The second map that sanctions drew
US export controls set out to cut off China's access to advanced chips. The result ran closer to the opposite. The domestic demand that rushed to fill the gap ended up copying the entire chain. GPU, foundry, equipment, memory: for each cell of the US chain, a Chinese counterpart among the companies this wiki tracks (its nodes) is standing up. China replicates the same chain in parallel. It is a second map forged by sanctions.
A Chinese counterpart node stands in every cell
Match the pairs one by one and the maps overlap. In AI chips, Cambricon stands where NVIDIA stands: on China's approved-hardware list, taking NVIDIA-alternative demand, with the x86-compatible Hygon beside it. In foundry, SMIC stands where TSMC stands, and Hua Hong covers the mature nodes. In equipment, where ASML and Lam Research stand, the etch makers Naura and AMEC and the cleaning-tool maker ACM Research supply domestic tools. Down to the memory interface (Montage) and niche memory (GigaDevice), nearly every cell of the US chain now has its counterpart. The demand side is guaranteed by government procurement and orders from Chinese CSPs. A closed home market with no Western alternative supports the revenue of each node.
The pair's return is the progress bar of replication
Set the pairs side by side by return and the gap shows. How much more the Chinese node rises than its US counterpart in each cell is the gauge of how far the replication has come. If the cycle is an axis of time, this is an axis of space, the one foreshadowed in Perez's technology cycle.
Start with AI chips. As NVIDIA slowed to +38.9% and +5.1% across 2025-26, Cambricon jumped +420% in 2024 and carried +120.1% in 2025 and +72.8% in 2026, turning its first profit. Hygon rose +106.8% in 2025. In foundry, against TSMC (+56% and +44.3%), SMIC ran +169.9% in 2024 and +98.4% in 2025, and Hua Hong +412.1% in 2025. In equipment, ACM Research posted +161.3% and +170.9% across 2025-26, level with Lam Research (+139.5% and +135.6%). Over the same span AMEC rose +93.5% and +77.1%, and Naura +70.6% and +71.2%. The same AI demand lifted different assets on either side of a border. The localization premium is the markup on the fact that Western alternatives were blocked.
Each layer grew a second runner too
Replication did not stop at one node per cell. Each layer grew a domestic second runner. In equipment, AMEC and ACM Research stand beside Naura; in compute, Hygon beside Cambricon; in memory, GigaDevice (+147.9% and +144.9% across 2025-26) beside Montage. Dependence on a single supplier collapses under one control order, but once multiple domestic alternatives stand in each layer, the chain becomes that much harder to sever. The thicker each layer, the denser the parallel map.
The cell where the maps overlap: optics
Not every cell splits cleanly. Optics is the exception. The world's No.1 transceiver maker InnoLight (+468.1% in 2025) and Eoptolink (+368.8%) are Chinese companies, yet more than half their revenue goes to NVIDIA's GB200. In this cell the Chinese node enters the US chain as a supplier inside it rather than a parallel copy of it. The two maps overlap here. The parallel copy is not yet total, and some cells remain interlocked across the border. If controls close this cell too, optics will split in parallel as well.
Verdict
A Chinese counterpart node now stands in each cell of the US chain, and the return gap in those pairs shows how far the replication has come. What separates this from the trickle-down seen in Did the AI cycle propagate along the value chain? is that it completed inside a single border. The ceiling, though, is still upstream. Cambricon's shipments are bound by SMIC's 7nm yields (~20%), Hua Hong faced a US equipment-supply halt in 2026, and SMIC's 2026 return slowed to +6.1%. Like the EUV optics that export controls put out of reach, the topmost layer of the chain cannot yet be crossed. The localization premium priced not the completion of self-sufficiency but the will toward it.
Yet once a parallel ecosystem stands up, the story does not end with one country. The moment a second map is drawn, the rest of the chain begins to be redrawn along the lines of export controls, tariffs and sovereignty. In The chain splits along borders we read the grammar of that redrawing.