For three years, the US-China AI competition has been fought on two fronts: chips and models. Washington restricted semiconductor exports.
Beijing built domestic alternatives. The US cut off access to frontier models. China open-sourced its way around the blockade.
A third front is now open: data.
In July 2026, the Financial Times reported that Beijing was considering a major expansion of its technology export controls. The target was not chips or semiconductors. It was the data used to train AI models — and the model weights themselves. According to the report, the Ministry of Commerce had held talks with Alibaba, ByteDance, and Zhipu AI to discuss restricting the transfer of core training data overseas and limiting foreign users' ability to download model weights.
The US had already moved first. In February 2024, President Biden signed Executive Order 14117, "Preventing Access to Americans' Bulk Sensitive Personal Data and United States Government-Related Data by Countries of Concern". The final rule, issued by the Department of Justice in December 2024 and taking effect in April 2025, specifically targets China, including Hong Kong and Macau. Its stated purpose: preventing countries of concern from using bulk sensitive data to "develop and enhance artificial intelligence capabilities".
Two countries. Two walls. One resource. Data has become the third front in the AI war.
China's New Data Wall
Beijing's approach to data control is not new. The Personal Information Protection Law (PIPL) took effect in November 2021. The Data Security Act came earlier. But the July 2026 discussions marked a significant escalation: from regulating how data leaves China to actively preventing AI training data from leaving at all.
The FT report revealed the scope of what Beijing was considering. The Ministry of Commerce had engaged with China's leading AI companies — Alibaba, ByteDance, and Zhipu — on a package of potential restrictions. These included limiting the transfer of key training data overseas, restricting foreign users from downloading model weights, and potentially prohibiting foreign chipmakers like Qualcomm and TSMC from manufacturing semiconductors based on designs developed by Chinese companies.
The rationale is straightforward: if data is the raw material of AI, then controlling the outflow of high-quality training data is as strategic as controlling the outflow of chips. Chinese regulators have watched US companies scrape the open internet for training data — and have concluded that China's proprietary, operationally grounded data should not follow the same path.
The US-China Economic and Security Review Commission, a congressional advisory body, issued a report in August 2026 warning that China was systematically collecting "enterprise, operational, and real-world data that cannot be accessed through web scraping" — and that this could give China a structural advantage in the AI race. The irony was not lost on observers: the US was warning about a data advantage that Beijing was actively trying to protect.
America's Data Wall
The US data wall is built on a different legal foundation but serves the same purpose: keeping data out of the other side's hands.
Executive Order 14117 represents the first comprehensive US data export control regime in history. It marks a complete reversal of the US's long-standing policy of "free flow of data across borders." The DOJ's final rule prohibits or restricts transactions involving bulk sensitive personal data and government-related data with six "countries of concern," including China.
The thresholds defining "bulk" data are specific: for personal identifiers, the threshold is 100,000 US persons; for financial data, 10,000; for precise geolocation data, 1,000. The rule applies broadly to US persons — including US companies, US-based subsidiaries of foreign companies, and even foreign individuals physically present in the US.
Data brokers, genomic data transactions, and certain vendor agreements are either prohibited or subject to strict licensing requirements.
The DOJ's rationale was explicit: countries of concern were using bulk data to "develop and enhance artificial intelligence capabilities and algorithms" in ways that threatened US national security. The rule was designed to cut off that supply.
For Chinese AI companies with US operations or US customers, the impact is significant. Any data transfer from a US entity to a Chinese parent company could trigger the rule's restrictions. Some companies have responded by isolating their US operations — keeping data within US subsidiaries and walling it off from Chinese headquarters. But even that workaround carries risks: if Chinese engineers based in the US access covered data, they are still US persons under the rule and subject to its restrictions.
The Model Weight Puzzle
The most complex piece of the data puzzle is model weights — the numerical parameters that determine how an AI model actually works.
The US moved first. In January 2025, the Commerce Department's Bureau of Industry and Security introduced Export Control Classification Number 4E091, which classified advanced closed-weight AI model weights as controlled technology. The threshold was models trained on 10²⁶ or more computational operations. The rule was later rescinded as part of the AI Diffusion Rule, but the precedent was set: model weights could be treated as export-controlled technology.
China is now considering its own version. The July 2026 discussions included limiting foreign users' ability to download Chinese model weights. The concern, according to reports, is that foreign users — particularly US developers — have been downloading Chinese open-weight models, customizing them, and building commercial applications without any technology transfer controls.
The result is a strange symmetry: the US wants to stop Chinese companies from accessing US model weights. China wants to stop US companies from accessing Chinese model weights. Both sides are building walls around their most valuable AI assets. And both sides are discovering that model weights are harder to control than chips. A chip can be blocked at the border. A model weight is just a file — and files can be copied, shared, and downloaded through a thousand channels.
The Business Cost
For global enterprises, the data wars are creating a new category of compliance cost.
Companies operating in both the US and China now face dual-directional data compliance — satisfying China's data export restrictions while simultaneously satisfying US data import restrictions. The US DOJ rule imposes obligations on US persons that can be contractually passed down to foreign partners through representations, warranties, and indemnity clauses. Chinese regulations require data export security assessments for certain categories of data.
The compliance burden is not trivial. One AI company told the FT that the cost of maintaining separate data infrastructure for US and Chinese operations had added "hundreds of millions of dollars" to its annual budget. Another estimated that the combined compliance cost of US and Chinese data regulations now exceeds the cost of the compute hardware itself.
The fragmentation is also affecting AI development. Companies that once trained models on globally sourced data now face the prospect of building separate training pipelines for different jurisdictions. A model trained on US data may not be deployable in China; a model trained on Chinese data may not be deployable in the US. The era of globally trained AI is ending.
The Third Front
The chip war is about hardware. The model war is about software. The data war is about the raw material that feeds both.
China is building a wall around its training data — the proprietary, real-world data that cannot be scraped from the open internet. The US is building a wall around its citizens' personal data — the sensitive information that powers consumer AI and personalization. Both sides are treating data as a strategic asset. Both sides are willing to sacrifice short-term commercial opportunity for long-term strategic control.
The data war is not a sideshow. It is the defining front of the next phase of the AI competition. How the walls are built — and who builds them better — will shape global AI for years to come.
Sources: Financial Times (July 2026); Sing Tao Headlines (July 21, 2026); Reuters (July 21, 2026); Ming Pao (July 22, 2026); White & Case LLP analysis of DOJ final rule (December 2024); Federal Register (January 15, 2025); USCC report (August 18, 2026); Lexology PIPL analysis (2021).
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.
Forecasts from third-party analysts can change with market conditions.
Cost and pricing examples are point-in-time estimates; actual rates vary.
Country and company comparisons rely on public reporting, not operational data.
This sector moves fast; timelines and deal terms may be updated later.
Company deals and regulatory rulings may evolve; verify current status.
AI infrastructure is changing quickly; claims can become outdated soon.
Sources
- Financial Times (July 2026)
- Sing Tao Headlines (July 21, 2026)
- Reuters (July 21, 2026)
- Ming Pao (July 22, 2026)
- White & Case LLP analysis of DOJ final rule (December 2024)
- Federal Register (January 15, 2025)
- USCC report (August 18, 2026)
- Lexology PIPL analysis (2021).
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.; Forecasts from third-party analysts can change with market conditions.; Cost and pricing examples are point-in-time estimates; actual rates vary.; Country and company comparisons rely on public reporting, not operational data.; This sector moves fast; timelines and deal terms may be updated later.; Company deals and regulatory rulings may evolve; verify current status.; AI infrastructure is changing quickly; claims can become outdated soon.