I. Two places
Most technological eras have a geography, and the current one is unusually narrow. The people who train frontier models, who build the tooling around them, and who decide what "state of the art" means in a given month number in the low tens of thousands. Nearly all of them work in two regions: the Bay Area, and a cluster of Chinese cities — Hangzhou, Shenzhen, Beijing.
This is not an accident of talent. Talent is distributed roughly evenly across the world. It is a consequence of density. Frontier work advances by the informal exchange of half-finished ideas between people who run the same experiments, read the same preprints the morning they appear, and leave the same companies to found the next one. That exchange has a radius of a few kilometres.
The two regions are not mirror images. California diffuses through APIs and capital: a lab ships a model behind an endpoint, a thousand startups build on it, venture money selects the survivors. China, partly by constraint and partly by strategy, has settled on open weights: Qwen, DeepSeek, GLM and their successors are released as files. Anyone with a GPU can run them, modify them, and keep them inside their own walls. Two transmission mechanisms for the same frontier, reaching the rest of the world at different speeds and with different strings attached.
II. What density actually does
It is tempting to describe the two regions in terms of what they have — capital, talent, compute. But the thing that makes them the frontier is not what they have. It is how often their people touch the technology.
In Hangzhou or Palo Alto, contact with a new model is not an event. It is Tuesday. A release drops, and by evening a few hundred people have run it against their own problems, found where it breaks, posted the breakage, and somebody has already shipped a workaround. None of this is organised. It is the by-product of a population that interacts with the frontier at high frequency, in public, with permission to fail.
This is the actual economic engine, and it is more mundane than the mythology suggests. Frontier regions are not smarter. They have simply arranged things so that the number of contacts between people and the technology per week is enormous, and the cost of each contact is close to nothing. Everything else — the startups, the funding, the papers — is downstream of that contact rate.
III. The contact rate elsewhere
Germany's contact rate with the frontier is low, and the reasons are structural rather than cultural.
The country's route for absorbing enterprise technology was built when trying something was expensive. A vendor certifies, a systems house implements, a consultancy signs off, a compliance officer files. Each step reduces uncertainty for the buyer, and each step also reduces the number of people who ever touch the tool directly. By the time a technology has passed through the channel, it has been handled by many and used by few.
For technologies that change slowly this was a reasonable trade. For a frontier that turns over every quarter, it means the version that arrives is not the frontier, and the people who receive it never develop the reflexes that come from having watched it fail. Industrie 4.0 is the precedent: a decade of programmes and reference architectures, a very large vocabulary, and almost no increase in the number of engineers who had personally broken something new.
None of this is a verdict on German industry, which remains excellent at what it does. It is a description of a rate, and rates can be changed.
IV. Changing the rate
If the frontier is a function of contact frequency rather than of resources, then the question for a region that wants to close the distance is not "how do we acquire the technology" — it is already available as a download — but "how do we manufacture contact at scale."
What that looks like is fairly concrete, because the frontier regions have already shown the shape of it.
It looks like laboratories where the current models are running on local hardware, permanently, and anyone with a real problem can bring it in and watch what happens. Not demonstrations — working sessions, with the person who owns the problem at the keyboard.
It looks like playgrounds: environments where a company's actual data, actual documents, actual processes can be pushed through a frontier system without a procurement decision, because the environment exists precisely so that trying is free.
It looks like open office hours — regular, unscheduled, walk-in — where the people who run these systems daily are available to the people who don't yet. The value is not in the answers. It is in the accumulation of small contacts that, over months, turns an industry's relationship with the technology from a decision into a habit.
And it looks like people who are paid to break things. Every frontier region has a population whose job is to push the new release until it fails and report where — responsibly, inside guardrails, but relentlessly. The Bay calls them red-teamers and power users; Shenzhen doesn't call them anything, because it is simply how everyone works. A region without them is a region that only ever learns about failure modes from someone else's blog post.
None of these require a national programme. Each of them raises the contact rate by a few hundred interactions a week, and contact rates compound. A region that runs enough of them for long enough stops importing the frontier and starts participating in it.
V. The outcome
The distance between Hangzhou and Saxony-Anhalt is real, and it is not a technology gap. The technology has already crossed. What has not crossed is the density of contact — the ordinary, weekly, unglamorous interaction between people with problems and systems that might solve them.
That density is manufacturable. It is made of rooms, hardware, hours, and a tolerance for things breaking. Regions that build it develop their own frontier reflexes within a few years; regions that don't remain customers of someone else's.
Which of those Germany becomes is not a question of talent or capital. It has both. It is a question of how many people, this month, actually touched the thing.