As the forced deleveraging cycle approaches its final stages, signs of a renewed artificial intelligence super-cycle are emerging, according to financial intelligence reports. Wall Street giant Morgan Stanley has upgraded South Korean equities to “overweight,” stating that the recent “position unwinding” offers investors a better entry point to re-engage with AI infrastructure and industrial super-cycle themes.
Another major Wall Street bank, Citigroup, has set an even more aggressive target for South Korea’s benchmark index. With foreign investors net buying approximately 7.2 trillion won of KOSPI stocks on July 31, setting a single-day record, and with local pensions and funds also turning to net buyers, Citi maintains a bold 10,000-point target for the KOSPI. The bank argues that memory chip profitability, extremely low valuations, and policy reforms are sufficient to support further market revaluation.
Samsung and SK Hynix are providing a crucial floor for the market. The deleveraging process is nearing its conclusion, shifting capital from speculative frenzy to the AI industrial chain’s super-cycle.
In a report, a team of Morgan Stanley strategists led by Daniel K. Blake stated that the KOSPI index still has a 36% upside potential to their 9,000-point target, following a significant unwinding of crowded trades and leveraged positions. Previously, the financial giant had given South Korean stocks a neutral “equal-weight” rating. After a record 18% surge in the previous session, the benchmark index fell as much as 5.5% on Monday, but losses narrowed to trade around 6,366 points.
The analysts said the recent sharp sell-off was “primarily driven by technical factors” and that “the deleveraging process for extreme leveraged ETF positions, hedge fund leverage, and retail margin accounts has passed the most violent midpoint of the sell-off.” The KOSPI, along with mega-cap stocks like Samsung Electronics and SK Hynix, remains highly volatile during this cleansing process. The index has fallen over 30% from its June peak.
In July, traders rapidly withdrew from the South Korean market, a key barometer for Asian AI demand. The surge in single-stock leveraged ETF holdings, alongside a heavy concentration of the index in Samsung and SK Hynix, exacerbated the decline. Following frequent circuit-breaker plunges in July, South Korean regulators have moved to restrict the use of such products, planning to limit retail investors’ exposure to a certain percentage of their total portfolios.
Morgan Stanley expects the KOSPI to trade in a wide range from 5,500 points to 10,500 points, believing that Samsung Electronics and SK Hynix, which together account for 50% of the index weight, will provide significant valuation support. The firm also anticipates tailwinds for stocks in the industrial, defense, and financial sectors. Morgan Stanley also upgraded Thailand’s stock rating from “equal-weight” to “overweight,” citing increasing investment opportunities driven by improved foreign direct investment and competitiveness. Conversely, the bank downgraded Australian stocks from “equal-weight” to “underweight,” citing limited upside potential following multiple interest rate hikes and tax reforms that have weakened local property investment incentives.
Citigroup is targeting 10,000 points for the KOSPI, recovering the pricing power for bullish sentiment. The recent correction in South Korean semiconductor stocks is being redefined by Wall Street as a dramatic liquidation of AI-related leverage and positions, rather than a reversal of chip fundamentals. The ambitious targets from Morgan Stanley (9,000 points) and Citi (10,000 points) are not a bet on a return to speculative frenzy, but rather a wager on the market’s shift from leverage-driven growth to the earnings, cash flow, and strategic scarcity of AI capital expenditure.
The massive 7.2 trillion won net foreign inflow on July 31, a record, combined with regulators raising the entry barriers for leveraged ETFs, signals that the negative feedback loop of “margin calls, forced selling, and amplified volatility” is weakening significantly. This is a key reason for Citi’s bullish view on fund flows. Crucially, the real demand for AI computing infrastructure has not collapsed alongside stock prices. South Korea’s July exports surged 62.8% year-on-year to $98.89 billion, with semiconductor exports soaring 179% and computer exports climbing 404%, directly confirming that AI data center investment, memory prices, and server demand remain in a high-growth phase.
Amazon has raised its 2026 capital expenditure plan from $200 billion to approximately $220 billion, while its cloud computing arm, AWS, reported 37% revenue growth, its fastest pace in 18 quarters. Despite this massive spending, management still expects they cannot meet all demand in 2026, with supply constraints persisting into 2027, and they describe the already-visible demand for 2028 as “staggering.” This suggests that the South Korean market’s rebound may not be driven merely by short-covering, but is beginning to receive fundamental confirmation from orders, exports, and cloud demand.
The “barbell” demand structure described by Amazon’s management reveals why the AI computing cycle may be steeper and longer. On one end, frontier labs like OpenAI and Anthropic are advancing inference, agentic AI, and recursive self-improvement, with AI increasingly involved in coding, experiment design, and model development. This creates a potential positive feedback loop where “using AI to develop stronger AI” consumes more computing power. On the other end, millions of enterprises have yet to fully launch production-level inference workloads, which will eventually spread from customer service, risk control, and code assistance to complete business processes. Anthropic, a leader in AI application development, has publicly stated that AI systems are taking on an increasing share of AI research and development. OpenAI has also disclosed that GPT-5.6 showed significant progress in evaluations measuring recursive self-improvement capabilities. The core of this AI infrastructure structure is that Labs AI works to increase the slope of the demand curve, while Enterprise AI works to expand the demand base and extend the cycle.
Comments