On Nanjing Road in Shanghai, the Huawei flagship store faces an Apple Store, and the back of the shop, roughly where Apple would put a Genius Bar, is filled with electric vehicles [9]. A shopper walks in for a phone and is sold a car. Earlier in the same trip the delegation visited a Xiaomi factory, another phone company that turned into one of China's largest EV manufacturers within a few years [10]. Neither of those needs a rival theory of intelligence to explain it. Both need a company that treated the retail floor and the assembly line as one distribution asset.
That is the part a China slide usually loses. The American version of this argument has run on oppositions, open against closed and LLMs against world models [17]. Saft, who publishes Rest of World, says he made the trip 18 months after DeepSeek's launch specifically to see whether the two-competing-models framing held up in person [2], and he reports that the dynamism read as familiar, in contrast to recent European trips where talk of stagnation left him rattled [16].
Take the account for what it is: a single delegation and one publisher's impressions, with no counts attached. The group rode a trolley through a robotic assembly line the way you would ride an attraction at Universal Studios [11], which is a description of a demo rather than a yield number. What a trip like that can honestly support is the shape of the conversation, and Saft's reading is that the deployment focus is deliberate, that a country cut off from leading-edge chips leaned into ground it could still win, and that the constraint may have become the strategy [18].
The one hard event in the essay is the reversal. Last summer, restrictions on EDA software were relaxed within weeks as part of a trade deal tied to rare earths [6]. Set that against the window the essay itself uses. Eighteen months is about 78 weeks, so if "within weeks" means six, the control held for roughly 8% of the post-DeepSeek period the trip was meant to assess [15]. The exact figure does not matter much. A lever that trades that quickly is not a foundation you can amortize across a three-year plan.
So here is the test worth running on your own China material. For each line, write down which map produced it: frontier capability, or deployment surface. Then ask which decision of yours changes if the other map turns out to be the right one. Hiring changes, along with where you source physical-world interaction data and which vendor you standardise on. If none of the lines move under either map, the slide is describing a mood, and you can strike it without losing anything you were going to act on. The uncomfortable version of this exercise is that the lines which do move tend to be about your own supply and staffing rather than about anybody's model scores, and you can answer those from your desk.
