After Unitree pricing, warehouses look at costs first

The most valuable takeaway from this article is that Unitree has pushed embodied intelligence from a primary market narrative to the secondary market, where it's now being priced in. The issue price of ¥150.80, an issued market cap of ¥60.99 billion, and a P/E ratio of 219.23x have given this industry its first public anchor point. But an anchor isn't a magic key. It proves the market is willing to discount humanoid robots as potential next-gen general-purpose terminals, but it doesn't mean all embodied AI companies can be valued using the same logic.

I wrote a piece back in late July titled 'The Valuation Myth of Tech Maniacs,' focusing more on teams, funding, and narratives then. Looking at it now, my thinking is more grounded. Once you've actually run things in a warehouse, you realize what truly makes robots valuable is their ability to perform stable, repetitive tasks a thousand times amidst shelves, aisles, forklifts, cardboard boxes, and temporary pallets. Executing the motion is just the start; responsibility boundaries, exception recovery, maintenance costs, and client IT integration are the real deep waters of delivery.

From a business model perspective, companies like Unitree look more like terminal manufacturers selling platform imagination. Warehouse robotics looks more like process automation, selling cost, efficiency, and verifiable metrics. Whether clients keep buying depends on eliminating one picking station, reducing mis-shipments by one instance, or cutting inventory discrepancies by one count. The humanoid chassis itself doesn't command a premium. Have you calculated deployment costs? This question sounds basic, but it's critical. Hardware BOM is only the first layer; on-site implementation, SLAM calibration, point cloud map maintenance, network latency, safety zone false alarms, and SKU verification all cost money.

Recently, I mounted a robotic arm onto a wheeled chassis. I used the arm for a month, the chassis for three weeks, and just tried out the ADAM dual-arm robot for three weeks too. The headaches centered around shelf position deviations, floor reflections, and temporary goods blocking aisles. Getting the model to run is just the baseline. I spent a month with point clouds and ROS2; while I could build the pipeline, running it stably every day still requires robust exception handling. I've been using LLMs and GPT-5.6 Luna for a month, mostly to help the team clarify requirements, write proposals, and organize delivery checklists. I only started touching Multiverse and Mistral five days ago—I'm a beginner, so I wouldn't dare call myself an experienced user.

Secondary market pricing will first reward scarcity, then punish non-replicability. That's my most direct impression after watching Unitree's IPO. What's scarce are humanoid robot stocks that can go public, tell big stories, and attract industrial capital. What's non-replicable is every individual warehouse site. The primary market loves general-purpose embodied AI because the space is huge. Once the secondary market has an anchor, it starts dissecting: who has real revenue, who has gross margins, who has delivery capabilities, and who is just editing demos into PPTs.

For startup teams, the metric for execution will change. Previously, we looked at funding amounts, number of prototypes, and launch event videos. Going forward, we might need to see if a company can run a single warehouse smoothly for three months, how failure rates are tracked, how intervention counts are reduced, and whether there's dedicated delivery staff on-site. I mainly use large models to integrate toolchains and improve delivery efficiency. For example, WorkBuddy is good for creating master process sheets, but key SKUs must be manually verified. If inventory data is wrong, no matter how pretty the robot is, it's useless.

So, post-IPO, the pricing of embodied intelligence will stratify. Humanoid full machines, embodied brains, scenario-specific robots, and components won't share a single multiple anymore. The warehouse robotics line may not seem as sexy in the short term, but it's easier to generate cash flow first. My advice to startups is simple: don't rush to brand yourself as a general-purpose embodied AI company. Find a narrow scenario first and solidify the safety boundaries, data flows, and fallback mechanisms for exceptions. If clients can calculate the ROI clearly, then capital can calculate the valuation clearly.

The difficulty lies in moving from someone buying a prototype to someone repurchasing. Unitree gave the industry a price anchor; the warehouse sector is still waiting for its own.