The choice is between AI applications and AI frontiers, Pg10

Key Highlights:

• India currently lacks competitive frontier AI models, unlike leading nations like the US and China.
• The country's focus has primarily been on developing AI applications for various industries.
• A significant shortage of computing power (GPUs) is identified as the primary bottleneck for India's sovereign AI model development.
• The IndiaAI mission has a limited pool of 45,000 GPUs, with only 4,096 allocated to Sarvam AI for its flagship model.
• A "compute tax" is proposed, requiring data centers in India to reserve 25% of their capacity for a national compute pool.

Detailed Insights:

• AI's impact is bifurcated into automating routine jobs through specialized applications and its strategic role in research, cybersecurity, and defence.
• The US has imposed export restrictions on advanced AI models like Mythos and Fable, aiming for global dominance in the sector.
• Countries that fall behind in frontier AI risk economic disempowerment, similar to those who missed the Industrial Revolution.
• Scaling laws dictate that AI model performance improves with increased size and computing power, making compute crucial for advanced models.
• The GPU capacity allocated to Sarvam AI is significantly smaller (about 50 times) than what is used to train global frontier models.
• New data centers in India primarily serve multinational corporations (MNCs) and offer limited employment, with significant environmental impact on local communities.
• The proposed "compute tax" would allow a central scheduler to allocate reserved computing capacity to Indian institutions.
• MNCs might resist the "compute tax," but their bargaining position is weakened by growing global hostility towards large data center installations.

Key Concepts Involved:

• Frontier AI: Advanced artificial intelligence models that push the boundaries of current capabilities and performance.
• AI Applications: Specialized AI tools built on existing models to solve specific problems or automate tasks within industries.
• Scaling Laws: Empirical principles in AI that describe how model performance improves predictably with increases in data, model size, and computational resources.
• Compute Tax: A proposed regulatory measure requiring data centers to dedicate a portion of their computing capacity for public or national use.