Shriram Asset Management Company (AMC) in its latest report noted that the global artificial intelligence (AI) investment boom should be viewed through the lens of returns on capital rather than fears of a speculative bubble, while identifying India's power and electrical infrastructure sectors as the country's biggest beneficiaries of the AI build-out.
In a report titled The AI Bubble Debate: A Unit-Economics Lens, the fund house said the central question for investors is whether revenues generated by AI infrastructure can earn an adequate return on the capital invested before those assets depreciate.
"The entire 'bubble or not' debate collapses into one question of unit economics," the report said.
Shriram AMC noted that the scale of AI investment is unprecedented. The four largest hyperscale technology companies committed about $1.08 trillion in capital expenditure between 2021 and 2025, while 2026 capital expenditure guidance alone stands at around $725 billion, up about 77% year-on-year. Citing Goldman Sachs estimates, the report said cumulative AI-related capital expenditure could reach about $5.3 trillion during 2025-30.
However, the report argues that the spending does not resemble the dot-com or telecom bubbles because it is being financed largely by some of the world's most cash-generative companies. Instead, the key risk is that returns on invested capital may remain depressed if revenues fail to justify the massive investments.
The report estimates the industry would need roughly $600-650 billion in additional annual revenue merely to generate a 10% return on current investments, while other estimates suggest AI revenues may need to approach $2 trillion by 2030 to support the capital being deployed.
Shriram AMC said AI infrastructure is currently being monetised through two distinct business models. The first involves cloud providers renting GPU capacity to customers, while the second involves operating proprietary AI models and selling access through tokens. The report said both models can generate attractive economics but depend on utilisation, pricing power and the pace of technology change.
It also contrasted the strategies of leading AI developers. While OpenAI has committed to significantly larger infrastructure investments through projects such as Stargate and partnerships across the semiconductor and cloud ecosystem, Anthropic has pursued a relatively lighter-capital strategy focused on enterprise customers and API services, targeting an earlier path to operating profitability.
According to the report, the economics of the AI investment cycle ultimately depend on three variables: GPU rental rates, pricing power for frontier AI models and the rate at which AI token prices decline. Stable GPU rentals and premium model pricing would support returns, while rapid commoditisation of AI models could compress profitability.
For India, however, the report argues that the most compelling investment opportunity lies elsewhere.
It noted that India has no listed frontier AI laboratories or hyperscale cloud companies, limiting direct participation in the AI model race. Instead, it said the country stands to benefit from the physical infrastructure required to support global AI expansion.
"The binding constraint on the entire build-out is not GPU but electricity," the report said.
Based on this view, Shriram AMC said it remains overweight on the power sector, with exposure across electricity generation, transmission, transmission EPC companies, power financiers, diesel genset manufacturers and the wires and cables ecosystem. It is also selectively positive on capital goods companies supplying equipment such as transformers, switchgear, grid infrastructure, cooling systems and cabling for data centres.
The report added that investors need not predict whether AI spending ultimately proves to be a bubble. Instead, it says they should monitor the evolving economics of AI infrastructure while positioning portfolios in businesses likely to benefit from the physical expansion of AI capacity regardless of which companies emerge as winners in the AI race.