Six days. That is the entire distance between a more than five-fold opening surge and a 45% drawdown.
Unitree, China's best known humanoid robot maker, debuted in Shanghai on 19 August. Shares surged past five times the offer price, then fell roughly 45%, cutting the valuation from about $66 billion to $36 billion. Reuters framed it as raising robotics bubble concerns, which is accurate and slightly beside the point.
The company did not change
Nothing about Unitree's products, order book, engineering team or manufacturing capacity changed between the surge and the fall.
Thirty billion dollars of value was created and removed without a single unit shipping differently.
That is not a comment on Unitree, which builds genuinely impressive hardware. It is a comment on what the number was measuring, which was the price of belief on a particular Tuesday.
Most coverage of this will be written for investors. Almost nobody reading this owns the stock, so let me write the version for the people who will be sold to.
Valuation finances roadmaps
Here is the mechanism that matters operationally.
Vendor roadmaps are financed by valuation. The three year product plan a robotics or AI company shows you in a procurement meeting was costed when the multiple was at its peak.
When the valuation halves, the roadmap gets rescoped. Quietly, two or three quarters later, usually announced as a sharpening of focus.
The features that get cut are rarely the demo features. They are the integration work, the regional support, the long tail of compatibility that your deployment happens to depend on.
I watched a version of this play out through the AI money getting strict earlier this summer. The pattern is consistent: capital discipline reaches customers as scope reduction, not as a price increase.
The question to ask this quarter
The diligence question is no longer what a vendor can build. It is what they can build at their current cost of capital.
Three things to ask, and to get in writing.
Cash runway at current burn, stated in months rather than described as strong. Revenue concentration, because a vendor with three customers carrying 60% of revenue has a different risk profile than the logo wall suggests. And what happens to your support tier if their next round prices flat.
That last question is the one that produces the most informative silence.
None of these are hostile questions. Any well run vendor has the answers prepared, and the ones that do not are telling you something they did not intend to.
None of this requires you to predict the market. It requires you to know which of your operations depend on a company whose plans assume a valuation it no longer has.
Why six days and not six months
Public markets reprice in hours. Private conviction takes quarters, because it only gets expressed through funding rounds, and those happen on a slow schedule.
Unitree's listing converted a private valuation into a public one. The public number is marked to market every single day by people with no relationship to the company and no obligation to be patient.
That is the mechanical reason a robotics valuation can halve in under a week while an equivalent private company holds its paper number for another year. Nothing about the underlying business is more or less real in either case.
For anyone doing vendor diligence, the useful conclusion is to treat a private valuation as an opinion with a long refresh cycle. It records what a small group of investors believed at the last round, not what the market believes this morning.
The pattern is not about robots
Look at the same week's other number. Lambda, the Nvidia backed cloud provider, is reportedly in talks to raise up to $3 billion at a valuation above $12 billion, on revenue estimated to reach $1.5 billion or more this year.
An eight times revenue multiple for an infrastructure business. By the standards of the last two years, that is disciplined. By the standards of 2022, it is aggressive.
Baselines move in both directions, and they move considerably faster on the way down. The Unitree correction took six days. The run up that preceded it took about eighteen months of accumulated conviction.
That asymmetry is the practical lesson. You have months of warning on the way up and days on the way down, which is a poor match for how procurement cycles actually run.
Robotics is not in a bubble because the technology is weak. Unitree ships real machines that do real work. The correction happened because the price ran ahead of the deployment timeline, which is a different failure and a much more common one.
The market recounted in six days. Your procurement cycle takes longer than that, and your vendor contracts run longer still.
Plan on the timeline you actually operate on, not the one the headline moves at.