#335 Rope Tricks

A Tariff and Technology Geopolitics Round-up

India Policy Watch: Tariff Pe Tariff, Judge Sahab

Insights on current policy issues in India

—RSJ

As we wrote last week, the US Supreme Court ruled against tariffs imposed by Trump under IEEPA (emergency powers) on various countries since the “liberation day” announcement last April. So, all reciprocal tariffs enumerated in that infamous chart, fentanyl-linked tariffs on Mexico and Canada, Russian oil buying penalties on India and China and random punitive tariffs on Brazil for putting Bolsonaro in jail, are invalid. Sector-specific tariffs and a few tariffs specifically targeted at China will continue.

More importantly, the Supreme Court ruling also directs the federal government to refund about US$175bn of tariffs collected so far. To whom and how? We don’t know that yet. It has passed the parcel on that problem to the lower courts to determine. Expect lawyers to make a lot of money unentangling that jungle of knots.

Trump, in response, ranted against the Supreme Court judges, then invoked Section 122 of the Trade Act (1974) to impose a flat 10 per cent tariff (or is it 15 per cent?) uniformly on all nations. Section 122 allows the President to impose a tariff of up to 15 per cent for up to 150 days to address a “fundamental international payments problem”. This is a provision that has never been used in the past, and no one knows the legal basis and evidence needed to justify its use.

In any case, does anyone know what the real applicable tariffs are at this moment or any time in the past year? I have mentioned in previous editions of having worked on lading and shipping processes of global logistics players. Even a minor change in tariffs between various countries can take months to deploy accurately on their systems and then to train processing agents to follow them based on the type of goods, port of origination, loading and destination and the arrangement between various parties on who will pay what part of the tariff, etc. I would be surprised if those systems had been upgraded since then to such an extent that they could take these frequent, random changes of tariffs on various goods and countries, and their combinations.

I suspect they are still testing their systems for some of the tariff changes made 6-9 months ago. This is evident from the effective tariff rate calculation done for the Oct-Dec 2025 quarter for US imports. It comes to about 11 per cent, which is way lower than the effective tariff rate should be if all tariffs were applied on the goods as the law stood during the quarter. This kind of significant leakage will persist because no one can really keep track and institutionalise so many changes on the system in such a short time. In a way, a uniform 10 per cent tariff across all nations must be a source of temporary relief to them compared to the complexity of Trump’s tariffs.

Where do we go from here on US tariffs?

Well, a useful place to start will be to define ‘here’ first. There are five different legal provisions available to the Trump administration to apply tariffs outside of the IEEPA provision. There is Section 301-310 of the Trade Act of 1974 (commonly called Section 301) that allows the USTR to impose tariffs or quota restrictions to counteract unfair trade practices by foreign nations. These have been used in the past two years to target Chinese EVs and batteries, with plans to extend them to other areas like medical products and semiconductors. These measures have to be country-specific, and USTR usually takes about 12-18 months to investigate and arrive at a report on unfair trade practices that could justify such retaliatory action.

Then there is Section 232, Trade Expansion Act (1962), which allows the President to impose tariffs on imports that are deemed a threat to national security based on Department of Commerce investigations. Trump has imposed these against imports of steel and aluminium from various countries, including China, India and Mexico, arguing for the need to protect the domestic industries in these areas that are critical for national defence.

There is also Section 201, Trade Act (1974), that allows the US president to impose a temporary tariff on imports that materially injure or threaten to injure a specific domestic industry. The President can impose a tariff of up to 50 per cent over the existing duties, usually for a duration of 3-4 years. A whole host of tariffs on solar panels and equipment from China was imposed by the Biden administration and later extended by Trump using this.

And finally, there is Section 338 of the Tariff Act of 1930 (or the good, old Smoot-Harley Act) that allows the President to impose ‘new or additional rates of duty’ on goods imported from a foreign country, which would impose unequal discrimination against US commerce or benefit other countries at the expense of US trade. In cases of persistent discrimination, the statute allows for the exclusion of imports from a foreign country. Tariffs enacted under this law have a cap of 50% and no fixed duration.

That’s a wide buffet of options available, and we know Trump loves to use tariffs as a panacea for the economy and a weapon for international diplomacy. So, he isn’t going to take the Supreme Court order lying down. The most likely scenario that is going to unfold now is this. The Trump administration will use the 150-day window available to it under Section 122 to re-negotiate all the bilateral deals it has signed with various nations and “intelligently” use the provisions of Section 301, 232, 201 and 338 to restore the status quo on tariffs as planned in those deals. This is a complex exercise involving another round of negotiations with perhaps less willing partners than in the past, to arrive at a solution that will be bad for those partners in any case.

I don’t think this will be done in 150 days, in which case Trump will renew Section 122 for another 150 days. Whether such a renewal needs Congressional approval, I’m not sure at this moment. In fact, Section 122 itself can be contested in the courts (as I expect they will be) since it was originally meant to address Balance of Payments issues in the pre-1974 era when there was a fixed rate system prevailing.

Can Balance of Trade of today be argued as a Balance of Payment issue? I suspect not. But then it will take much more than 150 days for any legal challenge to eventually reach the Supreme Court, where the final word on this will be heard. What all of this means is that we will have another year of chaos and negotiations, with random late-night messages from Trump imposing new tariffs on countries. The net gainer from all of these will only be China, whose significant overcapacity and dumping practices will seem saner and more predictable compared to whatever the US will offer. India will have further delay on finalising the US trade deal, and the fact that the relative disadvantage of China will go down in comparison to other nations, including India (under Section 122), it is likely that Indian exports will be under a lot more pressure.

This whole business is exhausting for everyone except Trump, and the expected drubbing that he is going to receive in midterms might make him go more ballistic on tariffs in the next 12 months. The pressure on INR because of this (and slowdown in IT exports) and the defence of the rupee by the central bank will continue to weigh heavily on liquidity in the system, which is the single biggest factor impacting credit growth. The four state elections will only aggravate the currency in circulation, draining liquidity further. I’m not sure many can explain the 7.6-7.8 per cent GDP growth forecasts based on the new GDP base, given such constraints. But then that’s the new great Indian rope trick.

---

Global Policy Watch: A Busy Week for Technology Geopolitics

Global issues relevant to India

—Pranay Kotasthane

Anthropic made it to the political page headlines this week, not once, but twice. The first instance relates to its scuffle with the Department of War on the usage of its models for military purposes. Negotiations between the Pentagon and Anthropic for a $200 million contract failed to meet the Friday deadline set by the US Secretary of War Pete Hegseth.

The US government’s demand was that it must have “full, unrestricted access to Anthropic’s models for every LAWFUL purpose in defense of the Republic.” Anthropic, on the other hand, was okay to license its technology, but insisted on two safeguards.

One: its models should not be used for mass domestic surveillance (it was okay with their use for “foreign intelligence and counterintelligence missions”). And two, since AI systems are currently not reliable for deployment in fully autonomous weapons, they should not be deployed just yet. This is a conditional reservation, not an absolute one, as Anthropic’s official statement acknowledges that it is willing to “work directly with the Department of War on R&D to improve the reliability of these systems, but they have not accepted this offer…They need to be deployed with proper guardrails, which don’t exist today.” Eventually, neither side budged, and the deal fell through. Soon enough, OpenAI got the contract and agreed to the US government’s conditions.

The dominant narrative over the last week portrayed Anthropic’s stance as a responsible one, where a technology company was pushing back against a capricious administration.

But that would be a misinterpretation. As I have highlighted above, Anthropic’s ‘red lines’ are specific conditional reservations, not a complete rejection of the use of powerful AI systems in US warfighting. Anthropic definitely doesn’t want US adversaries to have military AI, but it is pretty much okay with the US military deploying it, albeit with some guardrails. National governments have the upper hand on this question, and I expect Anthropic to fall in line soon. In any case, there are other models on offer, and we should have no doubt that the most powerful AI models will be deployed in American military systems, regardless of the Silicon Valley rhetoric.

Technology analyst Ben Thompson lists a separate reason that points to the same conclusion:

Anthropic talks a lot about alignment; this insistence on controlling the U.S. military, however, is fundamentally misaligned with reality. Current AI models are obviously not yet so powerful that they rival the U.S. military; if that is the trajectory, however — and no one has been more vocal in arguing for that trajectory than Amodei — then it seems to me the choice facing the U.S. is actually quite binary:

• Option 1 is that Anthropic accepts a subservient position relative to the U.S. government, and does not seek to retain ultimate decision-making power about how its models are used, instead leaving that to Congress and the President.

• Option 2 is that the U.S. government either destroys Anthropic or removes Amodei.

  [Anthropic and Alignment, Stratechery]

---

The other news involving Anthropic comes on the heels of the latest DeepSeek release. Last week, Anthropic published evidence that Chinese labs were ‘distilling’ Claude at a massive scale. The blog post reads:

“We have identified industrial-scale campaigns by three AI laboratories—DeepSeek, Moonshot, and MiniMax—to illicitly extract Claude’s capabilities to improve their own models. These labs generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts, in violation of our terms of service and regional access restrictions.

These labs used a technique called “distillation,” which involves training a less capable model on the outputs of a stronger one. Distillation is a widely used and legitimate training method. For example, frontier AI labs routinely distill their own models to create smaller, cheaper versions for their customers. But distillation can also be used for illicit purposes: competitors can use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently.”

The evidence is compelling, but the policy conclusion of doubling down on input controls (chips, API access) deserves serious scrutiny. The current approach of export controls restricts inputs (chips, API access) as a proxy for controlling use (military, surveillance, bioweapons). But these proxies have collateral damage. Restricting chips doesn't just slow weapons research; it also slows cancer research, climate modelling, and education.

If a French or Japanese lab had done this exact same distillation, would we be reading the same blog post with the same national security language? If not, the issue isn't distillation.

It's just that China is doing this. Anthropic frames rapid Chinese AI progress as largely dependent on distillation from American models. This conveniently implies that US dominance is natural and any catching up must involve theft. But competitive industries converge. That's just how technology works. "They used 24,000 fake accounts to get access" is presented as proof that the world needs tighter controls. I'd argue it’s proof that controls are already imposing enormous friction. The question is whether the marginal security gain of more restrictions is worth the marginal cost to global innovation.

We don't restrict electricity because someone could build an electric chair. We don't ban chemistry because someone could make poison. We create narrow, targeted rules around use. AI should be no different. General-purpose technologies need use-based governance, too. The FATF model isn't perfect, but it shows a path: multilateral agreements focused on specific dangerous applications, with monitoring frameworks adapted to the technology. It’s hard, but better than a blunt regime that treats every GPU crossing a border as a potential weapon.

The security framing does powerful rhetorical work. It makes disagreement feel unpatriotic. But conflating "a company lost competitive advantage through ToS violations" with "national security threat" is a move we should name clearly and evaluate honestly. I'm not arguing for no controls. I'm arguing that input-based controls on a general-purpose technology will always be leaky, always impose massive collateral costs, and always be one step behind. Narrow, multilateral use-based controls are better in the long run.

---

Another big development this week was the New York Times story titled The Looming Taiwan Chip Disaster That Silicon Valley Has Long Ignored. Tripp Mickle reports:

“A confidential report commissioned in 2022 by the Semiconductor Industry Association for its members, which include the largest U.S. chip companies, said cutting the supply of chips from Taiwan would lead to the largest economic crisis since the Great Depression. U.S. economic output would plunge 11 percent, twice as much as the 2008 recession. The collapse would be even more severe for China, which would experience a 16 percent decline..

If Taiwan’s factories were knocked offline, the impact would be immediate, the roughly 20-page report said. Economies would flounder. In China, the gross national product would fall by $2.8 trillion; in the United States, the drop would be $2.5 trillion.

Other reports, including one by Bloomberg Economics, a research service, estimate a conflict would cost the global economy more than $10 trillion.”

We covered this topic in edition #324. The economic impact of a blockade of Taiwan remains underestimated. I would be surprised if the US hasn’t made a contingency plan to evacuate the key leadership of TSMC in case of a Chinese blockade or invasion. Even so, replacing TSMC’s Taiwanese facilities elsewhere could take up to five years. This assessment remains valid even after accounting for the many mature node fabs being built worldwide and TSMC’s own new advanced node fabs in Japan and the US.

The only hope is deterrence. The economic loss due to Taiwanese fabs shutting down will be felt most severely by China, and perhaps this would discourage any misadventure. However, this scenario doesn’t consider the possibility that China would coerce TSMC to run under its control, while the world would fall in line just to avoid the adverse economic consequences.

This is precisely why US chip export controls might backfire. The controls force China to build domestic capabilities. Over time, this would mean its dependence on Made-in-Taiwan chips is reducing, implying that its marginal costs of occupying Taiwan become less than the marginal benefits. Ben Thompson of Stratechery again is right on the money here:

“…cutting China off from advanced chips doesn’t just reduce the likelihood that Chinese companies are dependent on a U.S.-based ecosystem, it also reduces the cost of destroying TSMC. More than that, if AI becomes as capable as Amodei says it will — the equivalent, or more, of nuclear weapons — then it actually becomes game theory optimal for China to do exactly that: if China can’t have AI, then it is, at least under current circumstances, relatively easy to make sure that nobody does.” [Anthropic and Alignment, Stratechery]

We need second-order thinking to understand the true cost of chip bans.

---

The last technology news snippet that caught my attention was the inauguration of Micron’s memory chip assembly plant in Sanand. Micron is an A-tier memory chipmaker, and the launch of its commercial operations marks the first major outcome of the India Semiconductor Mission. The first batch of assembled chips has been ordered by Dell for its laptops being made in India for the domestic market.

Before we get ahead of ourselves, it is important to contextualise what this assembly plant does and what it does not. This plant converts advanced DRAM and NAND wafers from Micron's global manufacturing network into finished memory products. Micron’s press release says that it expects to assemble and test tens of millions of chips at Sanand in 2026, scaling to hundreds of millions in 2027. Thus, the facility will be ramped up to full capacity over time.

Business Standard reports that Micron has shifted some of its machines from Penang, Malaysia, to India to kick off the plant setup, suggesting early-stage equipment deployment rather than full tooling.

Also, Micron has explicitly categorised what it’s doing in India as conventional, i.e., mature, established-process assembly, test and packaging. It is the older, less technologically demanding end of the packaging spectrum. The advanced stuff—next-generation packaging required for HBM (High Bandwidth Memory) and chiplet integration, which is critical for AI chips—is still going to happen in Micron’s US plants.

So, don’t be taken in by the hyped-up reports. Keep building. It’s a long way to go.

---

HomeWork

Reading and listening recommendations on public policy matters

1. [Podcast] Here’s a Puliyabaazi Show & Tell episode on how we are using AI tools for work. Researchers should also check out Chris Blattman’s damn good setup.

2. [Paper] On the Taiwan contingency question, this Takshashila paper presents three economic impact scenarios.

3. [Podcast] In the latest episode of the Ideas of India podcast, Pranay and Shruti discuss the political economy of critical minerals.

---