The Following Goose

Xi called AI China's lead goose. For a decade it has flown in the upwash of others. This is changing now.

The only help that the bird at the point of the V gets is some encouraging honks from the geese behind. Every goose behind it rides the rising air that spills off its wingtips, a cushion of lift the ornithologists call upwash, and the birds time their wingbeats to sit in it. The flock swaps the front position, matching the time each bird spends leading against the time it spends being carried, so that nobody free-rides and nobody burns out.

In 2018, Xi Jinping told a Politburo study session that artificial intelligence was China's 头雁, its "lead goose": the technology with the pulling power to drag every other sector of the economy forward behind it. It is a good metaphor.

Though for the whole of the decade since, China has not flown at the point. It has been the following goose, sitting in the upwash of the bird in front, while the American birds in front have spent hundreds of billions of dollars on the privilege of being first for a while each.

Here is what I predict.

Let's start with what China does not have.

It does not have the chips. Nvidia's most advanced processors cannot be sold there; Jensen Huang told investors in May to "expect nothing" on approvals. The Council on Foreign Relations calculated last December that Huawei would produce around 4% of the aggregate AI computing power Nvidia produces, and that America's best chips remained roughly five times more powerful than China's.

It does not have the capital. Private investors put $285 billion into American AI in 2025 alone; economists put state-led Chinese funding at $184 billion across the previous two decades. It has a shortage of specialised talent it is still years from closing.

Those are the constraints. Now watch what a creative culture does with constraints.

When I lived in China in 2012, I came across the counterfeit egg. Entrepreneurs in the south had worked out how to make one for less than a cent, against the two cents it cost to farm a real egg: six ingredients, sodium alginate and calcium carbonate among them, mixed with kitchen equipment from ingredients ordered online, with a shell that cracked. It was cheaper than nature. I have told that story for fourteen years as the purest piece of creative problem-solving I have ever encountered, and it usually gets the reaction the word "counterfeit" invites: China is a culture that copies, that takes what was invented elsewhere and makes it cheaper and faster. I have never accepted the word as an insult.

Abundance can distract. If you have the time, the money and the materials, you do not need a creative solution. You use brute force. Creativity is what happens when the thing you need is not available and the deadline does not move. It is the first ingredient of every good piece of work I have been near, and the egg is its cleanest proof.

DeepSeek is the counterfeit egg of AI. It built a model that trained on slower chips, with sharper architecture, and gave it away with open weights.

Xi has openly endorsed open-source models, a choice only a country with no moat to protect would make, and it has worked: Chinese models now consistently outnumber American ones on the OpenRouter leaderboard of the most-used systems in the world, and even Airbnb has adopted them.

Abundance does not need to be clever. That is precisely its weakness.

The second lesson is harder, because it gives the constraint a face. It is the competitor who, on paper, should beat you: bigger, richer, further ahead. Sometimes bullying you. The creative answer is never to fight on their terms.(There are two military conflicts in the world today proving exactly this, not to fight on the terms of the larger aggressor, but I won't go there…)

The taijiquan classics contain a phrase every Chinese schoolchild knows: 四两拨千斤, four ounces deflect a thousand pounds. In a fist fight, strength meets strength and the stronger wins. The taiji principle is the opposite. You do not resist the thousand pounds coming at you; you redirect it, and four ounces of well-placed pressure is enough. You do not need to be bigger. You need to know where the other person's weight is.

And the key to winning asymmetrically is patience. American firms think in terms of headlines, quarters and IPOs. China plans over 5 years, each 5 years is a LEGO brick in a larger plan.

China knows where America's weight is, because it manufactured most of it. Beijing controls roughly 94% of the world's sintered rare-earth magnet production, the magnets inside every electric motor, wind turbine and guided missile, and it runs its export licences as a dial rather than a switch.

And the upwash is the same move. America pays to fly at the front: the frontier models, the research, the compute. China rides the lift: open weights, distillation (OpenAI's own accusation against DeepSeek), architectures built to do more with less because the "more" was never on offer. The lead goose cannot profit from its own wake. Every bird behind it can.

The speech that started all of this, in June 2014, was not about language models. It was about robots. Xi had read a report predicting that AI-powered robots would remake the global economy and that China would be their biggest market. Biggest buyer was not enough, he told the assembled academicians. We would be the biggest maker too.

Twelve years later, more than half of all industrial robots installed on earth in 2024 went into a Chinese factory. China's factories run more than two million robots, more than four times Japan's stock. The plan says AI agents and tools should reach 90% of Chinese society by 2030, and it has wired the two industries together: a national exchange for training data, and the connections between model builders and machine builders mandated rather than negotiated.

Now the part that should worry anyone whose AI strategy is a chatbot. Yann LeCun, who shared the Turing Award for the deep learning that makes all of this possible, put it in one sentence three years ago: "On the road to AGI, Large Language Models are an off-ramp."

His argument is that language is "a very approximate representation" of what we perceive and think; the world is far richer than anything you can write down, and an intelligence that has only ever read about the world cannot understand it. He left Meta to build world models instead, systems that learn how physical reality behaves by watching and acting in it. Fei-Fei Li's World Labs and Skild AI have each raised a billion or more for the same bet.

This is why my prediction is that:

China stays the following goose in language models, by choice.

It costs less, it works, and a ten-year plan does not need a good quarter or an IPO. It can lose money on open-source models for as long as it takes to make the world's developers dependent on them, and keep spending on domestic chips that have not yet produced a major success, and simply keep going.

Geese do not rotate the lead because it is fair. They rotate it because the flock, not the leader, is the unit that arrives.

The frontier China leads is physical.

When the measure of intelligence becomes what a machine can do in a factory, a hospital or a home rather than what it can say, the country with two million robots, the largest supply of physical-world data on earth and a state that mandates the connections is not the following flock. It is already at the front.

In the face of this natural, mechanical, patient Chinese ascension there is one exception in the American sky, and to be honest it pains me to name it: SpaceX

In February SpaceX bought xAI in an all-stock deal that valued the pair at $1.25 trillion, putting Grok, Starlink and the world's biggest rocket programme under one roof and calling the result a vertically integrated innovation engine "on (and off) Earth". Tesla is not formally inside yet, but it is inside in every way that matters: its $2 billion stake in xAI converted into SpaceX stock, it sold the other two companies roughly $650 million of parts and services last year, Grok already directs the software version of its Optimus robot, and Tesla is closing the Model S and X lines to build humanoids on them instead. On his last earnings call Musk spoke of "more and more overlap" between the companies, which is what you say before a merger, not after one. Strip out the personality and look at the structure: a thinking company and two making companies, rockets and cars and robots, flown as one formation by a single controller who does not answer to a quarter. That is the Chinese model, built privately, inside the competitive flock. It pains me because I find the man at the front of it appalling as a public figure these days, having once deeply admired him. The structure does not care what I think. If any American formation can match a state-designed V on physical intelligence, it is not the five frontier labs taking turns. It is the one company that has stopped taking turns.

The flock is turning, away from machines that talk and towards machines that do, and the formation that has been quietly flying in that direction since 2014 truly has nobody in front of it.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 28 September 2026 in the Brandflow newsletter on LinkedIn.