What is the real state of the US-China chip war — sanctions, SMIC's progress, and the reshoring push
The US-China Chip War, Explained: Who Controls the Most Important Technology in the World?
Based on analysis of a 101-node, 272-edge knowledge graph mapping the relationships between US export controls, Chinese semiconductor development, and global technology competition.
What This Is About
Semiconductors — the tiny chips inside every phone, computer, car, and military weapon — are the most strategically important manufactured objects on Earth right now. The United States has spent the last several years trying to slow China’s ability to make advanced chips. China has been spending enormous sums trying to make its own. This is a map of how those two efforts interact, collide, and sometimes accidentally help each other.
The graph is not a prediction. It is a structural picture: here are the players, here are the connections between them, and here is what happens when you pull on different threads.
The Two Giants at the Center
Imagine a tug-of-war. On one end: the US government’s system of export controls — rules that say which chips, machines, and software American companies (and companies using American technology anywhere in the world) are allowed to sell to China. On the other end: China’s massive, state-funded effort to build its own chip industry so it doesn’t need to buy from anyone.
These two are the most connected nodes in the entire graph. Everything else — every policy, every company, every technology — connects primarily to one of these two. And the single strongest relationship in the whole graph is between them: US export controls directly amplify China’s drive for self-sufficiency. The harder the US squeezes, the more urgently China tries to break free.
That’s not a political opinion. It’s a structural feature of the graph.
The Paradox Built Into the System
Here’s the most counterintuitive thing the graph shows: US export controls partially undermine themselves.
Think of it like this. Imagine you have a neighbor who keeps buying groceries from your store, and you decide to stop selling to them. In the short run, they have fewer groceries. In the long run, they learn to grow their own food, build their own supply chain, and eventually don’t need your store at all.
The graph models this as a dedicated loop. Imposing controls generates what it calls a “compliance treadmill” — companies and governments constantly find ways around the rules, which forces new rules, which generates new workarounds, which forces yet more rules. That treadmill simultaneously weakens the controls and strengthens China’s motivation to invest in alternatives.
This isn’t speculation in the graph — there are separate nodes for the smuggling pipelines, the shell companies used to get around rules, the chip diversion networks running through third countries, and the Chinese AI company that trained a world-class model partly using chips that slipped through the controls.
The Machine That’s Slowly Breaking Down
One of the most important physical chokepoints in the graph is a type of chip-making machine called a DUV lithography system, made almost exclusively by a Dutch company called ASML. Without regular maintenance and spare parts from ASML, these machines degrade and eventually stop working.
The graph separates this into two different nodes: the policy decision to deny service (something governments can turn on and off), and the physical clock ticking on the machines already inside China. These are treated as distinct because they have different causal effects. The policy can be reversed by a phone call. The physical degradation cannot.
The clock node connects to China’s advanced chip production progress in a very direct way: the longer the machines go unserviced, the harder it becomes to manufacture cutting-edge chips at high yield. The graph predicts that if China’s domestic equipment programs — specifically a company called SMEE trying to build its own version of these machines — succeed before a certain deadline around 2030, it would be a phase transition, not a gradual shift. Many things downstream of that single chokepoint would change at once.
Where Everything Is Heading
The graph has a node called “Manufacturing Geopolitical Bifurcation Lock-In.” That’s a complicated phrase for a simple idea: the world’s chip industry is splitting into two separate ecosystems, one centered on the US and its allies, one centered on China, and the split is becoming harder to reverse.
What’s notable is the structure around this node. Almost every major player in the graph — American policy, Chinese state investment, Korean chipmakers, Taiwanese foundries, the Huawei-SMIC partnership — has edges pointing toward this node. Very few edges point away from it. In graph terms, it behaves like a drain: things flow into it from many directions, and not much flows out.
The one significant force modeled as slowing the split is the “mutual hostage” dynamic — the fact that China controls rare earth minerals that American chip factories need, and the US controls software tools and equipment that Chinese chip designers need. If both sides keep pulling, they both get hurt. That tension is modeled as a partial brake on the bifurcation, but it has not prevented it.
What Happened to US Policy Coherence
The graph shows three separate Trump-era policy nodes all carrying high-weight edges that undermine the US export control system. None of them amplify it.
The basic dynamic is a conflict between national security goals (keep advanced chips out of China’s hands) and economic goals (keep American chip companies selling to Chinese customers, because those are large and profitable markets). The graph models this as a genuine structural conflict, not a rhetorical one.
At the same time, a proposed escalation called the MATCH Act — which would give the US government a “kill switch” to remotely disable certain equipment inside China — carries a high-weight edge that would strengthen the controls. But it is modeled as proposed, not enacted. Meanwhile, the revenue-focused approach is modeled as already operational.
The graph also shows something unusual: the MATCH Act, if passed, would constrain the very alliance of Japan, the Netherlands, and the US that enforces the controls. The proposed escalation would require allies to comply with rules they have not agreed to, potentially straining the enforcement coalition.
Connections You Would Not Expect
A software architecture decision reducing the impact of a hardware ban. A Chinese AI company called DeepSeek built a model that requires far less computing power than expected to achieve high performance. The graph shows this directly reducing the strategic impact of US controls on a specific type of memory chip called HBM. The logic: if you need less computing power per unit of AI capability, the shortage of computing hardware matters less. A software breakthrough partially nullifies a hardware chokepoint.
An open-source processor design threatening a legal control mechanism. The US government’s ability to apply export controls globally rests on a legal concept: if a product contains US-origin technology or intellectual property, the US can regulate where it goes, regardless of where it’s manufactured. A processor design standard called RISC-V is open-source and not owned by any US company. The graph models Chinese adoption of RISC-V, combined with development of domestic chip design software, as the most significant long-term threat to this legal mechanism — not a physical workaround, but a legal one.
A Samsung competitive strategy depending on US controls remaining in place. Samsung’s foundry business (its chip manufacturing for other companies) is modeled as partly dependent on HBM export controls staying in effect. The connection is indirect but real: controls on memory chips disadvantage Chinese AI chip customers who would otherwise use Samsung’s manufacturing services, while also redirecting Samsung’s own investment priorities.
Government corruption reducing the chance of a decisive AI advantage. A node representing documented fraud and mismanagement inside China’s state semiconductor investment funds has a direct edge to a node representing the highest-stakes outcome: whether AI development leads to a decisive, compounding economic advantage. Corruption in state investment reduces the efficiency of R&D spending, which reduces the probability of achieving that advantage.
The Feedback Loops
A feedback loop is when A causes B, and B causes A. The graph has several, and they matter because they are self-sustaining — once started, they tend to continue without external intervention.
The most powerful one is symmetric: US controls trigger China to restrict rare earth minerals, and Chinese rare earth restrictions trigger further US controls. Both edges carry nearly equal weights. The graph treats this as a retaliatory cycle that can escalate independently of any specific policy decision.
The second important loop is the compliance treadmill already described: controls generate workarounds, workarounds weaken controls, which generates pressure for new controls.
There is also a loop involving Huawei specifically: the US extraterritorial controls triggered Huawei to use shell companies to access chip manufacturing through TSMC; that circumvention triggered tighter US controls; those tighter controls enable the extraterritorial mechanism that triggered the circumvention. Each round of the loop, the controls get stricter, and the circumvention gets more elaborate.
Bottom Line
The graph shows five structural findings that are not obvious from reading the news:
The controls accelerate what they are meant to prevent. The strongest single edge in the graph runs from US export controls to China’s self-sufficiency drive. The mechanism that tightens controls also strengthens the motivation and urgency of the thing being controlled against.
Three independent mechanisms are slowly dismantling the legal foundation of US global chip control. The US government’s ability to regulate chips manufactured anywhere in the world rests on the fact that most chip design uses American software and American processor architecture. China’s domestic design software programs, its adoption of open-source processor standards, and its investment in domestic chip design tools are attacking these foundations at different layers simultaneously.
The lithography chokepoint has a hard deadline. Whether China can maintain advanced chip production at volume is currently constrained by the condition of equipment it already has inside the country. That equipment is degrading. Whether China can replace it with domestic alternatives before the machines fail is one of the most consequential races modeled in the graph.
US policy is currently pulling in two directions at once. The revenue-focused policy approach and the national security escalation approach are both present at high weight, pulling in opposite directions, with no resolved outcome. The graph does not predict which wins — it shows they are in conflict.
The world’s chip industry is converging toward a split that becomes harder to reverse the longer it continues. Most of the major actors in the graph are contributing to that split, including the control mechanisms designed to prevent it. The graph models the split as a near-attractor state — a condition that, once approached from multiple directions simultaneously, becomes self-sustaining.