scaled battery-cell suppliers like CATL (a battery maker whose real business is EV and energy storage) rather than to robot OEMs, the companies that actually design and sell the finished robot. That's why CATL's own robotics exposure shouldn't be valued separately from its core business. The CATL/Galbot relationship makes the problem concrete: CATL is simultaneously the cell supplier, the plant host, and the strategic partner for the same product, which means there's no independent buyer anywhere in the chain to validate any performance claim. Until an audited robotics segment exists, CATL should be treated as a battery company, nothing more.
论点“心脏”在 CE-N 排名中位列 18 个瓶颈中的第 12，这一层的核心论点也异常简单：零部件层面的经济性归属于宁德时代这类规模化电芯供应商，而非机器人整机厂。宁德时代自身的机器人敞口不应被单独估值——它的数字属于电动车与储能业务，而非人形机器人。宁德时代与 Galbot 之间的结构进一步印证了这一纪律：宁德时代同时是电芯供应商、工厂东道主与战略合作伙伴，整条链条上没有独立买家来验证任何性能宣称。在出现经审计的机器人业务分部之前，应把宁德时代视为一家电池公司，仅此而已。
Products mapped to this system映射到本系统的产品
Lithium-ion and early solid-state cell, pack, and battery-management-system suppliers whose on-site profiles name robot or robot-adjacent (AMR, humanoid) applications. This is a cluster of mature EV/consumer-battery suppliers extending into robotics, not a dedicated robotics-battery category -- no cycle-life, yield, or cost verification for robot use has surfaced for any of them.
6 companies 6 家公司 4 manufacturers 4 家制造商
A charge-pump power-conversion IC explicitly framed for dexterous-hand robotics and mechatronic systems -- the only on-site power-electronics product with a stated robotics application; no thermal-management supplier for robot joints or edge compute is present in the site's data.
1 companies 1 家公司
Per the conference doc's company map, two names were surfaced for this system, both explicitly flagged as secondary exposure rather than primary Heart players -- one for power-management/power-electronics content that is secondary to its main business, the other because its battery content is part of a broader integrated component stack rather than a dedicated battery product. Neither is mapped to this site's heart system as a primary company; listed here only to preserve the doc's finding, not as a claim that either is a dedicated Heart-layer supplier.
+2 named in conference materials, unverified +2 项来自会议材料，未经核实
Companies mapped to this system映射到本系统的公司
Manufacture mapped to this system映射到本系统的制造
The clearest evidence from this beat is not a number, it is a structure. CATL now supplies the battery cell to Galbot’s S1 humanoid, hosts Galbot’s flagship deployment inside a CATL manufacturing plant, and has signed a global strategic agreement to commercialise Galbot internationally — meaning CATL is simultaneously the supplier, the customer, and the strategic partner for the same product, and every performance figure attached to it (an “8-hour” runtime claim, for instance) is a supplier’s claim about its own cell in its own building, with no independent buyer anywhere in the chain.
This is not unique to Heart: the identical pattern — Schaeffler as both investor in and flagship customer of Agility Robotics; Foxconn leading Agility’s PIPE while its own subsidiary sits on this taxonomy’s factory map — recurs across nearly every system this research covers, but Heart is where it is most visible, because there genuinely is no third party in the loop at all.
On the two named public and quasi-public P15, Battery, thermal, power proxies, nothing has moved: Amprius remains the only pure-play name where energy density could plausibly change a robot’s feasible design, and the taxonomy’s own verification list for it — cycle life, yield, cost — remains unanswered a full research cycle later.
Sell-side coverage of this beat is itself thin and mostly borrowed from adjacent themes rather than written about this problem directly.
Minth and Sanhua, both China-based auto-parts suppliers pivoting existing manufacturing lines toward humanoid robotics, illustrate the pattern: Minth raised its own 2026 humanoid-revenue guidance from RMB100m to RMB500m and its 2027 guidance from RMB500m to RMB1bn, but the same analyst note recording that guidance also states that peers Tuopu and Sanhua “have stopped giving humanoid-robotics revenue guidance” — a lone name still guiding into an unaudited line while its own peers pull back is a caution about the guidance, not a green light.
One new, genuinely on-topic private name did surface: Nyobolt, a fast-charging battery technology company, raised $60m at a valuation above $1bn explicitly framed around mobile-robot-fleet charging needs — a funding-round valuation, nothing more, and held to the same “committed capital is not received capital” discipline applied everywhere else on this site.
A UBS note ties physical-AI feasibility broadly to progress in energy density and efficiency, naming Amprius among its covered supply-chain names, but adds no new cycle-life, yield, or cost evidence beyond what the primary research above already carries. No thermal-management supplier for robot joints or edge compute appears anywhere in this batch either — the AI-data-centre liquid-cooling and industrial-hydraulics content that does surface is an adjacent market, not evidence the robot-thermal gap is being addressed, and conflating the two would be a category error.
Do not value CATL’s robotics exposure, for the reason the taxonomy already gives, but read the CATL/Galbot structure as a template rather than a special case — it is the cleanest illustration on the entire board of demand that may be substantially circular, generated by the same balance sheet that is supplying the capital and the components.
The sell-side batch corroborates caution rather than relaxing it: the clearest new datum is Minth guiding a specific number while its own peers un-guide the same line, exactly the kind of unaudited, hype-sensitive disclosure environment that justifies refusing to value CATL’s robotics exposure until an audited segment appears.
No thermal-management supplier for robot joints or edge compute appears anywhere in the current company map — a real gap, not just an unvalued one.
这一赛道最清晰的证据不是一个数字，而是一种结构。宁德时代如今向银河通用（Galbot）的 S1 人形机器人供应电芯，将 Galbot 的旗舰部署场地设在宁德时代自己的制造工厂内，并签署了一项全球战略协议，负责 Galbot 的国际化商业化——也就是说，宁德时代同时是同一产品的供应商、客户与战略合作伙伴，任何附带其上的性能数字（例如”8 小时”续航的说法）都只是供应商对自家电芯在自家厂房内表现的自我陈述，链条中根本不存在独立的第三方买家。
这并非”心脏”系统独有：Schaeffler 同时是 Agility Robotics 的股权投资者与其旗舰客户；富士康领投 Agility 的 PIPE 融资，而其子公司本身就是本分类法工厂地图上的一个节点——这一相同模式几乎在本研究覆盖的每个系统中都会重现，但在”心脏”系统中表现得最为明显，因为这里确确实实完全不存在任何第三方。
至于另外两个具名的上市及准上市 P15，电池、热管理与电源 代理公司，情况均无变化：安普瑞斯（Amprius）仍是唯一一家能量密度有可能切实改变机器人可行设计的纯正标的，而本分类法自己列出的核查清单——循环寿命、良率、成本——在整整一个研究周期之后依然没有答案。
这条赛道本身的卖方研究覆盖十分单薄，且大多借用自邻近主题，而非直接针对这一问题撰写。
敏实集团与三花智控这两家把现有产线转向人形机器人的中国汽车零部件供应商，正好说明了这一模式：敏实把自己 2026 年的人形机器人营收指引从 1 亿元人民币上调至 5 亿元人民币，2027 年指引从 5 亿元上调至 10 亿元，但记录这一指引的同一份分析师笔记也指出，同行拓普集团与三花智控”已停止给出人形机器人营收指引”——一家仍在为一条未经审计的业务线给出具体指引，而同行却在后撤，这是对该指引本身的警示，而非放行信号。
一个全新的、真正切题的私营标的确实浮现：Nyobolt，一家快充电池技术公司，以超过 10 亿美元估值募集 6000 万美元，明确围绕移动机器人车队的充电需求展开——这只是一轮融资的估值，仅此而已，需按本站在其他地方一以贯之的”承诺资本不是已收资本”原则来对待。
瑞银一份报告将物理 AI 的可行性广泛地与能量密度及效率的进展挂钩，在其覆盖的供应链名单中提到了安普瑞斯，但除了上文一手证据已经掌握的信息之外，并未提供新的循环寿命、良率或成本证据。这批研究中同样没有出现任何面向机器人关节或边缘计算的热管理供应商——其中出现的 AI 数据中心液冷与工业液压内容属于邻近市场，并非机器人热管理缺口正在被解决的证据，把二者混为一谈将是一种范畴错误。
不要为宁德时代的机器人敞口估值，理由与本分类法此前给出的一致；但应把宁德时代与 Galbot 的这套结构当作一个范本，而非特例——它是整张表里对”需求可能在很大程度上是循环的”这一现象最干净的例证，这种需求由同一张资产负债表同时提供资本与零部件而催生。
卖方研究批次印证的是保持谨慎，而非放松警惕：最清晰的新数据点是敏实给出一个具体数字指引，而其同行却在为同一条业务线撤回指引——这正是那种未经审计、对炒作敏感的披露环境，恰恰印证了在审计分部出现之前拒绝为宁德时代机器人敞口估值是正确的。
目前的公司地图上，找不到任何一家面向机器人关节或边缘计算的热管理供应商——这是一个真实的空白，而不仅仅是一个未被估值的空白。