The AI Policy Podcast — July 16, 2025

"China's AI Industrial Policy" with Kyle Chan

Host: Gregory C. Allen (CSIS)
Guest: Kyle Chan (Postdoctoral Researcher, Princeton; Adjunct, RAND Corporation)

Episode Overview

Theme:
Gregory Allen hosts Kyle Chan to dissect the evolution and current state of China’s AI industrial policy. The conversation traces China’s journey from the early days of economic reform through the emergence of tech giants, to the prioritization of AI as a national strategic objective. They draw on Chan's recent RAND report, "Full Stack: China’s Evolving Industrial Policy for AI," to explore the policy tools, motivations, and competitive pressures shaping China’s AI sector, how these contrast with US approaches, and what the future holds.

Guest Introduction and Personal Journey

[00:00–04:33]

• Kyle Chan’s background: Raised by Hong Kong parents in the US (Mandarin as household language), educated at UChicago in economics, but pivoted to sociology for a broader toolkit.
• Sociology as a lens for understanding China's industrial policy: allows both statistical, fieldwork, and systems perspectives.

  “Sociology gave me that systems level perspective that, you know, no shade on my economist friends, but was sort of unique among the social sciences.” — Kyle Chan [01:58]

What is Industrial Policy?

[04:33–06:55]

• Definition: State-driven strategies—investment, regulation, incentives—targeting the development of specific industries.
• China Context: Shift from command economy to market-influenced state intervention after the reform era, balancing top-down guidance with market mechanisms.

History and Structure of Chinese Industrial Policy

[06:55–11:30]

• Reform Era: State shifted from directly running industries to using incentives—joint ventures (JVs), regulatory frameworks.
• Examples:
  • Auto sector: JVs with Volkswagen, Toyota, GM required for foreign entry.
  • Mixed success; real breakthroughs came with focused efforts in electric vehicles, batteries, and protectionist policies enabling domestic champions (e.g., BYD).
• Five-Year Plans: Not top-down mandates, but directional signals for bureaucrats and entrepreneurs; align incentives across the system.

  “This sort of aligns the broader system, government, private sector, all the above, towards overarching national goals.” — Kyle Chan [11:30]

Rise of Digital Technology and Internet Giants

[11:56–18:00]

• Digital Giants: Alibaba (“Amazon of China”), Tencent (social media), Baidu (search) proliferated behind the Great Firewall, initially imitating western counterparts.
• Early Infrastructure: Built with foreign tech (e.g., Shanghai Bell JV), indigenization over time.
• Role of State: Infrastructure (telcos) was state-driven; internet firms often emerged “against the wishes or outside of the usual official framework.”
• Impact of Market Protection: Blocking western tech created sheltered domestic markets, allowing Chinese internet companies to mature and fend off global competition.

Emergence of AI in China

[18:00–26:39]

• Catalyst: 2012’s ImageNet competition marked a global AI inflection point; Chinese researchers and companies ramped up investment in AI.
• Early Focus: Machine vision and language processing prioritized—firms like iFlytek, SenseTime, Megvii, backed in part by surveillance needs under Xi Jinping (rise to power: 2012; focus on tech-driven state surveillance).

  “AI kind of solves that problem because there’s no labor shortage. AI is watching every camera all of the time. Obviously very attractive to totalitarian countries or authoritarian countries.” — Gregory Allen [21:15]

• Revenue Models: Chinese firms enjoyed robust government contracts for surveillance, unlike most American AI startups.

Making AI a National Priority — The 2017 Strategy

[24:56–31:05]

• 2017 Turning Point: Official release of the "New Generation AI Development Plan" (State Council).
  • Recognized AI as a strategic industry and cross-cutting enabler for the whole economy.
  • Goals: Become the “primary center for AI innovation,” turbocharge productivity and social services (health, education, manufacturing, governance).

    “AI was seen as a meta layer on top of China’s broader techno-industrial policy." — Kyle Chan [29:30]

  • Local governments crafted their own AI strategies, often mirroring the national plan.

The AI Industrial Policy Toolbox

[31:05–42:54]

1. Compute Infrastructure

• Goal: Build a national grid of data centers; utilize inland regions with cheap/renewable energy (“Eastern Data, Western Computing” project).
• Implementation: Much at local/provincial level, often in partnership with telecom giants (China Mobile, Huawei).

2. State-Backed AI Labs

• Purpose: Basic research, standards, training talent, open-source tools (with prominent labs in Beijing, Shenzhen, Zhejiang, Shanghai).
• Collaboration: Mix of local governments, universities (Tsinghua, Zhejiang), and sometimes private players (e.g., Alibaba).
• Outcome: Spun off high-profile startups (e.g., Zhipu from Beijing Academy of AI).

3. Talent & Education

• Massive investment in university- and academy-based AI centers, government grants, public-private research.
• Results: Surge in Chinese-authored papers at top global conferences.

4. Guidance Funds

• State-run, VC-style investment vehicles ("trillion RMB guidance funds"); co-invest with local governments and private LPs.
• Crowding-in effect: Central capital attracts private sector investment into AI and semiconductors.

Startup Benefits: Direct investment/loans, talent pools, subsidized access to compute (“compute vouchers”), regulatory support (fast-tracking permits/incorporation), cheap/free office space in government-owned zones.

Did the 2017 Strategy Work? Where Is China Now?

[42:54–47:03]

• Achievements: Robust talent pipeline, major R&D output, vibrant startup landscape, strong foundational infrastructure.
• Challenges:
  • Rapid AI paradigm shift with US-led breakthroughs (ChatGPT) caught China off-guard.
  • Shift from computer vision focus to “catch-up” mode in generative AI, especially large language models (LLMs).
  • US export controls on advanced chips (Nvidia) threaten compute capacity—a mounting gap.

  "Previously China had been focused...on sort of computer vision and things like that. Generative AI...caused a big pivot.” — Kyle Chan [43:17]

China's "Full Stack" AI Policy: Present Day

[47:03–55:30]

• Comprehensive Approach: Industrial policy tools applied everywhere—hardware (chips), compute (data centers), foundational models, applications (software & physical/robotics), open-source efforts.
• Chips:
  • Domestic alternatives (Huawei Ascend series) prioritized, but face technical and ecosystem hurdles vs. Nvidia (CUDA).
  • Reluctance to switch to domestic chips due to maturity gap and software compatibility. Market still demands even degraded Nvidia chips.
• Applications:
  • Major firms (Tencent, Alibaba/Ant, Meituan) and startups develop AIs for verticals (medical diagnostics, personal finance, edtech).
  • Highly competitive, pro-startup environment—government-provided compute, office space, business support.

  “It still requires some of these physical inputs...proximity to other researchers and tech companies...local government initiatives play a role.” — Kyle Chan [55:05]

Competitive Ferocity: Copycat Markets and Industrial Bubbles

[55:30–66:02]

• "Copycat Gladiators": Ruthless domestic competition, enabled by weak IP protection early on; “trial by fire” for innovative entrepreneurs.
• Industrial Bubbles: Industrial policy often leads to intentional over-investment (“better to have too much than too little”), producing eventual globally competitive survivors (BYD in EVs as a model; trend visible in AI too).

  “The right amount of investment is over-investment.” — Gregory Allen [64:24] “Pour a bunch of resources...that maybe results in redundancy...but ultimately at the national level...allows you to then translate into international [competitiveness].” — Kyle Chan [62:51]

Strategic Goals: Ranking Priorities

[66:02–72:20]

• Ultimate Goal: “Techno-economic power”—achieve comprehensive, up-the-value-chain success, moving beyond vulnerability to foreign “malicious interruption.”
• Self-sufficiency and indigenous alternatives prioritized (sometimes at the cost of short-term technical superiority).
• Adoption:
  • Chinese government (esp. SASAC) directs massive state-owned enterprises to deploy AI (e.g., DeepSeek mandated across SOEs).
  • Municipalities roll out AI adoption plans across public services.
  • Open source push (models/tools) to maximize diffusion, lower barriers for ecosystem uptake.

Future Trajectories: Crystal Ball

[72:20–82:16]

• If Trends Hold:

  • China could keep pace with, or even edge ahead of, the U.S. in some industrial sectors (EVs, renewables), while AI remains a mixed field owing to the chip gap.
  • U.S. competitive in AI due to massive private investment (OpenAI’s Stargate, $300B+ annual compute spending), better chips, and an “arms race” in model scaling.
  • China’s catch-up depends on whether domestic chip makers (esp. Huawei) can reach “viable” levels in design and ecosystem, or if export controls continue tightening.

  “Huawei is going to go from non-viable to viable—and that actually might end up being a more important strategic transition.” — Gregory Allen [81:23]

• Limitations and Unknowns:

  • Compute bottlenecks may define China’s AI performance ceiling.
  • Upcoming AI paradigm shifts (agents, new model types) could either raise or lower the bar for competitors.

Notable Quotes

• “Sociology gave me that systems level perspective...unique among the social sciences.” — Kyle Chan [01:58]
• “AI kind of solves that problem because there’s no labor shortage. AI is watching every camera all of the time.” — Gregory Allen [21:15]
• “In China, there's a coliseum...everybody's dying like flies out there. BYD is the tiger that survived the longest in this coliseum.” — Gregory Allen [58:43]
• “The right amount of investment is over-investment.” — Gregory Allen [64:24]
• “Better to step on the gas harder rather than...carefully modulate.” — Kyle Chan [66:02]
• “Techno-economic power… national power broadly.” — Kyle Chan [67:10]
• “We take for granted that the next Nvidia or Google will be started in the US...now that’s not as certain.” — Kyle Chan [76:42]

Key Timestamps

• [04:33] — Defining industrial policy
• [06:55] — Structure of China’s industrial policy
• [13:54] — Rise of digital technology and internet giants
• [18:00] — Modern AI’s emergence in China
• [24:56] — 2017 national AI strategy
• [31:05] — Policy tools: compute, labs, guidance funds
• [41:42] — How startups benefit from policy
• [42:54] — Did the 2017 strategy work?
• [48:09] — Full stack policy today
• [55:30] — IP protection, copycats, and hyper-competition
• [66:02] — Strategic goals and priorities
• [72:20] — Future speculation and policy recommendations

Takeaways

• China’s AI industrial policy is deeply integrated, multifaceted, and methodologically aggressive—leveraging overinvestment and robust local competition.
• Successes in domestic R&D and company formation are countered by urgent bottlenecks in high-end chip supply, exacerbated by US export controls.
• The competition is dynamic: AI may yet prove the U.S.’s last bastion of tech supremacy, but China’s capacity for policy mobilization and industrial scaling means the race is far from over.
• Both host and guest stress that innovation policy, research openness, and compute access are critical variables for AI leadership.

Full RAND report and Kyle Chan’s newsletter "High Capacity" on Substack recommended for further reading.