Tariffs Down, Capital Screened: How a $50,000 Threshold Rewrites the Layer of the US-China Truce Beijing Says Was Never Agreed

Treasury's proposed rule implements Subtitle C of the Comprehensive Outbound Investment National Security Act, enacted on December 18, 2025 as Title VIII of the Defense Production Act. Published in the Federal Register on October 26, 2026, with comments due December 7 and a final rule expected in 2027, it widens the outbound screen from three technology sectors to six and converts the AI tier from notifiable to prohibited for systems Treasury calls superintelligence.

The five statutory sectors are semiconductors and microelectronics, AI systems, quantum information technologies, high-performance computing and supercomputing, and hypersonics. Biotechnology joins them because the Secretary may add technologies that enable military, intelligence, surveillance or cyber capabilities at a country of concern. In practice the additions bite where capital is scarcest: biotechnology, high-bandwidth memory, radiation-hardened and photonic chips, and hypersonics at Mach 5 and above.

The commerce ministry responded on Saturday with \"grave concern\" and \"firm opposition\", called the proposal a discriminatory broadening of scope and an abuse of national security, and reserved the right to take \"corresponding measures\". It also noted that restrictions on investment in China \"would not only harm China-U.S. economic and trade cooperation but also hurt the interests of U.S. firms\" — the argument of a government that has decided the screen is a commercial issue rather than a security one.

The timing is the point. Two days before Beijing's objection, the same two governments had pledged to cut tariffs on $60 billion of goods each way, extended the trade truce to January 10, 2027 and published \"30-for-30\" lists of products recommended for reduced tariff treatment. Goods flows were being opened while private capital flows were being closed, and the capital layer is precisely what MOFCOM says the leaders' consensus never covered.

The mechanics are where the trade is. The $50,000 de minimis applies only to transactions that would otherwise be notifiable, so prohibited technology is barred at any size and cannot be structured around with small tickets. Limited partners get a $2,000,000 carve-out only if they give a binding contractual assurance that the fund will not fund a covered transaction. Secondary purchases of covered foreign person equity are not excluded; only the services around them — bank lending, custody, underwriting, debt rating, prime brokerage and equity research — are excepted.

And there is no registry to screen against. Treasury says it is still considering a public database of covered persons and expects a \"reasonable and diligent inquiry\" in the meantime, which makes this a compliance bill for allocators rather than a ban on buying China. US-origin capital flows into Chinese technology had already been declining through 2025 and into early 2026, so the screen accelerates an existing retreat rather than causing one.

Read the screen from the top of the stack and the transmission is short. Arm licenses the designs that Chinese AI chipmakers sell, so Arm's PRC royalty line is a direct read on whether Chinese design budgets are being funded or throttled. Its fiscal2026 filing puts China at about 18% of revenue — down from 19% a year earlier and 22% two years ago — even as PRC revenue grew 17% on royalty strength, and it explicitly warns that the outbound rule \"has not had a material effect on Arm to date but could potentially have a future impact on our PRC customers\".

That is the mechanism in one disclosure: the screen reaches its customers' funding rather than Arm's contracts, and because the screen binds US persons rather than assets, the first-order listed exposure sits in US, Hong Kong and mainland China listings, with no direct channel for Korean or Japanese names.

The clearest downstream expression of capital substitution is Chinese AI silicon. Cambricon's revenue grew 453% in 2025 to ¥6.50B with its first full-year profit of about ¥2.06B, and trailing twelve-month revenue reached ¥11.4bn — all of it achieved while US capital access was being narrowed.

The constraint has not moved to capital. Third-quarter revenue of ¥3.53B, up 104.5%, with net profit up 148%, still missed analyst expectations because of high-bandwidth memory and manufacturing supply limits. The screen loosens money; it does not loosen HBM.

Chinese platforms are the balance sheet of last resort. Tencent spent ¥79.2B of capital expenditure in 2025 and is backing a $12B-plus DeepSeek round with CATL, and Alibaba's ¥66.5B of research and development in the year to March 2026 needs no US private equity at all. That is why the domestic pools, not the private market, absorb the money the screen displaces.

The biotech mirror is WuXi AppTec: with US growth capital in Chinese biotech now prohibited outright and no notification tier to fall back on, the venture-funded developer pipeline that feeds its CRDMO order book tightens, even as FY2025 revenue rose 15.8% to ¥45.46B and backlog reached ¥58.0B. Alibaba is the same trade with a caveat — fiscal 2026 net income fell to ¥103.6B from ¥130.1B as AI spending absorbed profit.

The two Hong Kong AI listings that US corporate venture once seeded show both sides of the screen. SenseTime, which signed a strategic investment agreement with Qualcomm in November 2017, now earns its first profit — ¥620M in the first half of 2026 on revenue of ¥2.91B at a 41.4% gross margin, with second-half 2025 EBITDA of ¥380M — but its future private rounds have to come from Hong Kong placements or Chinese pools.

Horizon Robotics faces the same gate with less room: FY2025 research and development of ¥5.15B ran well ahead of revenue of ¥3.76B, the company swung to a ¥10.47B net loss, and cash of ¥20.2B is the buffer that buys time rather than a solution.