The Compute Challenge

In the global race for technological supremacy, access to computing infrastructure has been a critical differentiator for nations maintaining status quo in terms of growth and strategic autonomy. For India, a country with 1.4 billion citizens, a growing tech ecosystem and digital transformation goals-the “compute challenge” represents both a hurdle and an unprecedented opportunity

- Rohit Chintapali

POLICY EXPERTS at Carnegie India offered the best definition of “compute” in their February 2024 essay. They explained that in technical terms, compute measures a processor's calculation capacity, typically expressed as floating-point operations per second (FLOPS). For context, the world’s most powerful supercomputer achieves nearly 1.2 exaflops - exceeding a billion-billion calculations each second.

The experts explained that compute can also be understood as an integrated technology stack consisting of three key components: a hardware layer comprising graphic processing units (GPUs), an infrastructure layer of data centres and server optimisation algorithms, and a software layer featuring various development frameworks.

Today, despite being a software and startups powerhouse, India faces a big deficit in high-performance computing infrastructure. According to industry estimates, India possesses less than 2 per cent of the world’s AI compute capacity, severely lagging behind the United States, China and the European Union. This disparity becomes particularly concerning as AI applications and systems increasingly drive economic value across sectors and the need of compute infrastructure grows.

According to the International AI Safety Report 2025, released at the AI Action Summit in France this February, the trajectory of AI development points toward extraordinary computational demands. The report forecasts that by late 2026, certain general-purpose AI models may require approximately 100 times more training compute than the most resource-intensive models of 2023. This exponential growth is projected to accelerate further, potentially reaching 10,000 times the 2023 compute levels by 2030.