Why FIIs Are Selling India: The AI Chip Bubble Behind It

Samsung Electronics reported second-quarter 2026 revenue of KRW 171.5 trillion, an all-time quarterly high, up 130% year-on-year and 28% quarter-on-quarter. Operating profit reached KRW 89.5 trillion (roughly $62 billion), up an extraordinary 1,814% year-on-year, beating analyst estimates. The Device Solutions division, which houses Samsung's memory chip business, generated KRW 89.2 trillion of that operating profit on the back of surging demand for high-bandwidth memory (HBM) used in AI servers.

SK Hynix posted its own record quarter: operating profit of KRW 60.54 trillion (roughly $41.5 billion), up 557% year-on-year, on revenue of KRW 79.32 trillion (roughly $54.3 billion), up 257%. Its operating margin hit 76%, an all-time record for any memory chipmaker, though the company actually missed analyst estimates, a reminder that even historic results can disappoint when expectations have run further ahead. Together, the two companies earned roughly $104 billion in operating profit in a single quarter.

What makes this quarter more than just a big number is how concentrated it is. Samsung is a sprawling conglomerate spanning memory chips, smartphones, displays, consumer electronics, and telecom equipment. Yet the Device Solutions division alone generated nearly the entirety of the company's operating profit. The rest of Samsung combined, mobile, display, consumer electronics, and networks, contributed close to nothing. Samsung's mobile division, the Galaxy smartphone business, actually posted its first ever operating loss this quarter.

The market's reaction to this data reflects that unease. On the day these results were announced, South Korea's KOSPI index triggered back-to-back circuit breakers, the first time in the exchange's history that has happened on consecutive trading days. Extraordinary profit and extraordinary volatility arrived together, which is usually a sign that a narrative has become the dominant force in pricing, more than the underlying fundamentals alone.

A meaningful share of this year's FII selling in Indian equities hasn't been about India specifically. It has been global capital chasing the AI chip trade in Korea and Taiwan instead. India's weight in the MSCI Emerging Markets Index fell from roughly 20% in mid-2024 to around 11% by mid-2026, while Korea and Taiwan's weights climbed to roughly 23% and 26% respectively. That rotation shows up directly in index returns: KOSPI is up sharply for the year and Taiwan's index has rallied strongly too, while Sensex and Nifty have both been in negative territory for stretches of 2026.

Samsung and SK Hynix's results are the clearest possible illustration of why that capital moved. When two companies can generate $104 billion in three months on a single product category, global index-tracking and momentum-driven capital follows the earnings, almost mechanically. The question worth asking is not whether that capital was right to move, it was chasing a genuine and enormous profit pool, but whether a profit pool this concentrated in two companies and one product category is a stable place for that capital to stay indefinitely.

A bubble has never been about whether the underlying technology or demand is real. AI memory demand is genuinely enormous right now, Samsung and SK Hynix's order books and multi-year supply contracts confirm that this isn't a one-quarter fluke of accounting. A bubble is a question of crowd psychology and pricing: whether too much capital has piled into too narrow a theme with too much enthusiasm, setting up a sharp repricing whenever sentiment eventually turns, regardless of how real the underlying business is. Railways in the 1800s were real. The internet in 1999 was real. Both were still priced as bubbles at their peak.

The back-to-back circuit breakers on KOSPI, the fact that Samsung's non-memory business is contributing almost nothing to profit, and the sheer scale of $104 billion concentrated in two companies and one product category are all consistent with a market that has become narrative-driven rather than purely fundamentals-driven. That doesn't mean a correction is imminent, and nobody, including the companies themselves, can reliably call the timing. It means the risk is asymmetric: a lot of capital is now betting on a continuation of an extraordinary trend holding for years, and history suggests these trends eventually normalise, even when the demand driving them was real all along.

The case for a return: if a market this concentrated eventually corrects, even partially, some of the capital that left India chasing the AI chip trade could rotate back into other large, liquid emerging markets, India among them. Index weights that shifted mechanically on the way out can shift mechanically on the way back in, once the underlying trigger (extreme relative earnings growth in Korea and Taiwan) fades.

The case against a quick return: FII allocation decisions are slow and benchmark-driven, not something that reverses in days. India's own market has just been through a rough patch of its own (a geopolitics-and-oil-driven selloff, covered in our last two digests), which doesn't automatically make it the more attractive destination the moment Korea and Taiwan wobble. Relative conviction matters as much as any single market's weakness.

None of this is a reason to avoid technology or AI-linked exposure altogether, most well-diversified Indian and global equity funds already carry some exposure to the broader theme through their existing holdings. It's also not a reason to try to time an FII return to India, since nobody, including institutional investors themselves, can reliably call that timing. What it is a reason for is resisting the temptation to concentrate fresh money into a single narrow theme purely because the headline numbers are extraordinary, whether that's an AI-chip-themed fund, a Korea or Taiwan-focused allocation, or the inverse bet that India is now "cheap" simply because FIIs have been selling.

A standard, diversified SIP already gives you a reasonable, risk-managed way to participate in the parts of this story that are durable (AI infrastructure spending is real and will likely continue for years) while staying insulated from the parts that are concentrated and narrative-driven (two companies, one product category, one quarter's extraordinary number). Continuing that discipline, through Korean circuit breakers, Indian selloffs, and whatever comes next, remains a steadier way to compound wealth than trying to correctly predict which market or theme wins next.

Quick Questions