#158 From Prohibition to Tollgate
In this edition of Technopolitik, Arindam Goswami explores what the Nvidia-AMD chip saga in China tells us about technology denial regimes.
This newsletter is curated by Anwesha Sen.
The Nvidia-AMD China chip story shows that export controls are rarely simple or lasting. These controls depend on deals shaped by politics, national security, and business interests. In April 2025, Nvidia was required to obtain export licenses for its H20 chips, which caused a $4.5 billion loss in one quarter. The ban was lifted by July. After that, Nvidia and AMD agreed to pay 15% of their China revenues for access, in a deal reportedly made directly between Jensen Huang and the President, who first asked for 20%. By December 2025, the deal included the H200 chip, and the US share rose to 25%. As of early 2026, Nvidia has not sold any H200 chips in China because Beijing’s approval process is delayed. Both companies and policymakers are facing a changing situation, not a settled policy.
The Conceptual Shift: From Prohibition to Tollgate
In my earlier paper on Technology Denial Regimes, I looked at export controls based on two factors: supply chain control and working with other countries. The four parts of the Imposing Country Strategic Effectiveness Framework - Iron Gate, Fragile Bloc, Paper Wall, and Hollow Threat - all assume the main goal is to block access. The Nvidia-AMD deal challenges this idea.
The US is no longer blocking China from getting advanced AI chips. Instead, it charges for access and takes a share of the profits. This change shifts the approach from blocking access to setting up a tollgate.
Farrell and Newman’s weaponised interdependence theory describes two ways countries use their network power: the panopticon effect, which is about surveillance, and the chokepoint effect, which is about blocking access. The revenue-sharing deal adds a third way, called the tollgate effect, where the controlling country charges for access instead of blocking it. The idea of club goods changes too. Before, there were two groups: insiders with full access and outsiders with none. The Nvidia deal creates a third group: conditional insiders who pay a regular fee.
Who Benefits - And How
By taking 25% of H200 revenues, the US government earns several billion dollars each year without taking on business risks. It also gets information, since every approved Chinese buyer is identified and every shipment is tracked.
It can be said that charging for access lasts longer than outright bans for dual-use technologies. Bans push the targeted country to act quickly, but paid access lowers that urgency. If China can buy H200 chips at a 25% markup, it feels less pressure to speed up its own development. The Sputnik paradox, where a country denied access innovates faster, does not apply if access can be bought. The CFR notes that Huawei does not plan to make an H200-level chip before late 2027.
These benefits depend on certain conditions. First, it should be planned through a clear, multi-agency process, not created under business pressure. Second, it must follow the law - the Export Control Reform Act bans fees on export license applications, and the current tariff workaround is still being challenged. Third, it should involve allies, but the Netherlands, Japan, and South Korea were not consulted, leaving the coalition without a shared reason for their restrictions. Fourth, it should focus on technology that is already widespread enough that a ban would not work - a decision that should come before the arrangement, not after. The current deal meets few of these standards. It was reactive, legally improvised, announced by the US alone, and applied to a chip - the H200 - that the US had recently called a national security risk.
Options for the Targeted Country
In the Targeted Country Adaptive Response Framework, China is seen as a Resilient Transformer. It has strong local research and development and can form other partnerships. The option to pay for access changes how China responds to the situation.
China has logical reasons to accept the current situation. The H200 is almost six times more powerful than the H20, which was the best chip available before this regime. Paying 25% to the US government is cheaper than waiting two to three years for a domestic alternative. The sensible short-term move is to buy access while continuing local development.
China can also refuse. As of early 2026, Beijing has not approved H200 imports under US terms. This could be a negotiation tactic, using delays to push for easier approval. It might also mean China is concerned that agreeing would make the tollgate model permanent.
A third option is for China to act as a Determined Innovator: treat access as temporary and speed up its own development. This can happen alongside buying access.
Evolution or Collapse?
A key question is whether the Nvidia deal is an evolution of technology denial policy, shifting to a more advanced tollgate model, or if it shows the denial regime collapsing under political pressure and only looking like strategic innovation.
The idea of evolution uses the OPEC example. After the 1973 embargo, OPEC found that setting prices lasted longer than outright bans. Bans led other countries to find substitutes, while pricing brought steady profits. From this perspective, the US is learning a similar lesson.
The collapse view is this. The full export ban was lifted in seven weeks, after Nvidia lost $7 billion in one quarter. The percentage was decided not by strategy, but by talks between a CEO and a president. Since the Export Control Reform Act bans fees for export license applications, the deal had to be set up as an import tariff through US territory. This workaround shows the political motives behind the arrangement.
The collapse argument also highlights a major strategic problem. By setting 25% as the price to turn a ban into permission, the US has revealed its minimum price. Now China knows the chokepoint can be bought, not just used as leverage.
The evolution view also affects the coalition. The Biden-era export control group, which includes the Netherlands, Japan, and South Korea, was based on a shared goal: stopping Chinese AI compute for security reasons. That reason falls apart if the US starts selling access for money without consulting its allies. For example, why should the Netherlands now keep limiting ASML’s EUV equipment sales at a big cost?
Implications for Semiconductors and AI
This episode has four lasting effects on how technology denial works in semiconductors and AI.
First, the Iron Gate quadrant is now basically empty for AI chips. It requires both supply chain control and multilateral cooperation. The revenue-sharing model has weakened supply chain control by providing access and damaged cooperation by showing that US policy can change under commercial pressure. The semiconductor denial regime has shifted from Fragile Bloc toward Paper Wall.
Second, the pattern of raising fees is risky. The deal increased from 15% on H20 chips to 25% on H200 chips in just five months. The CFR estimates that if this approach is used for Blackwell-generation chips, China’s AI computing power could grow by over 1000% compared to using only local chips. Setting a percentage fee for each new chip generation matters more than any single deal.
Third, the policy now includes public choice issues. The US government gains financially from keeping the deal in place. This makes it less likely that future rules will become stricter, even if strategy suggests they should.
Fourth, for countries like India, which have historically been targets of technology denial in nuclear, space, and supercomputing, the monetised access model opens a new space for negotiation. A country offering supply chain partnership commitments could negotiate better terms, such as lower or zero revenue share in exchange for semiconductor manufacturing investment. This is a very different way of dealing with technology denial compared to the adaptive strategies I described in my earlier paper.
The Nvidia-AMD case does not settle whether this is evolution or collapse. A future administration with stronger legal backing, better coordination with allies, and the right institutions could create a true tollgate model.
Technology denial regimes only work if they are credible. If a company can negotiate the rules down in seven weeks because of a bad earnings report, the regime loses credibility.
The Post Graduate Programme in Public Policy is a comprehensive 48-week hybrid programme tailored for those aiming to delve deep into the theoretical and practical aspects of public policy. This programme emphasises both academic knowledge and hands-on experience, blending conceptual learning with policy simulation exercises. Beyond the online webinars, students also get the added advantage of engaging with Takshashila’s esteemed subject matter experts and its vast alumni community.
Overall, this intensive programme will arm you with all the essential skills for policymaking, analysis, persuasion, and communication, preparing you for your chosen career path in public policy.
And before you go-
Check out Grammar of War, a newsletter by Adya Madhavan, that looks at advanced military technologies!