#285 Of Controls and Insecurities

The Export Control Week and the Mission Mode Fallacy

Global Policy Watch: Back to Arms Race

Insights on global issues relevant to India

— RSJ

The Biden administration isn’t spending their final weeks in quiet contemplation. In the last week, it came out with two fairly sweeping policies to anchor the degrees of freedom the incoming Trump administration might have on these issues. The new administration might have a different view on them, but it will be difficult to disregard the measures suggested entirely and start afresh. The first, somewhat less surprising, move was the sweeping fresh sanctions on the Russian energy sector. The other announcement was the unprecedented “Export Control Framework for AI Diffusion’ that seeks to regulate the export of AI technology and its potential misuse. Both these moves will significantly impact India whichever form it eventually takes.

Let’s start with the Russian energy sanctions. Till the beginning of last year, Russia had found a way to circumvent the somewhat half-hearted measures that were in place to curb their energy exports. Shutting Russia out completely without finding what could substitute that supply would have meant a significant disruption to the global energy markets, which could fuel inflation. This meant Russia had built or was allowed to build, an alternative supply chain for supporting its oil fields, a shadow fleet of containers which could circumvent the $60 per barrel price cap that was imposed on Russian crude, non-dollarised trade deals and bilateral insurance agreements to continue business-as-usual. This took some time to take shape, but after the initial hiccup, it ran well, especially with its two biggest export markets - China and India. The Biden administration has been tightening the net around commercial entities that deal with Russia in strategic areas like defence, finance, and energy throughout last year. Some of these moves specifically called out Indian firms and local offices of Russian entities and cautioned Indian banks from supporting Russia-related trades.

This has been escalated a few notches up now. The new sanctions target 183 tankers in the shadow fleet outside of G7 jurisdiction and also port operators who engage with them. This will make circumvention of the price cap difficult because no Chinese or Indian port operator wants to take in a sanctioned vessel and invite compliance risk for the rest of their business. Russia will either have to go back to non-sanctioned ships that operate under G7 regulations, which will mean a significant price discount on their oil exports, or it will have to build additional capacity for a shadow fleet, which can go on for some time without attracting sanctions. Even this will take time and cost good money. The specific bans on two of the most prominent Russian oil producers - Gasprom Neft and Surgutneftegas - will also hurt as they account for 1 million bpd of Russian exports. The workaround of finding an intermediary who will buy it from them and then distribute it globally will take time and might still attract compliance risks to those buying from them.

Russia could divert the exports of these two companies to meet their domestic demand and rework the supply chain so that the other domestic gas companies can then start exporting. This will be short-lived, too, because I would like to believe the US will see through this manoeuvre. The ban on the provision of oil field service to Russia by US companies will mean, in the medium term, maintenance of older wells and development of new wells will take a hit. With the addition of two new Russian insurers, the whole of the Russian marine insurance sector is under sanctions. This will further increase the cost of exports for Russia, which will have to find new ways of insuring the supply logistics. The degrees of freedom for Russian oil exporters have been severely curtailed, and while they may find and build alternatives soon, it is unclear why those alternatives, too, won’t be placed under sanctions. This squeeze will hurt unless Putin makes a deal with Trump to get out of this. This is going to be a rough couple of quarters for Russia as the war economy starts to stutter.

Anyway, in response, unnamed Indian officials have maintained that India has a two-month wind-down period to find alternatives and that it expects Russia to offer wider discounts or come under the G7 price caps to continue its supply. That’s being a bit too sanguine. India has an oil problem. It imports almost 90 per cent of its requirements, of which about 40 per cent is sourced from Russia (up from (12 per cent in 2021). The 183 tankers that have been sanctioned were quite regular on the Indian shores. Almost 30 per cent of their volumes were for Indian refiners. India will have a problem to contend with after the current in-transit supply winds down. Unless Russia figures out an alternative fast, India will have to go back to its traditional sources in the Middle East for oil. This will mean a jump in its import prices. That apart, the reduction in Russian supply in the oil market will mean a spike in prices in the near term. This oil price jump works well for the US, too, since they are mostly self-sufficient in their energy needs, and a price hike further increases the viability of their oil fields. So, India will have to prepare for an increase in oil prices, which will feed into higher prices across industrial goods and products. Any hopes of inflation coming closer to the 4 per cent range, which could encourage the RBI to cut rates, has disappeared for the next two quarters. How much the cheap Russian oil was a factor in keeping inflation down is anyone’s guess. However, without it, the 5 per cent inflation print will also be challenging to maintain. This means even if growth comes in below expectations in the next few quarters, the RBI won’t have the elbow room to ease liquidity or cut rates to spur growth. This is before Trump comes in and puts in a fresh set of sanctions on Iranian crude, which is another source of oil for India. And I haven’t yet begun talking about the weakness of the rupee against the dollar, which could worsen after Trump is inaugurated.

Tough times ahead.

Moving on to the news about AI becoming a tool of diplomacy. There are two lenses through which the Biden administration has viewed ’AI Diffusion’ as it terms it.

First, there is a tremendous opportunity for American companies to dominate AI technology and its downstream applications, which can generate huge revenues, boost US exports, and create a new AI-led sphere of influence around the world. This is critical as China, despite its domestic challenges, continues to dominate global exports (its trade surplus reached $1 trillion last week) and has already taken a pole position on green tech. The large markets of India, Gulf states, and Southeast Asia are keen to invest in data centres and AI technology. However, some harbour deep suspicions about Chinese technology and its ultimate intentions. The way to safeguard long-term domination of any tech space is to write the rules of the game, especially around usage protocols and standards, which, once globally established, will act as an entry barrier for others. US policymakers understand this well from their experience of the multi-cycle tech domination they have had since WW2.

The other lens is this: unlike other tech cycles, AI is a different beast because it’s not merely a technological advance in the conventional sense. Powerful AI systems can be tools of strategic national and geopolitical importance. A country with a massive edge in AI technology can pose national security risks to its rivals through ways that aren’t fully comprehensible now. AI-driven cyber attacks, surveillance capabilities, and defence infrastructure are just the tip of the iceberg. Any desire for a broader diffusion of AI technology for commercial gains will have to be weighed against the possibility of handing over a critical advantage to an adversarial power that might use it against you.

The framework announced this week attempts to find a middle ground between these two countervailing forces. The rules will establish a licensing regime for the export of advanced chips and the model weights of AI systems by creating a three-tier system of restrictions to govern their sales and usage. Tier 1 has a group of eighteen closest US allies who will have unrestricted access to AI technology. Tier 3 will have key adversaries, namely, China, Russia, Iran and North Korea, who will be blocked from importing any of these technologies (which is already the policy now). In Tier 2 will be the rest of the world, including India, who will have to essentially play by US rules to access advanced computing power or set that up on their soil through US tech giants like Microsoft, Google and Amazon. So, US companies can treat the Tier 1 countries as “homeland” and deploy as much computing power there and use them as a base to expand to Tier 2 countries so long as they follow the security standards set in these rules. This allows these companies to maximise commercial opportunities in Tier 2 countries within a tight governance framework.

Additionally, there is an overall quota system on computing power which means these companies will have to ensure they keep 50 per cent of their total capacity on US soil and no more than 25 per cent of the total outside of Tier 1 with the maximum limit of 7 per cent in a single tier 2 country. The quota system keeps any Tier 2 country from running away with compute capacity because they have the money and the intent to push for multiple AI use cases. In a way, if this becomes a long-term bipartisan policy, the US will have full control over the extent of AI capabilities any country can build. This holds true until that country builds its own AI computing power or goes to China for an alternative ecosystem. Both those options are remote right now because, despite all the bluster from China, it still tries to smuggle in as many high-powered chips for its own AI use because its domestic capabilities aren’t up to the mark. With the tightening of screws on knowledge sharing and exports, it is difficult to see how China could quickly innovate on cutting-edge chips that not only satisfy its domestic demand but also leave enough surplus for exports. It is safe to assume that for the foreseeable future, the US will hold a monopoly over the export of AI technology. These rules then provide a tightly governed framework for US companies to continue with their investments in Tier 1 and 2 countries without the ambiguity of seeking approvals from the US government in every case. Tier 2 countries like India won’t face a scarcity of computing power in the medium run (the demand is still low) as long as they work with these companies to set up data centres on their soil. Additionally, homegrown companies in Tier 2 can also access computing power if they apply for the national validated end-user status (NVEU) that confirms that they will not use these chips for the wrong reasons. This would mean that a validated company would almost be treated as a company on Tier 1 soil and its chip imports won’t count towards the country’s cap.

Additionally, there are limits to the export and overseas training of AI model weights that have been laid out. Companies using these models will have to abide by US-determined standards to host the model weights of AI systems in their own countries. These restrictions don’t apply to open-weight models. The legislation almost guarantees US domination of AI technology because, eventually, half of the total computing power will have to be developed on US soil, unlike the current plans of many tech giants that envisaged opening large centres in Tier 2 countries. Where will the US get the energy to build such capacity is a separate question for the future. But expect relaxation of all kinds of green energy sourcing norms that constrain data centre buildout in the US currently.

Obviously, the US chip makers haven’t taken kindly to being shut out of China completely, their largest overseas market. Plus, there’s this lingering fear that if China manages to break through in chip technology any time soon, it possibly won’t have such restrictions placed on its companies in gaining markets in Tier 2 countries. Even outside of computing power, the lure of lower restrictions and easier compliance will have many Tier 2 countries look at China for the lower ties AI chips and access to AI models. So, there are vocal critics of this piece of legislation already. The incoming Trump administration might have a different view on this, but this legislation's core underlying principles are unlikely to be reversed. It won’t want China to circumvent the current chip embargo through friendly countries that are currently in Tier 2. And the foolproof way of doing this is to enforce global standards that the US will manage and enforce directly. I can’t see how Trump would be displeased with having that kind of bargaining power. In a way, the new administration will have to use this framework as the basis for any changes it might want to make. Much of that will be bilateral negotiations of the kind Trump likes rather than an overarching framework like this one. This framework is here to stay.

China will be left with only one option. Building chip development capabilities that quickly bridge the current gap and keep it ahead of the US.

The AI arms race has officially begun.

---

Addendum

— Pranay Kotasthane

Export controls had an IPL week. The Biden administration introduced graded restrictions on the export of AI chips and models, tightened the screws on TSMC for producing China-designed chips, added 25 Chinese companies to the Entity List, imposed a ban on Chinese Connected Car Technologies, and put in place rules to prevent Chinese and Russian companies from accessing dual-use biotech laboratory instruments.

Not to be outdone, China added a few American companies to the Unreliable Entity List and indicated an expansion of its export controls. In a role reversal, China is now considering anti-dumping duties on legacy chips imported from the US!

This week gives a sneak peek into a future where trade wars are likelier to be science and tech (S&T) wars at their core. Nuclear deterrence makes large-scale conflicts like the previous two World Wars unlikely. Similarly, the mutual gains from globalisation have made any large-scale economic decoupling unfeasible. Thus, the primary domain of international conflict, confrontation, and cooperation is now S&T. Given its perceived importance to national power, governments are willing to incur the costs of S&T decoupling to get ahead of challengers. This means countries are not content with domestic S&T investment and talent development; they actively want to degrade the adversary’s technological capabilities.

Over the last few editions, I have commented that American policies increasingly seem indistinguishable from Chinese ones. These export controls only buttress my case. It’s pretty surreal to see the US turn its back on the rhetoric of permissionless innovation and instead burden its companies with an untenable and byzantine licensing regime. What began as a “small yard, high fence” approach to technology controls has unrecognisably expanded in scale and scope. It now covers items like laboratory equipment, and the AI rules have even neatly carved out the world into Cold War-esque blocs.

Through the AI Compute Rationing rules (codename: AI diffusion), the US wants to ensure that no other country builds a frontier model in the future. But unlike the nuclear industry, which can be largely indigenised and has short supply chains, global connections are crucial for AI research, chip manufacturing, and company revenues. Thus, the rules have various exemptions allowing American businesses to sell their wares globally. Finally, this AI planning scheme is about getting ahead of China; hence, additional provisions exist to weed out Chinese companies from anything cutting-edge and American.

Trying to achieve these three goals with one policy instrument violates Tinbergen’s Rule (one policy instrument, one target)—no wonder the rules document seems as lucid as an Indian Income Tax Returns Form. On most occasions, such attempts end up achieving none of the goals. This particular licensing regime faces the additional challenge that the implementing agency—BIS—doesn’t seem to have any capacity to handle the volume of filings, licenses, waivers, and authorisations the rules will generate. At a time when the incoming administration wants to make the government more efficient, the necessity of surge hiring in the BIS is sure to produce some fireworks.

Secondly, these ever-expanding export controls can win the US followers but not partners. Even those in Tier 1 will now try to find domestic alternatives and collaborate less with each other. After all, what’s their insurance against Trump’s tantrums? What if a lousy Twitter war throws them out of the club? The previous strategy of interdependence made US companies indispensable to the world. In doing so, the US set the pace for what others could achieve. However, with this new strategy where insecurity is the norm, every country is on its own.

Expect industrial espionage, cybersecurity breaches, and shady deals as the stakes for getting hold of top-end AI models rise. Just as the Nuclear Non-proliferation Treaty (NPT) couldn’t prevent India, Pakistan, Israel, and North Korea from building their nuclear weapons programme, expect determined countries to get around these restrictions as well.

Also, expect genuine technological breakthroughs. The rules assume that GPU compute will remain the bottleneck for other countries. But that’s hardly a given. Companies working on other chip architectures, decentralised training, and small language models will get a boost. The layers of a future AI ecosystem will look very different from the ones these rules seek to protect today.

As for India, its position in Tier 2 indicates America’s assessment of India’s geopolitics and technological power. Some Indian companies smuggling chips to Russia would have affected the American calculation adversely. More importantly, the categorisation makes it clear that India lacks technical leverage—the US doesn’t see the benefits of having India in Tier 1, nor is India in a position to inflict any pain on the US.

In such a situation, India spending taxpayer money to buy Nvidia GPU clusters now makes even less strategic sense. The ownership of these chips in Indian data centres will remain perpetually conditional on following American geopolitical priorities—hardly a recipe for Sovereign AI. Instead, the strategy should be to find one leverage point in the AI supply chain that makes India indispensable to the AI economy.

---

India Policy Watch: The Mission Mode Fallacy

Insights on current policy issues in India

— Pranay Kotasthane

This is also the week of the Mahakumbh Mela in Prayagraj. You will likely hear news about how this incredible organisational feat showcases the Indian State's ability to outdo itself in mission mode. Come the general election or the Mahakumbh, the Indian State rises to such challenges and delivers spectacularly.

All of which is true. But the lesson that we take away from such feats is often wrong. We tend to think that the solution to all our governance problems is mission-mode programmes. If we can organise Mahakumbh Mela reasonably well, surely we can improve our public schools and hospitals only if we tackle them in mission mode.

This is what I call the mission mode fallacy. It comes in many versions. The general idea is that things are hard to do, so the only way to get things done is to carve out special programmes where complicated rules don't apply. In public administration, this is called the single-window clearance phenomenon. In the judiciary, this idea morphs itself into fast-track courts. In industrial policy, this thinking creates Special Economic Zones (SEZs) and Production-linked Incentives. When these “missions” sometimes succeed, we glorify the individuals involved without appreciating the ingredients that went into making the missions deliver.

An excellent Business Standard editorial on this issue explains the pitfalls of conventional thinking on mission mode programmes:

State capacity in India has always been severely limited. Historically, this constraint has been overcome not through broad expansions of ability and upskilling of state functionaries but through prioritisation and the creation of “missions”. In the 1970s and 1980s, India had Project Tiger for wildlife preservation, and Operation Flood, which transformed India’s dairy sector. Subsequently there was the oilseeds mission; the Delhi Metro, which carved out its own administrative space; and multiple others. These successfully achieve limited aims. But they do not always create a broad spillover of expertise into other related domains of policy and governance. In fact, they often suck up the most skilled, experienced, and forward-thinking individuals from elsewhere in government. Their successes are more a reflection of the fact that political prioritisation allows them to short-circuit political and regulatory obstacles and to create new, if temporary, institutions. The wrong lessons are taken from their success: People assume that this shows the capability of the Indian state rather than showing the need for removing such constraints overall….

… the need is to identify the learning that has external validity. For example, if a mission has been successful because bureaucrats involved have been chosen for their expertise, how can this be replicated across government? If mission mode programmes manage to avoid the delays imposed by certain regulations, then should not those regulations be revised in general, to improve project execution across the board? If the links between the public and private sector are managed well in such mission mode programmes, can similar mechanisms not be implemented elsewhere? Instead, the success of missions is often personalised; they are seen as reflecting the abilities of specific administrators or bureaucrats, rather than demonstrating a functional problem with how the Indian state approaches problems. [Business Standard editorial]

A classic illustration is Maruti’s success. Even today, many believe that Maruti proves that the Indian government can run manufacturing businesses. All it needs is political will and a strong leader. But the lesson from Maruti’s success is precisely the converse. Maruti succeeded despite the prevailing manufacturing policies. It succeeded because the government allowed it a choice of a foreign partner, permitted foreign investment, and granted a special manufacturing license to produce 100,000 cars—policy conditions that no other company could enjoy at that time. Thankfully, some of these lessons were applied to all manufacturers through the pro-market reforms of 1991.

It’s unreasonable to believe we can mission-mode our way through governance challenges. Mission modes work like whack-a-mole games without addressing three leverage points—fiscal powers to local governments, pro-market reforms, and law and order improvements. Improving each of these is a generational marathon.

P.S.: More details on the Maruti story are in edition #227.

---

HomeWork

Reading and listening recommendations on public policy matters

1. [Podcast] In this Puliyabaazi, we discuss the life and work of India’s finest engineer-administrator, Sir M Visvesvaraya, with historian of science Aparajith Ramnath.

भारत के सर्वश्रेष्ठ टेक्नोक्रैट कौन? How Visvesvaraya pioneered engineering led administration in India? ft. Aparajith Ramnath

2. [Article] This take on the AI Compute Rationing rules is a must-read.

   Unstable Diffusion

3. [Article] A Goodhart’s Law explanation of India’s air quality problem on a new publication platform, The Plank.

---