Expert Data Export Controls

Expert data export controls are the proposed or debated restrictions on selling high-value training data, expert-written content, reinforcement-learning feedback, or knowledge pipelines to foreign AI labs. In Google’s AI Brain Drain, SpaceX’s Huge Quarter, Airtable’s 90% Collapse, US Data Fuels China AI, Jason Calacanis cites Forbes reporting that U.S. data companies such as Surge AI and Mercor sell similar datasets to U.S. labs, government customers, and Chinese firms.

The episode presents a live policy split. Jason argues that expert datasets can help China catch up in AI, especially as Kimi, Qwen, and GLM become stronger. David Sacks answers that ordinary data labeling may be a commodity, that Chinese firms can recreate much of the work, and that controls should meet a high bar tied to proprietary, dual-use, or military-relevant data.

Key Claims

• Data controls are harder to define than chip controls because the controlled object may be text, feedback, workflow traces, expert answers, or labeling labor.
• The strategic case is strongest when the data is proprietary, difficult to recreate, militarily relevant, or directly improves frontier-model capability.
• Overbroad bans can invite reciprocal restrictions and may not stop competitors from recreating commodity data pipelines.
• Voluntary supplier choices, such as Micro1 declining Chinese customers in the source account, can become informal governance before formal policy exists.

Connections

• Surge AI, Mercor, Micro1, OpenAI, Anthropic, Tencent, ByteDance, Alibaba, and Moonshot - source company cluster.
• China, AI Export Controls, AI Data Infrastructure, Frontier Model Access Restrictions, and Chinese Open-Weight AI Strategy - policy and competition context.
• Open Source AI Models, Model Distillation / 模型蒸馏, and Model Sovereignty / 模型主权 - adjacent routes by which data, models, and access shape AI competition.