【钠瞳 Sodium Pupil】一：为什么是这件事

一、为什么是这件事

1.1 算电协同很贵，野外节点却是最便宜的验证场

以大模型为代表的人工智能计算正在重塑能源格局。数据中心单机柜功率从早年的十余千瓦，快速提升到百千瓦量级，并向兆瓦级推进。算力基础设施的约束已经从"芯片够不够"，变成"电够不够、稳不稳"。

但在机房里验证算电协同，成本极高：要先建机房，再搭测试台；收益要折算到全生命周期度电成本；反馈周期以季度或年计。

而一台挂在树上的野外相机，恰恰具备算电协同的全部要素：

能量闭环——一块电池加一块光伏板，就是它的全部预算；
任务价值明确——每一次识别与回传，都有可计价的收益；
反馈迅速——一天就能看出策略差异。

它是算电协同最小、最廉价、也最容易验证的闭环。

1.2 三种死法，其实是一种病

我们在稿纸上把野外相机的死法分了三类：

死法	机制	后果
冻死	夜间低于 -30℃ 时容量骤降；锂电低于 0℃ 禁止充电	白天有太阳，电也充不进去
累死	风雪、枝叶、晒热的石头都会触发全尺寸连拍	几千张废帧，淹没有效画面
过放死	电量跌到阈值后硬性关机	需人工上山复位，最关键那一晚恰好不在线

三种死法表面上不一样，其实是一个病：它不会做取舍。它知道自己快没电，却不知道该把最后的电，花在哪一件事上。

1.3 为什么是现在

钠离子电池已经从"实验室新闻"走进规模化。宁德时代与海博思创签下三年 60 GWh 的储能钠电战略合作，并计划于 2026 年底规模化量产（文献[1]）。储能行业采购钠电，图的正是同一组特性：宽温、耐过放、高倍率、免维护。

而把这四个词念一遍就会发现——它们也正是那台挂在树上的相机最需要的东西。

I. Why This Problem

1.1 Compute-power coordination is expensive to prove, while a field node is the cheapest testbed

Artificial intelligence computing, represented by large models, is reshaping the energy landscape. The power of a single rack in a data center has risen from roughly ten kilowatts in earlier years to the hundred-kilowatt level, and is moving towards the megawatt level. The binding constraint on computing infrastructure has shifted from "are there enough chips" to "is there enough power, and is it stable".

However, verifying compute-power coordination inside a data center is very expensive. A machine room has to be built first and a test bench set up afterwards; the benefits have to be converted into a levelized cost of energy over the full life cycle; and the feedback cycle is measured in quarters or years.

A wildlife camera hanging on a tree, by contrast, contains every element of compute-power coordination:

An energy closed loop — one battery plus one small solar panel is its entire budget;
Clearly priced task value — every identification and transmission has a value that can be accounted for;
Fast feedback — a single day is enough to reveal the difference between strategies.

It is the smallest, cheapest and most verifiable closed loop of compute-power coordination.

1.2 Three ways to fail, one root cause

On paper we sorted the ways a field camera can fail into three categories:

Failure	Mechanism	Consequence
Frozen to death	Capacity collapses below -30℃ at night; lithium cells are forbidden to charge below 0℃	The sun is up, but the energy cannot get in
Worn out	Wind, snow, branches and sun-heated rocks all trigger full-resolution bursts	Thousands of useless frames drown the useful ones
Deep-discharge death	The device shuts down hard once the charge falls below a threshold	Someone has to climb the mountain to reset it; the most important night is exactly the one it is offline

These three look different, but they are the same illness: the device cannot make trade-offs. It knows its battery is running out, but it does not know which event deserves the last of its energy.

1.3 Why now

Sodium-ion batteries have moved from "lab news" into volume production. CATL has signed a three-year, 60 GWh strategic cooperation agreement on sodium-ion batteries for energy storage with HyperStrong, and plans volume production by the end of 2026 (Ref. [1]). The energy storage industry is buying sodium-ion batteries for exactly one set of properties: wide temperature range, tolerance of deep discharge, high rate capability, and maintenance-free operation.

Say those four words again and you will notice that they are precisely what the camera hanging on that tree needs most.