Congress wants to override the White House on AI chip exports to China, and the first vote lands Wednesday. If it passes, lawmakers—not federal agencies—could block every future license for Nvidia H200 shipments to Beijing.
The House Foreign Affairs Committee will vote Wednesday on the AI Overwatch Act, a bill that would hand Congress direct veto power over shipments of advanced AI accelerators to China and other U.S. adversaries.
Introduced by Representative Brian Mast (R-FL) in December, the measure arrives just weeks after the Trump administration quietly approved licenses for Nvidia’s flagship H200 AI chips to Chinese end-users, a move that triggered bipartisan alarm on Capitol Hill.
If enacted, the bill would force the Commerce Department to pause any license for high-performance AI silicon for a mandatory 30-day review window. During that period, the House Foreign Affairs and Senate Banking committees could reject the shipment outright—no presidential override, no agency appeal.
Why This Vote Matters Now
- Timing: The vote lands amid a White House push to loosen export curbs in exchange for Beijing’s purchase commitments on U.S. semiconductor equipment.
- Scope: The bill targets any chip above 1,200 TOPS (trillion operations per second) and 600 GB/s memory bandwidth—thresholds that capture Nvidia’s H200, AMD’s MI300X, and every next-gen accelerator slated for 2026.
- Precedent: Congress has never asserted statutory control over individual export licenses; the act would create a new legal chokepoint between Silicon Valley fabs and foreign customers.
Inside the Legislative Text
The 14-page draft defines “covered AI chips” as any processor designed primarily for machine-learning workloads that exceeds 1,200 TOPS and 600 GB/s. That language sweeps in Nvidia’s Hopper H200, the upcoming Blackwell B200, and AMD’s CDNA 3 lineup.
Licenses would be automatically suspended once either committee chair issues a disapproval resolution. The pause lasts until both chambers vote to lift it, a process that could stretch for months in a polarized Congress.
White House Pushback Intensifies
David Sacks, the administration’s AI czar, publicly criticized the bill last week, reposting an X thread that labeled the effort a “Never-Trump sabotage plot.” The thread accused Anthropic CEO Dario Amodei of orchestrating the legislation through ex-Biden staffers.
Amodei, speaking at the World Economic Forum in Davos, doubled down: “It would be a big mistake to ship these chips. It’s a bit like selling nuclear weapons to North Korea.”
Mast fired back, citing Trump’s own 2025 block on ASML’s high-NA EUV tools to China: “You can advise him to sell H200 chips to China if you want; I advise the opposite.”
Industry Impact at a Glance
- Nvidia: Roughly 18 % of its fiscal 2025 data-center revenue came from Chinese hyperscalers. A congressional veto could erase $6–8 B in annual sales.
- AMD & Intel: Both firms have pending MI300X and Gaudi-3 export waivers. The bill would freeze those applications indefinitely.
- Cloud Giants: Alibaba, ByteDance, and Tencent have pre-ordered thousands of H200s for 2026 delivery. Cancellation penalties could exceed $1 B.
- Start-ups: Chinese model labs rely on U.S. GPUs. A supply choke would accelerate Beijing’s home-grown replacement timeline by an estimated 24–36 months.
What Happens Next
Wednesday’s committee markup is procedural but pivotal. Republican leadership counts 27 of 46 members as firm “yes” votes; only three GOP defections are needed for passage. Democratic staffers say the party will largely vote “present,” ensuring the bill advances to the House floor.
From there, the path narrows. The Senate Banking Committee has not yet scheduled a companion hearing, and Majority Leader John Thune has signaled he will not fast-track any export-control legislation without White House sign-off. Still, Mast retains leverage: the annual National Defense Authorization Act offers a must-pass vehicle later this year.
Bottom Line for Builders
Developers stateside face near-term uncertainty. If the bill becomes law, cloud providers must segregate Chinese workloads onto older A100 or H100 GPUs, capping model size and training speed. Expect tighter quota systems, higher spot prices, and longer lead times for GPU instances labeled “export compliant.”
For open-source maintainers, the chilling effect is already visible: Hugging Face pull requests that mention “Chinese training data” now carry export-control disclaimers, and several model cards flag “ITAR-restricted” weights.
Meanwhile, Beijing is doubling subsidies for local fabs. SMIC’s new 7-nm line in Shenzhen is ramping to 20 k wafers per month—still two nodes behind Nvidia’s 4-nm process, but closing fast.
Stay locked to onlytrustedinfo.com for the fastest breakdown of tomorrow’s committee vote and real-time analysis of how Capitol Hill’s AI chip showdown rewires the global semiconductor map.
