India risks falling behind in AI infrastructure race as global spending soars: MUFG

India risks falling behind in the global race to build artificial intelligence (AI) infrastructure unless it exponentially scales investments in computing power, data centres, semiconductors and energy, according to a report by Mitsubishi UFJ Financial Group (MUFG), which argues that the next AI battle will be fought over physical infrastructure rather than software.

In its 107-page report, Bottlenecks to Scaling AI Computational Power, MUFG says AI has triggered the "largest capex supercycle in history" but warns that shortages of advanced chips, electricity, memory, strategic minerals and manufacturing capacity could emerge as the biggest constraints to the technology's growth.

"Chip supply chains will shape geopolitics more than oil over the next 50 years," the report quotes former Intel CEO Pat Gelsinger as saying, underscoring the strategic importance of semiconductor supply chains in the AI era.

India trails global leaders in AI investment

The report projects that the United States will invest $2 trillion in AI infrastructure, software and services in 2026, nearly 20 times India's projected investment of $95 billion. China is expected to invest $355 billion, followed by the UK ($205 billion), Germany ($180 billion), Japan ($155 billion) and France ($110 billion).

The gap is equally stark in corporate spending. According to MUFG, AI companies account for 23% of total corporate capital expenditure in the US, compared with just 0.6% in India, highlighting the relatively limited scale of AI-led investment in the country. Japan stands at 4.1%, the European Union at 3.3%, China at 2.6% and the UK at 1.3%.

The findings come as India pushes ahead with the IndiaAI Mission and seeks to establish itself as a global AI hub, while simultaneously expanding semiconductor manufacturing and domestic data centre capacity.

The next AI race is about infrastructure

Rather than focusing on advances in AI models, MUFG argues that future leadership will depend on who can build and operate the physical infrastructure needed to train and deploy increasingly complex AI systems.

The report identifies 10 critical bottlenecks that could slow AI adoption, including semiconductor manufacturing, memory chips, electricity, data centres, strategic minerals, export controls and geopolitical risks centred around Taiwan.

One of its key findings is that the United States currently controls around 75% of global frontier AI computing power, while China accounts for about 15%, leaving the rest of the world—including India—with only a small share of advanced AI compute capacity.

"The United States currently controls about 75% of global frontier AI compute, a core competitive advantage in the global AI arms race," the report said.

Taiwan remains the world's AI chokepoint

MUFG also warns that the global AI ecosystem remains heavily dependent on Taiwan despite efforts by countries to diversify semiconductor manufacturing.

The report estimates that TSMC manufactures roughly 90% of the world's advanced AI chips, describing Taiwan as the world's most important "silicon chokepoint". It cautions that supply-chain diversification is progressing more slowly than geopolitical risks, leaving countries dependent on a highly concentrated manufacturing base.

Power, not processors, may become the biggest constraint

The report argues that electricity, rather than computing algorithms, could become the defining bottleneck for AI expansion.

It projects that capital expenditure by the world's five largest hyperscalers—including Amazon, Microsoft, Alphabet, Meta and Oracle—will reach $775 billion in 2026 before crossing $1 trillion in 2027. Memory chips alone are expected to account for roughly one-third of those investments.

MUFG also points to the unprecedented scale of emerging AI infrastructure. Meta's planned Hyperion data centre in Louisiana, for instance, is expected to span 2,250 acres, consume around 5 GW of power—equivalent to a mid-sized American city—and house millions of GPUs by the end of the decade.

For India, the report's message is clear: success in AI will depend not only on software talent or foundation models but also on sustained investments in chips, energy, computing capacity and digital infrastructure.