What the Union Budget 2026 Can—and Should Not—Do for AI

The global race for AI today is a competition driven by scale, with a handful of companies in the US and China leading the way. These firms spend tens of billions of dollars on computing infrastructure, rapidly increasing their capabilities far beyond what most governments or startups manage. 

Ahead of the Union Budget 2026, this raises an important question: Given the substantial compute costs required to train and operate advanced AI systems, how can the Indian government make the most effective use of its resources to maximise impact?

The IndiaAI Mission, approved in 2024, aims to build domestic capability across talent, datasets, applications, and computing infrastructure. To support domestic innovation and expand affordable access for Indian startups and researchers, it has deployed 38,000 GPUs so far. However, this is minuscule compared to the scale at which global tech leaders are expanding. A Graphics Processing Unit is an electronic circuit designed for image and accelerated data processing.

Meta alone operates around 1.3 million Nvidia H100-class equivalent GPUs, which are Nvidia’s flagship AI chips. At current prices of roughly $25,000 to $40,000 per unit, the total cost is estimated at $32–52 billion. This excludes other expenses, such as power, cooling, land, and networking.

Other major players are on a similar trajectory. Google, for instance, has announced that it will double its compute capacity every six months. 

By contrast, the budgetary allocation for all of IndiaAI Mission’s objectives so far stands at $1.2 billion. This raises the need to examine the role of the state and to ask whether ownership of AI infrastructure utilising public money is the appropriate battleground. India cannot win the AI race by owning compute at scale. It can only succeed by shaping demand and incentives, and by enabling last-mile access for consumers and developers alike. The Budget must reflect this reality. 

Even if India could afford to spend at this scale, the economic trade-offs are difficult to justify. India’s overall R&D expenditure remains around 0.6% of GDP, far below that of 2.56% for China  or 3.59% for the US . Within this constraint, allocating substantial sums to GPU ownership reduces funding for applied research, early-stage product development, and the adoption of AI across public services–areas where government spending has the highest multiplier. 

Additionally, compute is a capital-heavy asset that yields limited spillovers unless it is fully and continuously utilised. Given the pace at which AI hardware advances, there is a risk of rapid obsolescence, making chips purchased today less competitive in a few years and requiring substantial additional investment just to keep up. For a Budget that has to fund the country’s infrastructure, welfare, and defence needs, this is a large, upfront and impractical demand. 

Treating asset ownership as the primary policy lever reverses the logic of market development. Compute capacity does not lead innovation; it follows demand. Firms invest billions in GPUs because they see clear use cases, customers, and revenues that justify that scale. Where such demand is absent, compute sits idle. 

India has seen this before. High-performance computing facilities built under earlier public programmes have often struggled, with usage rates averaging 65-70%. GPU utilisation within the national supercomputing facilities has been even lower in recent years - around 60% in 2023, falling to about 57% in 2024. Treating GPU ownership as the primary policy lever risks repeating this pattern. 

To be sure, access to affordable compute remains a fundamental constraint for Indian innovators. High-end GPUs are prohibitively expensive for most startups, academic institutions, and small firms to purchase themselves. Depending on the use case, cloud-based AI supercomputing can also be costly. 

India’s share of global AI compute capacity is estimated to be under 2%, while the US and China together account for close to 60%. Given the strong link between compute availability and research output, some degree of government support is necessary to prevent domestic innovation from falling behind. Without sustained demand, however, publicly owned compute risks becoming an expensive but underutilised input. 

The IndiaAI Mission aims to support national priorities by strengthening domestic skills, datasets, applications, and infrastructure. When allocating resources, the Union Budget should focus on renting compute power for startups, researchers, and companies rather than trying to match Big Tech in buying GPUs. 

To maximise impact, the government could encourage early adoption of domestic AI products and services through public procurement. This approach would allow public spending to attract private investment, help build local expertise, and give firms the confidence to invest and grow in line with India’s AI goals.