The outgoing Biden administration, on January 13, has published a new document titled “Framework for Artificial Intelligence Diffusion.” The purpose of the document is to elucidate the American government’s policy regarding emerging technologies.

With this document, it seems that the American political establishment seeks to put a process into motion which will dictate the framework for export and terms of engagement with several countries for the AI market, likely to power the most complex systems of the future.

The interim final rule published by the Bureau of Industry and Security, Department of Commerce puts restrictions on the import of Graphic Processing Units (GPU) and essentially gives AI chips geopolitical overtones. GPUs are critical for training and building large-scale AI models, essential for advanced AI applications.

Data centre GPUs are crucial for advanced use applications to include machine learning, modelling, and cloud computing. It further envisages a tiered approach with eighteen countries which include the five eyes intelligence partnership countries, key NATO allies and some countries as part of intrinsic AI value chain in the first tier.

A simple explanation is that technology hyper scalers like Google, Amazon and Microsoft would be able to deploy as much computing power as required in specified tier one countries. Most of the world would form part of tier two in which GPU manufacturing technology giants will seek requisite permissions. Whether this policy is a good business case for entities who manufacture this cutting-edge technology remains to be seen.

The implied aspects are to seek licenses and deploy not more than seven percent computing power in tier two countries and no more than a quarter in tier one countries, besides keeping fifty percent computing power captive for the United States. China along with North Korea and Russia form part of tier three countries which have been included in a blanket ban.

There is a designated capping on the prospective computing power for tier two countries. Individual companies located in the so-called tier two countries can apply for their own national validated end user status to access a larger number of AI chips with physical and cyber assurances. Essentially it means that their high-end technology proliferation and usage in cutting edge futuristic systems will be restricted in addition to adopting US standards.

This further limits the export and overseas training of proprietary AI model weights above a certain threshold, which no existing model aligns to. Post one year, these companies will have to abide by security standards to host the model weights of powerful AI systems in tier one or tier two countries.

The 168-page document is a lengthy and complex piece of policymaking which will require a number of legal, technical and business experts to decipher as to what lies beyond the obvious. There is clarity on four issues in certainty. Firstly, the policy will probably change the proverbial “grey mode” nation states to choose “monochrome” and pick sides between the US and China.

Secondly, it will protect the interests of US companies and instead of technology proliferation, it would foster technology restriction. Thirdly, this move will envisage preferential partnerships in technology and lastly, it will certainly affect countries with ambitions to build AI data centres.

It is important to note that India, with a well-established iCET and being a key ally in Indo- Pacific along with some NATO allies is placed in tier 2 of this “Silo Structuring”. Japan and Australia as part of the strategic QUAD framework are placed in tier 1. India AI Mission was launched in 2024 which aims to establish a robust AI computing infrastructure in India to support the development and testing of AI systems.

The Ministry of Electronics and Information Technology has been allocated Rs 551.75 crore in the budget of 2024-25 to enhance AI infrastructure, including the procurement of high-performance GPUs. The Union Cabinet has also approved Rs 10,372 crore with an aim to establish a computing capacity of over 10,000 GPUs and develop foundational models with a capacity of more than 100 billion parameters trained on datasets covering major Indian languages for priority sectors like healthcare, agriculture, and governance.

Effective AI deployment requires advanced cloud computing infrastructure. There could be a need to re-look at our Design Linked Incentive (DLI) policy with a larger tranche and think of more enabling policies with an aim to replicate NVIDIA and others in this decade.

There is a likely dearth of AI professionals in the world and India should aim to become the talent powerhouse of the world with its unique demographic dividend and STEM education prowess. More emphasis could be allocated to creation of Indic language datasets and models for a greater reach and harnessing a bigger talent pool in the future. Skilled manpower will be at a premium and will come with a price.

There could be a time and possibility, when some nation states could propagate an artificial deficit of trained manpower in this domain, putting a premium on their own skilled workforce. As is the trend, there may be a time in the next two decades which will give options to some nation states to use the skilled manpower in this space to further their own agendas and interests.

These restrictions are a stark reminder of the past in domains of space, atomic energy and some other critical technologies. At the core of the Indo-US technology landscape is the US-India Initiative on Critical and Emerging Technologies (iCET) which was signed in May 2022.

This was aimed to strengthen bilateral cooperation in technologies, defence innovation, future telecommunications and space sectors, besides building a resilient semiconductor supply chain. It will be prudent that this remains the centrepiece for a further reappraisal and reconfiguration of this policy by the new administration.