If AI Bubble Bursts Indians Need To Worry

Chennai: If the AI bubble in the US market bursts, it may not spare equity markets across the world, including India. Foreign portfolio investor (FPI) outflows could put pressure on the rupee, fuel imported inflation, and trigger volatility in Indian markets. A freeze in debt-funded investments by hyperscalers could also slow hiring by Global Capability Centres (GCCs), affecting employment, real estate and consumption in cities such as Bengaluru, Hyderabad and the National Capital Region.

Concerns over an AI bubble have intensified following the correction in the South Korean stock market, the decline in SpaceX's valuation, and warnings from policymakers. The RBI Governor has cautioned that a sharp correction in AI-driven global technology stocks could spill over to India. With AI leaders such as OpenAI and Anthropic also preparing for IPOs, the debate over whether AI valuations are sustainable has become increasingly relevant.

To understand the risks and implications, we spoke to Vineeth Vijayaraghavan, who has worked closely with private equity investors and AI startups.

Everyone is talking about an AI bubble. With both the RBI Governor and the Chief Economic Advisor expressing concerns, the issue has come much closer home. Why are investments in AI companies increasingly being described as a bubble? Is there a disconnect between earnings and valuations?

There is certainly a disconnect. Take OpenAI, which is currently valued at roughly 40 times sales despite generating about $20 billion in annual revenue and still remaining unprofitable. In fact, estimates suggest it could lose between $15 billion and $20 billion this year alone. Anthropic, too, is valued at around 20 times sales while continuing to report losses.

Now compare this with established companies. TCS, for instance, trades at around 25 times earnings while generating consistent profits. AI laboratories are being valued at 40 times sales despite burning cash. That itself tells you why investors are worried.

At the same time, hyperscalers such as Amazon, Google and Microsoft are collectively investing nearly $750 billion in capital expenditure this year, a jump of nearly 35 per cent. A significant portion of this demand has become circular. Nvidia supplies chips to OpenAI, OpenAI buys cloud infrastructure from Oracle, Oracle buys Nvidia chips, and the cycle continues. When demand starts feeding itself rather than being driven by end consumers, investors naturally begin questioning whether valuations have run too far ahead of fundamentals.

Another perspective is startup funding. The entire Indian AI startup ecosystem attracted roughly $650 million in funding last year. Anthropic alone raised around $65 billion in one funding round. The comparison shows just how extraordinary valuations have become. The concern is that capital raising has moved much faster than earnings.

So, do you believe we are in an AI bubble?

I think there are certainly pockets of a bubble. Whether it bursts dramatically or deflates gradually is something only time will answer. The technology itself is real and transformational. The debate is more about whether expectations have become unrealistic in terms of timing rather than direction.

We have recently seen a sharp correction in the South Korean market and a decline in SpaceX's valuation. What do these developments indicate?

The market appears to be correcting excessive optimism. Investors are overestimating how quickly AI revenues will arrive rather than whether they will arrive at all.

Today, nearly $750 billion of hyperscaler capital expenditure is chasing AI revenues that are expected to be only around $70 billion in 2026. That mismatch is significant.

Companies are already making strategic decisions based on expectations of rapid AI monetisation. TCS, for example, has announced an AI-first strategy and reduced around 23,400 jobs, betting that AI-led revenues will emerge faster. While companies like Anthropic have increased revenues dramatically over the last 18 months, much of that revenue remains concentrated in a handful of firms. Downstream participants continue to burn cash.

Every transformative technology goes through a phase of excessive optimism. How do you distinguish genuine innovation from speculative frenzy?

One principle I have followed since the dot-com era is very simple: follow who is paying for the capital expenditure.

If companies are funding investments through operating cash flows generated by profitable businesses, that usually reflects genuine innovation. If investments increasingly depend on debt or complicated financial structures, it becomes a warning sign.

During 2024 and 2025, hyperscaler investments were largely funded through internal cash generation. This year, however, companies such as Meta and Oracle have increasingly turned to long-term debt. That shift is important. Interestingly, the RBI's Financial Stability Report has also flagged similar concerns.

There are, however, differences from the dot-com era. Data centres require enormous amounts of electricity. Around 16 gigawatts of projects have been announced, but only about five gigawatts are actually under construction. Delays of 50 to 70 per cent are common, while grid connections can take four to seven years. Energy constraints may actually prevent the kind of reckless overbuilding seen during the dot-com boom.

Another useful test is to examine where revenues originate. If revenues come from genuine customers, that signals healthy adoption. If companies increasingly derive revenues from businesses funded by the same investors who finance them, I become much more sceptical.

Whenever we discuss an AI bubble, comparisons with the dot-com boom are inevitable. What are the biggest similarities and differences?

There are several similarities. Market concentration is one. During the dot-com era, capital became concentrated in Nasdaq stocks. Today, a similar concentration exists in the S&P 500, particularly around the Magnificent Seven and semiconductor companies.

Vendor financing is another similarity. Nvidia financing customers resembles strategies adopted by Lucent during the dot-com period.

We are also witnessing peak private-market valuations and an IPO rush, both classic late-cycle characteristics.

However, there are equally important differences. Nvidia trades at around 38 to 45 times forward earnings, whereas Cisco traded at well above 100 times earnings during the technology bubble.

More importantly, Nvidia generates approximately $25 billion in quarterly cash flow. Cisco's quarterly cash generation at its peak was only around $1 billion.

Infrastructure is another key difference. During the dot-com era, laying excess fibre-optic cable was relatively inexpensive. Today, building data centres requires massive investments in power infrastructure, making overcapacity much harder to create.

Many experts argue that early investors in companies like OpenAI and Anthropic could eventually sell to retail investors through IPOs, triggering a correction. Is that a valid concern?

That risk certainly exists.

SpaceX provides a useful example. Its valuation surged from roughly $135 to nearly $225 within days before falling sharply. Many investors who entered after the initial excitement are now barely breaking even.

Whenever private-market valuations transition into public markets, there is often an initial wave of enthusiasm followed by genuine price discovery.

If OpenAI seeks a trillion-dollar valuation or Anthropic comes to market at similarly elevated levels, retail investors need to be extremely cautious. By definition, they are entering at the highest valuation in the company's history.

Personally, I would wait until lock-in periods expire before making any investment decision.

Do you expect valuations to peak around the IPO and then correct?

Predicting market timing is always difficult. These companies will probably command rich valuations when they list. Could prices move even higher after listing? Certainly. SpaceX demonstrated that valuations can rise dramatically in a matter of days.

But markets usually return to fundamentals over time. So while further upside is possible, a subsequent correction is equally likely.

The RBI has warned that inflated AI valuations could pose risks to financial stability. How could an AI correction affect India? Why should Indian investors be concerned?

The first point to understand is the sheer scale of global exposure. Foreign ownership of US equities has risen from around $7 trillion during the pandemic to approximately $21 trillion today. That means any significant correction will have worldwide consequences.

India has already witnessed substantial FPI outflows. If investors become risk-averse globally, emerging markets will almost certainly face additional capital withdrawals. That would weaken the rupee and increase imported inflation.

The second-order effects could be even more significant.

Many hyperscaler investments are increasingly debt-funded. If capital expenditure slows, Global Capability Centres could freeze hiring. We are already seeing slower recruitment across India's IT and GCC sectors.

That has implications well beyond technology. Cities like Bengaluru, Hyderabad, Gurugram and other NCR markets depend heavily on IT employment. Slower hiring would affect office demand, housing markets and consumer spending.

The broader employment trends are already concerning. Freshers' salaries have declined from roughly ₹3.5 lakh to ₹2.4 lakh in many cases. The top five Indian IT companies have collectively reduced jobs this year, while campus hiring remains at multi-year lows. Many graduates from the 2024 batch are still waiting for placements.

The trigger for this discussion itself came after Accenture lowered its guidance, wiping out nearly ₹1.5 lakh crore in market capitalisation from Indian IT stocks in a single trading session. That illustrates how sensitive Indian markets have become to global AI sentiment.

India currently has a relatively small presence in the global AI ecosystem. Could an AI correction eventually benefit Indian technology companies?

Ironically, yes.

Indian IT stocks have already corrected sharply. Nifty IT is down substantially from its peak, and the combined market value of India's largest IT companies is now only marginally higher than Reliance Industries.

If AI infrastructure costs fall following a correction, India's compute costs, GPU costs and talent costs would also decline. Cheaper intelligence ultimately means lower input costs for Indian companies.

There will undoubtedly be pain over the next two to four years. However, over the longer term, lower AI costs could make Indian companies far more competitive.

The biggest beneficiary may actually be India's Global Capability Centres rather than traditional IT services companies. GCCs already employ around 2.3 million people out of India's six million IT workforce and continue to grow at 11-13 per cent annually.

There may initially be a broad sell-off in technology stocks, but once markets stabilise, India's cost advantage could become even stronger.

Indian investors may soon get an opportunity to participate in IPOs of AI leaders such as OpenAI and Anthropic. What would be your advice?

Many sophisticated investors have already gained exposure through global venture capital and private equity funds at valuations of $3 billion, $4 billion or $5 billion. Those investors entered much earlier in the growth cycle.

Retail investors are entering at a completely different stage.

Personally, I would avoid rushing into these IPOs. At the very least, investors should wait until the lock-up period expires and markets have had time to establish a realistic valuation. There will always be opportunities after the initial excitement fades.