As the global race to build artificial intelligence (AI) infrastructure gathers pace, India’s biggest opportunity may not come from building advanced AI models or launching the next ChatGPT. Instead, the country’s strongest investment opportunity could lie in supplying the electricity and physical infrastructure needed to power AI data centres around the world, according to a new thematic report by Shriram Mutual Fund.
The report, titled “The AI Bubble Debate: A Unit-Economics Lens,” stated the AI investment cycle is “first and foremost a power-demand story” for India. It said that while companies continue investing billions of dollars in AI chips and data centres, electricity has become the biggest constraint in the global AI expansion.
“The binding constraint on the entire build-out is not GPU but electricity. Whatever the resolution of the unit-economics debate, whichever lab wins, however fast tokens deflate, electricity will get consumed,” the report said.
Based on this view, Shriram Mutual Fund said it continues to maintain an overweight position in India’s power sector, including companies involved in power generation, transmission, transmission EPC, power financing, diesel generator manufacturing, and the wires and cables ecosystem.
Electricity is the real bottleneck
According to the report, the world is witnessing an unprecedented AI infrastructure investment cycle. It said four of the world’s largest hyperscale technology companies committed around $1.08 trillion in capital expenditure between 2021 and 2025, while their planned spending for 2026 alone stands at nearly $725 billion. Goldman Sachs estimates cumulative AI-related capital expenditure could touch $5.3 trillion between 2025 and 2030.
While most discussions focus on powerful AI chips such as GPUs, the report stated that electricity has become the real bottleneck. It said nearly 60% of an AI data centre’s construction cost goes towards GPUs and networking equipment. However, once the facility starts operating, electricity becomes the largest recurring requirement. According to the report, a single AI-grade gigawatt facility can cost between $20 billion and $50 billion and consume as much electricity as nearly 750,000 homes.
The report stated that this creates a long-term opportunity for countries like India, which can supply the physical infrastructure needed to support AI growth. It also showed that India has no listed frontier AI laboratory or hyperscale cloud company, limiting its direct exposure to the competition among companies such as OpenAI, Anthropic, Google and Meta.
“India has no listed frontier lab and no hyperscaler,” the report said. Instead of trying to compete in the AI model race, the report says India should benefit from industries that support AI infrastructure. It describes businesses involved in switchgear, transformers, grid infrastructure, cooling systems and cables as the “picks and shovels” of the AI boom because every AI data centre depends on them regardless of which company eventually dominates the market.
Is AI investment boom another technology bubble?
The report also addressed growing concerns that the massive spending on AI infrastructure could create another dot-com-style bubble. It stated that the debate should not focus only on the size of investment. Instead, investors should ask whether AI companies can generate enough revenue from their infrastructure before it becomes outdated.
“The entire ‘bubble or not’ debate collapses into one question of unit economics. Does the revenue thrown off by the compute earn an adequate return on the capital sunk into building it, before that capital depreciates? Everything else is just headline numbers,” the report said.
Unlike the dot-com and telecom booms, the report says today’s AI spending is largely funded by some of the world’s most profitable technology companies rather than borrowed money.
“On the evidence today, it is not a solvency bubble since the spend is overwhelmingly self-funded by the most cash-generative companies on earth. It is instead a return-on-capital question, and that question is still genuinely open,” it said.
The report stated that India does not need to predict which AI company will emerge as the global winner. Instead, it can benefit from supplying the electricity and infrastructure that every AI company requires.
“If the unit economics prove out, demand compounds and so does the expected capex. If demand for tokens disappoints, the self-funded build-out still likely continues, but in a smaller way for some years anyway. It is the corner of the AI trade where India can long the physical infrastructure while being a bystander everywhere else,” the report said.