Korva is a scroll of the internet’s best ideas.
China adds under 10% of compute
Where chips come from
After big regulations against China and a dramatic rise in America, China's domestic production and Nvidia purchases stayed small, and some of its chips end up in places like Malaysia. This year China still leans on smuggled chips, on chips TSMC made for companies it thought were not Huawei but turned out to be, and on HBM that Samsung ships. By 2028 China's total AI compute is likely 30 gigawatts or less.
The domestic ramp
That starts to change in 2027 and 2028, when fabs from SMIC and CXMT come online and domestic production reaches many millions of units a year, adding five to ten gigawatts of domestic chips in 2028. Those chips are worse than what Nvidia, Google, or OpenAI will have. Adding 50 gigawatts in 2029 is reasonable, and if most are domestic chips, that 50 gigawatts is worth about 20 gigawatts of American chips.
Money and subsidies
Export controls explain part of the gap, and finance explains the rest. American investors pour money into startups readily. Chinese financial systems pick an industry and then subsidize it much harder, and China's semiconductor industry gets more subsidies than the rest of the world's semiconductor industries combined. A slower takeoff would give China time to catch up on semiconductors, and compute follows.
How did US export controls on AI chips begin?
In October 2022 the US Commerce Department restricted sales of advanced AI chips and chipmaking equipment to China. It widened those rules in October 2023 and again in later rounds. Chipmakers responded with versions of their chips designed to fall under the limits, which the rules were then adjusted to cover.
Why does HBM matter for AI chips?
High-bandwidth memory stacks DRAM dies vertically beside the processor, giving it far more data throughput than ordinary memory. AI chips spend much of their time waiting for data, so memory speed often limits performance. Only three companies, SK Hynix, Samsung, and Micron, make the stacks at scale.
How do subsidies shape the chip industry?
China created its National Integrated Circuit Industry Investment Fund, known as the Big Fund, in 2014 to finance domestic chip companies. The US passed the CHIPS and Science Act in 2022, with about 52 billion dollars for chip manufacturing and research. Chipmaking requires such large upfront spending that national subsidies have long shaped where fabs get built.
Why do domestic Chinese chips deliver less?
Without access to EUV machines, SMIC has produced 7-nanometer-class chips using older deep-ultraviolet tools and repeated patterning steps, as teardown firm TechInsights reported in 2022. That route is slower and yields fewer good chips. Chips also depend on memory bandwidth, where domestic supply has trailed.
More from Dwarkesh Podcast
