Chinese Innovation Ecosystem (Patent Evidence)
What almost 14 million Chinese patents say about the country’s technological rise. A series of recent papers — centered on Lerner, Narain, Papanikolaou, Seru & Xu’s “Chinese Sputnik Moments?” project (HBS Working Paper 27-009, July 2026) — built the first economist-grade dataset of Chinese patenting and used it to ask whether China’s critical-technology surge is real, who is producing it, and whether it depends on imported American know-how. The answers are empirically sharp and politically uncomfortable in both Washington and Beijing.
The dataset
Chinese patent data is a mess: the same patent is published multiple times, sometimes under different ID numbers, and no single source is complete. The team merged PATSTAT (EPO), CNIPA, and Google Patents, built a map linking 24 million patent IDs, and deduplicated to 16.3 million unique patents (14.2M with domestic assignees). Each patent gets a modal four-digit IPC primary class, an assignee sector (company, university, hospital, non-profit, individual), and flags for venture-backing (PitchBook: 103,445 patents across 9,755 assignees), state ownership (TianYanCha registry: 28,124 SOEs holding 497,576 patents), and PLA affiliation — including the “Seven Sons of National Defense” universities, which alone hold 176,629 patents.1
Critical technologies are identified from the U.S. DoD’s Critical Technology Areas list: take U.S. patents linked to DoD funding in the Defense Technical Information Center, extract the CPC subclasses covering 80% of those awards plus category-unique bigrams/trigrams, then match any Chinese or U.S. patent whose primary CPC/IPC subclass and title/abstract text both hit a category.2
What the data shows (Lerner et al.)
- Patenting tracks real innovation. Weighted Chinese patenting correlates with top-journal articles at the technology-year level with slope ≈ 0.98 — patents are a reasonable innovation indicator in China, not just paperwork.3
- Not a Huawei story. Critical-tech patents are not concentrated in corporate giants. Universities play a far larger role than SOEs or government facilities — the share of university/non-profit patenting in China is much greater than in the U.S.4
- Homegrown, not imported. Fewer than 1 in 10 Chinese critical-technology patents involves an inventor with U.S. experience or training. Inventors who moved from large U.S. innovative firms (DISCERN 500) appear in only ~1.4–1.6% of breakthrough patents — less than among non-breakthrough patents — and their spillover effect on receiving firms’ subsequent breakthrough output, substantial in 2008–2013, collapses after 2013.5
- No quality decline. Four text-based quality measures — disruptive bigrams (Kalyani et al.), top-decile “breakthrough” patents by Kelly-Papanikolaou-Seru-Stoffman (KPST) textual novelty scores, and a cross-national pooled KPST variant — all agree: the rise of Chinese critical-tech patenting has not come with falling quality relative to U.S. awards.6
- Secrecy doesn’t drive it. Declassified-patent analysis in both countries finds abundant similar never-classified patents by the same inventors/assignees (≈90% have non-secret siblings), so hidden military patenting doesn’t overturn the picture.7
The authors frame the tension against the received Washington wisdom (McMaster: “the largest theft of intellectual property in history”) and offer three explanations: China covered the tracks of transplanted inventors; China keeps building on Western tech through public channels (disclosures, corporate ties, cyber-espionage) rather than people; or — their tentative lean — internally generated innovation has risen since the mid-2010s.8
The quality debate
The “no quality decline” finding has a serious rival:
- Ang, Jia, Yang & Huang (Comparative Political Studies, 2024): 4.6M patents across 333 cities, 1990–2014. After the 2006 indigenous-innovation campaign attached numerical patent targets to local-government evaluation, total patents doubled their growth rate but the share of novel patents (first-ever combination of two previously uncombined technology domains — a deliberately game-proof measure) fell from 10.2% to 3.7%. Producing one batch of novel patents in 2014 took 27,000 total patents and 6.2B yuan vs 11,000 and 3B yuan in 2007. Cities under tighter intra-provincial competition produced more patents and fewer novel ones: systematic target-gaming, not innovation.9
- Boeing, Brandt, Dai, Lim & Peters (2024): LLM-scored patent importance from abstracts, 1985–2019, plus registry-based ownership types. Importance declined 2002–2010, then reversed upward; private firms (PIEs) drove the growth as overseas patentees shrank from ~60% to <10% of filings; dependence on foreign knowledge fell steadily; Chinese and foreign portfolios converged in composition.10
The studies reconcile more than they fight: different eras (Ang ends in 2014; Lerner’s internally-generated surge is post-2015; Boeing’s U-turn is post-2010), different samples (all patents vs the DoD-critical subset), and different measures (combinatorial novelty vs textual importance vs citation/KPST). The composite picture: China ran a quantity campaign that produced measurable junk in the late 2000s, and then — after target incentives shifted and capability accumulated — began producing genuinely important critical-tech work in the 2010s, disproportionately from universities rather than the state sector.
Open questions
- Does the post-2013 collapse in mover spillovers reflect genuine capability independence, or better concealment (alias filing, tracked inventors going dark)?
- How much of the critical-tech output is dual-use by design, given the Seven Sons universities’ PLA role (176K patents)?
- Will U.S. mobility restrictions and Entity List expansion accelerate internal generation (the quality-ladder model the authors sketch) or starve it?
- Patent targets were reformed after CNIPA’s 2018–2021 crackdowns on abnormal filings — is the Ang et al. gaming equilibrium actually over?
Cross-domain connections
- scientific-idea-diffusion-decline — the other big textual-corpus measurement of idea flow; China’s reduced citation-dependence on foreign knowledge (Boeing et al.) is diffusion decline at the international scale.
- institutionally-constrained-technology-adoption — state-led innovation as an institutional-incentive problem: Ang et al.’s target-gaming is exactly ICTA’s logic (rulers optimize internal metrics over external effectiveness) run through a patent office.
- gongkai — the ethnographic complement: bunnie Huang’s account of Shenzhen’s informal IP-sharing networks is the unmeasured substrate of capability accumulation that predates and prefigures the formal (post-2015) internal-generation surge this page documents.
- specification-gaming-openai-hf-incident — Goodhart’s law in two substrates: bureaucrats optimizing patent-count targets, models optimizing benchmark scores. Measurement becomes the target, the measure decays.
- demand-governance-wedge — public enthusiasm for technology vs distrust of its governance; here the state is the demand side, with its own wedge between quantity and quality.
- deepseek — the firm-level embodiment of the post-2015 internal-generation thesis: R1’s January 2025 reception was the archetypal Sputnik moment, and its 1/20th-compute efficiency claims and Ascend pivot are critical-tech capability in corporate form.
Sources
- Lerner, J., Narain, N., Papanikolaou, D., Seru, A., & Xu, Z. W. (2025). “Sputnik Moments” — ASSA/Econometric Society presentation of the “Chinese Sputnik Moments?” project (HBS WP 27-009, Jul 2026). 2025
- Lerner, J., Narain, N., Papanikolaou, D., & Seru, A. (2024). “Creating the China Patent Dataset” (NBER conference paper). 2024
- Boeing, P., Brandt, L., Dai, R., Lim, K., & Peters, B. (2024). “The Anatomy of Chinese Innovation: Insights on Patent Quality and Ownership” (IZA DP 16869 / Toronto DP 24-016). 2024
- Ang, Y. Y., Jia, N., Yang, B., & Huang, K. G. (2024). “China’s Low-Productivity Innovation Drive: Evidence From Patents” (Comparative Political Studies 57(1); abstract archive — full text paywalled). 2023 — China’s Low-Productivity Innovation Drive: Evidence From Patents