Jared Cohen, the young historian turned president of global affairs at Goldman Sachs, once said, “When people think of digital diplomacy, they think of government tweeting. It is not what it is. That is public diplomacy only.”
In the immediate aftermath of President Donald Trump’s triumphant swing through the Gulf states, the details of the administration’s artificial intelligence agreements remain under wraps. Yet, the broad contours are already clear. By pledging to build hyperscale data centers in Saudi Arabia and the United Arab Emirates stocked with cutting‑edge U.S. chips, Washington has publicly reaffirmed the global reach of the American AI stack — semiconductors, algorithms, and the data that feed them — at the expense of China’s rival offering, manifest most visibly in Huawei’s stalled contracts across the region.
The pageantry of this new entente has eclipsed a more technical but equally consequential decision: Trump’s revocation of the Biden‑era “AI Diffusion Network” export‑control regime. I criticized that framework when it appeared in January for the simple reason that it converted every overseas shipment of high‑end silicon into a Kafkaesque license chase. Understanding why the rule existed, however, is indispensable for grasping the reservations now voiced by some China hawks in Congress and even a few MAGA fellow‑travelers — and for designing something more rational in its place.
The story begins with the scaling laws that dominate contemporary machine learning: more computation reliably begets more capable models.
Alarmed by Beijing’s progress, the Biden administration in 2023 embargoed the sale of Nvidia’s H‑series accelerators and their peers to China. However, we realized 18 months later that through back-door deals in Malaysia or sometimes with Middle East middlemen, the Chinese had continued to get access to critical U.S. chips and components. The DeepSeek moment briskly led to the realization that basic H100 Nvidia chips had certainly been used as part of the model training (it is estimated that this company alone could have hoarded up to 50,000 chips, which is the equivalent of a modern data center in the USA).
This export control failure forced the Biden administration, in its last days, to enact a new, cumbersome, unwieldy, and bureaucratic export framework for chips. Countries were sorted into three bins. Allies could import without limit. Embargoed states were shut out completely. Everyone else, including Israel and the Gulf monarchies, faced a quota of 350,000 H100‑equivalents through 2027, layered with 7% capacity caps and license carve‑outs that only an apparatchik could love. The H100 is the historical standard in AI chips, although these days, the GB200 Nvidia (2.5 times more powerful) is becoming the new generation standard. The result was paralysis. Thus, it is no surprise that Trump just rescinded the text and opened up the gates of this new alliance with Middle Eastern funds.
How can we strengthen these deals and reassure those who fear for national security, and the exported chips eventually finding their way to China? We need simpler, stricter, and more enforceable export controls, while the Bureau of Industry Standards at the Department of Commerce is seriously understaffed for its mission. Sen.Tom Cotton (R-AR) introduced the Chip Security Act on May 9, which, if enacted, will require “a location verification mechanism on export-controlled advanced chips.” A bipartisan House companion bill was introduced on May 15 by Reps. Bill Huizenga (R-MI), Bill Foster (D-IL), John Moolenaar (R-MI), and Raja Krishnamoorthi (D-IL).
I have always advocated location verification because this is more feasible than what some people believe and does not encroach on privacy. Nvidia already embeds an on-die secret key on the chips. We could create a communication route between the chips and some landmark data centers. Nvidia would only need to issue a firmware update enabling communications between the landmark and the tracked chip. This would give early warning to the BIS and allow on-site verification procedures, which we should include in our deals with our Saudi and Emirati friends.
The bipartisan bill would require companies to implement location verification techniques to approximate the location of exported chips and to notify the Department of Commerce if chips are found to be diverted or smuggled. Eventually, the text directs the department to study additional “chip security mechanisms” — new technologies that could help us prevent theft or misuse of advanced semiconductors.
I would go further and ask that any company receiving a large quantity of chips should sever financial or ownership connections to entities in China, Russia, or other adversaries of the USA, and be barred from using Huawei technologies. And if our Middle Eastern partners use the U.S. AI stack, they should follow the same standards regarding data breaches and leakage. We should send technical emissaries to train them. A ”trust but verify” approach should complement physical tracking with regular visits and reporting, and obviously U.S. authorization if the chips are transferred to a new entity or re-exported for some reason. Construction companies involved should not be Chinese, and no Huawei equipment should be mixed with our AI stack.
A TECHNO-INDUSTRIAL POLICY PLAYBOOK FOR THE US
I believe there is still a lot to do in terms of export controls and with our allies, particularly Europe. The old continent is not as wealthy as the Gulf States, but it has incredible human talent, and many AI researchers in the U.S. historically came from Europe.
This AI diplomacy has the potential to reshape our global alliances. If so, I would suggest not rushing to finalize the Middle East deals. The right mindset here should be to intertwine national security and AI, and anchor the monarchies into the Western camp for good.
Sebastien Laye is an economist and AI entrepreneur.