Congress moves to tighten control over exports of cutting‑edge AI chips to China, advancing two bipartisan measures that would pull key decisions away from the executive branch and place them under direct congressional oversight.
On April 23, the House Foreign Affairs Committee approved two bills-the AI Overwatch Act and the Chip Security Act-underscoring growing frustration on Capitol Hill with how the Trump administration has handled sales of advanced Nvidia semiconductors to China and other geopolitical rivals. All but two committee members backed the AI Overwatch Act, and the Chip Security Act also cleared the panel with broad bipartisan support.
AI Overwatch Act: Treating AI Chips Like Arms Sales
The AI Overwatch Act, authored by Foreign Affairs Committee Chair Brian Mast, is designed to give Congress a formal veto over the export of high‑end AI accelerators to so‑called “countries of concern.” The bill would create a 30‑day review period for export licenses covering advanced AI chips destined for China, Russia, Iran, North Korea, Cuba, and Venezuela, mirroring mechanisms lawmakers already use to review and potentially block arms sales.
Crucially, the measure would not just apply to future approvals. It would revoke all existing export licenses for AI chips to those countries until the administration delivers a detailed strategy explaining how such exports will be prevented from enhancing foreign military or intelligence capabilities. Only once that plan is submitted and deemed satisfactory could licenses be re‑evaluated.
Mast framed the issue in stark national security terms, arguing that AI hardware is now as strategically sensitive as traditional weapons. “We are in an AI arms race, and it’s important that we know where the AI arms dealers are selling,” he said, suggesting that chips powering generative models and battlefield analytics should be treated with the same seriousness as missile systems or advanced radar.
A companion version of the AI Overwatch Act already has bipartisan backing in the Senate, with sponsors spanning the ideological spectrum, including Jim Banks and Elizabeth Warren. The cross‑party alignment reflects a widening consensus that access to top‑tier AI hardware is now central to geopolitical competition, especially with China.
Nvidia’s Role and Concerns About “Ghost Datacenters”
The debate has been complicated by comments from Nvidia CEO Jensen Huang, who recently warned that China already operates “ghost datacenters” equipped with infrastructure capable of supporting frontier‑level AI systems. His remarks suggested that Beijing may be closer than many in Washington realize to matching US capabilities, even as Nvidia has simultaneously lobbied against tighter export controls.
This dual message-highlighting China’s rapid progress while arguing against additional restrictions on US chip exports-has unnerved lawmakers. For them, Huang’s warning is evidence that the window to slow or limit China’s access to state‑of‑the‑art AI compute is narrowing, and that current export regimes may already be undermined by past sales and insufficient tracking.
At the same time, Nvidia and other industry players warn that overly broad restrictions could push Chinese companies to accelerate domestic chip development or rely more heavily on non‑US suppliers, potentially eroding US dominance in the AI hardware market over the longer term.
Chip Security Act: Closing the Diversion Loophole
While the AI Overwatch Act focuses on licensing and political oversight, the Chip Security Act targets a more technical problem: once an advanced chip leaves the United States, how can Washington be sure it stays where it is supposed to?
The bill would mandate that high‑end AI chips exported from the US include built‑in mechanisms capable of verifying their physical location. These could involve secure on‑chip modules or cryptographic attestation systems that periodically confirm where the hardware is operating.
In addition, exporters would be legally required to notify the US government if a chip is detected at an unauthorized site-such as a Chinese military research lab or an intelligence‑linked facility-rather than the originally approved end user. Together, these provisions are intended to plug what lawmakers describe as a “verification gap” in current export rules.
Right now, the US can approve a chip sale to a seemingly benign buyer in a friendly or neutral country, but it has no reliable, technical way of ensuring that hardware is not quietly rerouted, resold, or transferred to an adversary. The Chip Security Act aims to transform export controls from a paper‑based licensing system into one backed by real‑time, hardware‑level visibility.
Market Repercussions and the AI-Finance Nexus
Semiconductor export rules are not just a matter of foreign policy; they also move markets. Nvidia’s disclosure of an expected 5.5 billion dollars in charges in April 2025, when Washington began requiring export licenses for its H20 chips destined for China, illustrated how quickly regulatory shifts can translate into major financial hits.
Each new round of restrictions has tended to trigger sharp swings in Nvidia’s share price, as investors reassess revenue prospects in the vast Chinese market. Broader tech and AI‑linked assets, including companies building on AI infrastructure and adjacent digital‑asset markets, have also shown sensitivity to changes in US-China chip policy.
For investors, the two new bills represent not only a potential change in who holds the pen on export approvals, but also a signal that Washington may be moving toward a more durable, legislated framework for AI hardware controls-one that could be harder for any single administration to loosen.
White House and Nvidia Push Back
Despite strong bipartisan support in committee, the bills face determined opposition from the White House and from parts of the tech industry. White House AI adviser David Sacks publicly criticized the AI Overwatch Act, amplifying arguments that the proposal would constrain the Trump administration’s ability to maneuver strategically in its broader contest with China.
From this vantage point, too much congressional micromanagement over export licenses could slow decision‑making, introduce political gridlock into a fast‑moving technological arena, and reduce the president’s flexibility to use export control policy as a bargaining chip in negotiations.
Nvidia’s leadership has mounted its own lobbying campaign. Jensen Huang has argued to lawmakers that US strategic interests are actually served by proliferating American chips worldwide, including in China. The more global AI infrastructure is built atop US‑designed hardware and software stacks, the argument goes, the more entrenched US firms become in the ecosystem, giving Washington long‑term leverage and influence.
Chair Mast, however, has publicly pushed back, saying that the talking points he heard from Sacks closely mirrored those promoted by Nvidia, implying that industry interests and executive‑branch arguments are aligning in ways that may not fully reflect security concerns.
A Struggle Over Who Controls Export Policy
If enacted, the AI Overwatch Act and Chip Security Act would mark a significant shift in how the United States governs exports of sensitive technology. Historically, the executive branch has enjoyed broad discretion to set and administer export rules, with Congress providing a framework but rarely intervening in individual licensing decisions.
These bills would change that balance. By giving Congress a formal review window and by imposing detailed technical requirements on hardware, lawmakers would be asserting a much more direct role in shaping the flow of critical AI components abroad. In practice, that could mean slower approvals, more public debate around controversial exports, and a higher likelihood that politically sensitive sales are blocked altogether.
For the executive branch, this represents not just a policy disagreement over China, but a structural challenge to presidential authority over national security and foreign economic policy. For Congress, it is a chance to lock in a tougher, more durable stance on AI export controls that cannot be easily reversed by a future administration.
Why AI Chips Are Being Treated Like Strategic Weapons
Behind the legislative push lies a deeper shift in how policymakers think about technology. Advanced AI accelerators are no longer viewed simply as commercial products; they are increasingly seen as core enablers of military power, intelligence gathering, cyber operations, and influence campaigns.
High‑end GPUs and specialized AI chips enable rapid model training for applications ranging from automated target recognition and drone swarms to electronic warfare, satellite analysis, and codebreaking. In this context, restricting access to frontier‑grade hardware becomes a form of strategic denial, akin to limiting exports of advanced fighter engines or missile components.
Lawmakers worry that if China or other rivals gain comparable compute capacity, they will be able to build AI systems that rival or surpass US capabilities in critical defense and security domains. That concern is driving a shift from reactive export controls to a more proactive strategy: limit access now, before the capability gap closes.
Technical and Practical Challenges Ahead
Even supporters of the bills acknowledge that implementing hardware‑level location verification is technically complex. Any on‑chip tracking or attestation mechanism must be secure enough to resist tampering by sophisticated state‑level adversaries, while not imposing such high overhead that customers reject the product.
There are also questions about privacy and commercial confidentiality. Verification systems that constantly report location and usage could raise concerns among foreign buyers about giving the US government visibility into their infrastructure. Some countries may respond by encouraging domestic alternatives or by favoring suppliers from jurisdictions seen as less intrusive.
Enforcement will be another hurdle. Detecting that a chip has been diverted requires reliable telemetry, but adversaries may attempt to isolate or spoof devices, disable verification features, or reverse‑engineer them. As controls tighten, a black market in unlabeled or modified hardware could emerge, mirroring what has occurred historically with weapons and sanctioned technologies.
Global Reactions and the Risk of Fragmentation
Other major tech powers are watching Washington’s moves closely. If the US successfully mandates location‑aware chips and more stringent licensing, allied countries may face pressure to adopt compatible regimes to prevent “backdoor” access via their jurisdictions.
This could push the global semiconductor ecosystem toward more fragmented rules and technology stacks. Countries aligned with the US may adopt similar controls, while others may seek to build independent supply chains insulated from US regulatory reach. Over time, this could accelerate trends toward technological decoupling between rival blocs.
Such fragmentation would have broad consequences: duplicated R&D efforts, higher costs, and slower diffusion of cutting‑edge tools across borders. For AI researchers and startups outside the US and China, navigating this increasingly politicized landscape could become a core operational challenge.
What Businesses and Investors Should Watch
For chipmakers, cloud providers, and AI‑centric firms, the trajectory of these bills matters not just in principle but in day‑to‑day planning. Companies with significant exposure to Chinese demand will need to model scenarios where exports of top‑tier chips become far more constrained, and where future products must incorporate secure location verification from the design stage.
Investors should track not only whether the legislation passes, but also how aggressively regulators define “advanced AI chips” in implementing rules. A narrow definition might only affect the latest, highest‑end accelerators; a broader one could sweep in a large share of data‑center and high‑performance computing products.
Firms operating in sensitive sectors-defense, telecom, large‑scale AI services-will likely face stricter scrutiny on cross‑border partnerships, joint ventures, and cloud‑based access to compute. Even where hardware does not physically leave US soil, regulators may increasingly ask how remote access to powerful US‑based AI clusters is being controlled.
The Road Ahead in Congress
For now, both the AI Overwatch Act and the Chip Security Act have cleared only an initial hurdle. To become law, they must pass the full House, win approval in the Senate, and secure the president’s signature-all amid vocal pushback from the White House and well‑resourced industry lobbying.
Still, the overwhelming bipartisan vote in committee shows that concern over AI exports to China bridges many of the usual partisan divides. As strategic competition in AI intensifies, Congress appears determined to assert itself, even at the cost of a bruising institutional fight with the executive branch and friction with some of the country’s most powerful technology companies.
Whether or not these specific bills make it across the finish line, the direction of travel is clear: AI hardware has moved to the center of national security debate, and US export controls on advanced chips are likely to become tighter, more technical, and more politically contested in the years ahead.