A New York Times investigation published November 15 revealed the shape of the US Commerce Department's proposed AI chip export controls: a rule covering just seven countries, with a carve-out that exempts data centers already filled with American hardware. The disclosure arrived as the conflict between Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent over Nvidia's China exposure broke into public view, and one day before Nvidia is scheduled to report its first quarterly earnings since CEO Jensen Huang's resignation on November 12.
The rule's design choices, taken together, would leave most of the world's data center market untouched by new chip licensing requirements, while locking in an exemption for the largest tranche of Nvidia GPU capacity already committed outside US borders. That the Commerce Department's proposal is narrower than what other agencies wanted is not a secret: Bessent has confronted Lutnick directly about the gap. The question is whether the Commerce Department will finalize rules before two deadlines that reshape the political calculus: a congressional accountability date of December 1 for Lutnick, and the expiration of the US-China trade truce on January 10, 2027.
What Commerce's Seven-Country Proposal Permits and Skips
The proposed Commerce Department rule would apply to data centers in seven countries: Singapore, Malaysia, Thailand, Indonesia, Brazil, South Africa, and Norway, according to the November 15 New York Times investigation summarized by ShiftDelete. Those seven were chosen because American intelligence agencies have flagged them as jurisdictions where advanced Nvidia chips are being or could be rented to Chinese customers training large AI models. The critical omission: every other country gets a free pass under this rule.
Three design choices in the Commerce draft compound its narrowness. First, existing data centers already filled with US hardware are explicitly exempt, meaning any facility that raced to procure Nvidia GPUs before the rule takes effect faces no restriction. Second, companies operating in the seven listed countries would be required to notify the government about Chinese customers training AI models on their infrastructure, rather than applying for a license, reducing the government's leverage to actual enforcement after the fact. Third, the computing power threshold that triggers the rule has been set higher than what national security officials in other agencies wanted, cutting many mid-sized deployments out of scope entirely.
The rule is being developed under the Bureau of Industry and Security, the Commerce Department's export enforcement arm, which administers the AI diffusion rule framework first proposed in January 2025 under the Biden administration. That framework divided the world into tiers, with the most advanced AI chips fully accessible only to Tier 1 allies (Australia, Japan, UK, and others), subject to caps and reporting requirements in Tier 2 countries, and effectively banned for Tier 3 (China, Russia, and embargoed nations). Lutnick's proposed rule would narrow the second-tier controls to cover only seven countries, restricting active enforcement to the seven named countries while leaving the broader cap structure in place but unenforced for everyone else.
The notification-not-license approach also matters legally. A license requirement means Commerce can deny a shipment before it happens. A notification requirement means Commerce only learns about Chinese customers training AI models in covered data centers after the fact, and must separately build a case for enforcement. National security officials in other agencies have argued this creates a one-way ratchet: once chips are deployed and training has occurred, there is no remedy once damage has occurred. Letters from the Senate Banking Committee sent to Lutnick pressed the Commerce Department to explain why Megaspeed International had not been placed on the entity list despite documented evidence of chip diversion. A second letter co-signed by Senators Warren and Hawley asked Commerce to explain its oversight framework for overseas data centers receiving US AI hardware. Neither has received a detailed public response.
How Bitdeer's .7 Billion Malaysia Hub Fits the Exemption
The practical consequences of the existing-hardware exemption are clearest in one facility: Bitdeer's A201 data center in Johor, Malaysia. The 21.7 megawatt campus has committed more than 70 percent of its capacity under contracts worth over $1.7 billion in revenue over five years, with power-on expected in early 2027. The facility has already received more than 100 Nvidia GB300 NVL72 racks, meaning it was physically equipped with restricted hardware before any rule would take effect. Under the Commerce draft, it would be grandfathered and face no new licensing requirements.
Cantor Fitzgerald, the investment bank that Lutnick founded and whose stake he transferred to his family in October 2025, has acted as sales agent for two Bitdeer stock offerings: up to $1 billion in 2026 and up to $250 million in 2024. The likely anchor tenant at the Johor facility, according to sources cited by the New York Times, is Tencent. Tencent is a Chinese technology company whose AI research division would trigger notification requirements if the rule applied, but only if the facility had not already been built and filled with hardware. Commerce has said the NYT found no evidence that those investments influenced Lutnick's decisions.
The existing-hardware exemption creates a structural incentive that the proposed rule does not resolve. Any data center operator planning to serve Chinese customers from outside China can accelerate procurement to get inside the exemption window. Singapore, Malaysia, and Thailand are home to some of the fastest-growing AI data center markets in the world. Singapore's total installed data center capacity exceeds 1.5 gigawatts, with more than a third of that added since 2024. Malaysia's Johor corridor, where Bitdeer's A201 sits, has attracted more than $20 billion in announced data center investment in the past 18 months. The exemption's geographic and temporal structure could perversely accelerate the buildout it is meant to control.
Nvidia has disclosed cloud service commitments exceeding $36 billion across hyperscalers and cloud operators worldwide. Each of those commitments involves GPU shipments that may already be in transit or installed before any rule takes effect. The question the Commerce Department has not answered publicly is what threshold of installed capacity qualifies for the exemption, and whether a data center that has received some but not all of its GPU order can claim full exemption for the remainder. That ambiguity creates its own incentive: get the full rack order placed before the rule, and the question becomes moot.
China's $3.2 Billion Nvidia Pipeline and What the Rule Misses
A Reuters investigation identified projects in China worth 23 billion yuan (about $3.2 billion) that specify Nvidia processors subject to US export restrictions, according to Decrypt's summary of the Reuters findings. The projects span multiple provinces. In Fujian, a proposed computing center valued at 6.2 billion yuan ($873 million) names H100 servers and is still seeking financing partners. In Shaanxi, a state-owned engineering group won a contract tied to a 3 billion yuan center specifying H100 and H200 chips. A project in Inner Mongolia calls for 625 H200 servers. Reuters stresses that the $3.2 billion represents total project investment including buildings and power infrastructure, not confirmed chip purchase orders.
IDC estimated Nvidia's share of China's AI accelerator shipments in the first half of 2026 at 34 percent, with Huawei holding 28 percent and gaining. Z.AI, a Chinese developer, recently released an image-generation model it says was trained entirely on Huawei chips, a proof-point that domestic alternatives can produce competitive outputs. ByteDance, Baidu, and Alibaba have all announced increased procurement of Huawei's Ascend 910C processors, which deliver roughly 60 percent of the floating-point performance of an Nvidia H100 at a price below what sanctioned H100s trade for on grey markets.
The bear case for tight export controls, however, is straightforward: restricting Nvidia's access to China may accelerate Huawei's development timeline without slowing Chinese AI progress in any measurable way, while costing American semiconductor companies tens of billions in revenue that will flow to Huawei instead. Critics of the hawkish position argue this is a losing trade. The price of a black-market H100 in China reached an estimated $40,000 to $80,000 per chip in mid-2026, according to grey-market price trackers, compared to a list price of roughly $30,000 for the legitimate version. That premium funds the diversion infrastructure that Commerce is trying to close: stricter rules raise the premium further, making diversion more profitable for intermediaries, unless paired with aggressive entity list enforcement. That is precisely what Bessent has demanded and Lutnick has not delivered.
The Situation Room Fight and Why Bessent Wants a Different Rule
The administrative clash behind the proposed rule became visible on November 5, when Treasury Secretary Scott Bessent reportedly confronted Lutnick in the White House Situation Room. Coinotag's account of the meeting describes Bessent asking why Megaspeed International, an Nvidia reseller accused of diverting chips to Shanghai data centers, had not been placed on the entity list despite multiple agency reviews flagging it as a risk. Lutnick reportedly said he did not know. Axios reported that Lutnick warned stricter chip controls would start a trade war with China during earlier discussions, and that Trump had backed chip crackdown efforts three separate times before Lutnick stalled each time.
Bloomberg separately reported that banned Nvidia chips ended up in a Shanghai data center linked to Alice Huang, citing Megaspeed International and a Singapore-based Nvidia Cloud Partner called Aperia Cloud Services in the chain. Some servers were originally sold to Aperia and later moved to the Chinese facility. Nvidia has not been shown to have acted improperly, but the reporting has intensified congressional scrutiny of the company's oversight procedures and the adequacy of Commerce's enforcement. Bessent gave Nvidia's board a deadline, understood to fall around November 25, to answer export compliance questions, according to reporting cited by ShiftDelete.
The Megaspeed investigation is the central case study in the Commerce Department's credibility gap. Lutnick's Commerce Department has not placed Megaspeed on the entity list despite that reporting, and multiple senior officials in other agencies have pressed for it. The Commerce Department's response has been that it cannot comment on enforcement proceedings, a standard position that provides no timeline for action. Senators from both parties have signed letters to Lutnick, and the House Armed Services Committee separately requested a classified briefing on AI chip diversion for November 18. That briefing would put the Commerce Department's export control posture on record before its own proposed rule is finalized. If the December 1 congressional deadline passes without a substantive response, lawmakers have signaled they will pursue legislative fixes that bypass Commerce's discretion entirely, potentially mandating entity list additions by statute.
Nvidia's November 17 Earnings and the Three Deadlines That Follow
Nvidia reports earnings on November 17, the first results since Jensen Huang's resignation on November 12 and the Bloomberg reporting on banned chips in Shanghai. CFO Colette Kress, now interim CEO, inherits a data center segment that generated roughly $26 billion in a recent quarter, with China explicitly excluded from forward guidance. Nvidia has already excluded China data center chip revenue from its earnings guidance, citing regulatory uncertainty. The practical question is whether the board has identified the full scope of the compliance problem before Bessent's deadline, or whether the November 25 response will reveal additional gaps in Nvidia's oversight of its cloud partner ecosystem. TechCrunch's November 15 analysis raises whether the leadership vacuum at Nvidia creates a policy vacuum at the same moment the administration is trying to finalize export rules.
Three imminent deadlines will shape the outcome. Congressional lawmakers have set December 1 as the date by which Lutnick must answer for Megaspeed, according to CryptoBriefing. Nvidia's export compliance response to Bessent is due around November 25. The US-China trade truce that has moderated tariff escalation lapses on January 10, 2027, removing a constraint on both sides taking harder positions on semiconductor trade. If the Commerce Department has not finalized its rule by then, whatever regulation exists will be tested against a changed diplomatic context in which both the United States and China have less incentive to accommodate each other's commercial interests.
The broader compute infrastructure market has priced Nvidia's dominance as durable. Its roughly 80 percent share of AI GPU compute globally means that any rule affecting Nvidia's ability to ship to third-country data centers has an outsized impact on the entire buildout. If the final rule is as narrow as Commerce has proposed, data center developers in unaffected jurisdictions would face no new barriers, and the Nvidia supply story would continue largely uninterrupted. If Bessent wins the internal fight and the rule goes global, with real licensing requirements rather than notifications, the data center pipeline outside the seven listed countries would face the same friction that China-based operators have faced since mid-2025, repricing a large portion of the announced investment in AI compute infrastructure globally. The November 17 earnings call will be the first public forum in which the new Nvidia leadership has to answer, on the record, which outcome it expects.
A rule that exempts data centers already filled with US chips is not a chip restriction. It is a permit for the hardware that is already there.
Key Takeaways
- Seven-country scope: Commerce's proposed rule covers Singapore, Malaysia, Thailand, Indonesia, Brazil, South Africa, and Norway; all other countries face no new chip controls under the draft.
- Existing hardware exempt: Any data center already filled with Nvidia GPUs before the rule takes effect faces no license requirement or compliance obligation, creating a race-to-procure incentive.
- Cantor Fitzgerald conflict: Commerce Secretary Lutnick's former firm holds up to $1.25 billion in Bitdeer stock sale mandates; Bitdeer's Malaysian facility with 100-plus GB300 NVL72 racks is grandfathered under the proposed rule.
- China pipeline at risk: Reuters identified $3.2 billion (23 billion yuan) in Chinese AI projects specifying restricted Nvidia processors; Nvidia holds 34 percent of China's AI accelerator market versus Huawei's 28 percent.
- Three deadlines converge: Nvidia compliance answers due November 25, congressional deadline for Lutnick December 1, US-China trade truce lapsing January 10, 2027.
Questions Worth Asking
- If Commerce finalizes a rule that exempts existing hardware, does that simply accelerate the race to procure Nvidia GPUs before any future rule closes the window, making each regulatory cycle less effective than the last?
- Nvidia holds 34 percent of China's AI accelerator market. Would stricter export controls slow Chinese AI development, or just hand the remaining share to Huawei while costing American chipmakers tens of billions?
- The Lutnick-Bessent fight is about jurisdiction, but the deeper question is enforcement: even a broad rule means little if Commerce refuses to add violators to the entity list. Who actually enforces the rules once they exist?