$2.8 billion. That’s the net amount South Korean retail investors piled into Chinese AI-related assets in the first half of 2023. Not institutional. Not hedge funds. Mom-and-pop traders, chasing a story.
I’ve seen this movie before. In 2018, I watched retail rush into ICOs built on whitepapers that smelled like rotten fish. The pattern is identical: a flashy narrative, a geopolitical hook, and a complete disregard for fundamental metrics. The only difference? The asset class.
Hook: The Data Point That Stops You Cold
Between January and June 2023, Korean retail investors net bought over $2.8 billion in Chinese stocks, ETFs, and ADRs tied to AI. The top picks? Shares of Cambricon Technologies—the so-called "China’s NVIDIA"—along with chip equipment maker NAURA Technology Group (Northern Huachuang), foundry SMIC, battery maker CATL, and local AI startup MiniMax. The flows overwhelmed the broader market: Korean retail net sold $800 million in U.S. stocks during the same period.
This isn’t diversification. It’s a concentrated bet on a single geopolitical narrative: that China can build a parallel AI stack from chips to applications, bypassing U.S. sanctions. Retail is voting with their wallets on a story, not on earnings.
Context: Why Now, and Why Korea?
South Korean retail investors are famously active, high-leverage, and trend-chasing. They flooded into U.S. tech stocks in 2021, then pivoted to Chinese assets in early 2023 as the reopening story gained traction. AI mania—sparked by ChatGPT’s launch—added rocket fuel. By April, Chinese AI stocks had already surged 40% YTD. The FOMO was real.
Globally, the narrative around China’s AI sector shifted. Investors framed it as the only viable alternative to NVIDIA’s monopoly, given mounting export controls. Korean retailers, lacking the ability to verify on-the-ground execution, adopted the simplest heuristic: buy the proxies. The result? A torrent of capital into a handful of unprofitable, high-valuation names.
Core: The Technical Underpinning—A House of Cards
Let’s trace the money. $2.21 billion went into Chinese A-shares and Hong Kong-listed stocks plus ETFs specifically targeting AI and chips. Another $670 million bought individual Chinese AI stocks through ADRs. The top individual stock purchases were: - Cambricon Technologies ($380 million): A designer of AI accelerators, still deeply unprofitable, with revenues heavily dependent on government contracts. The "China’s NVIDIA" tag is more marketing than reality—Cambricon’s chips target inference for surveillance and edge, not massive training clusters. No CUDA-like ecosystem exists. No viable path to challenging NVIDIA in the cloud. - NAURA Technology Group ($340 million): A semiconductor equipment maker. Beneficiary of domestic fab expansion. Real business, but valuation priced for 5 years of double-digit growth in a single quarter. - SMIC ($260 million): China’s largest foundry, stuck at 7nm and bleeding profits due to sanctions. The retail bet is that SMIC will leapfrog to advanced nodes. Physics says otherwise. - CATL ($280 million): A battery maker, not an AI company. This inclusion exposes the "AI basket" as a catch-all for China tech optimism. - MiniMax (via ADRs, $110 million): A startup building large language models. Little public data on revenue, user base, or API pricing. Valuation set by whisper numbers.
Breaking this down, $2.8 billion wasn’t distributed rationally. It concentrated in names offering the boldest narrative—regardless of unit economics.
From my experience running arbitrage strategies on Uniswap V2, I learned that liquidity chasing hype is the fastest way to get trapped. When the TVL on a pool was entirely driven by yield farmers, the exit was always a crash. Same pattern here: retail liquidity is the last wave before the reversal.
Contrarian: The Blind Spots Nobody Wants to Discuss
The mainstream take is bullish: Korean retail sees the future, and they’re buying the dip. But here’s the counter-narrative.
First, the "China’s NVIDIA" thesis is a fantasy. Cambricon’s latest AI chip (Siyuan 590) reportedly achieves around 10-15% of the performance of NVIDIA’s A100 in training tasks, and software compatibility is minimal. Enterprises won’t migrate unless forced. The narrative ignores the multi-year lock-in of NVIDIA’s CUDA ecosystem and the lack of any Chinese alternative for large-scale training. This isn’t 2019 Intel vs. AMD; it’s a 10x performance gap.
Second, the valuation disconnect is extreme. At the time of peak influx, Cambricon traded at 100x forward sales. NAURA at 15x sales, vs. Applied Materials at 5x. Retail isn’t calculating; they’re anchoring on a "China premium" that may evaporate overnight if U.S. export controls tighten further.
Third, the regulatory and geopolitical risk is completely unhedged. If the U.S. bans all AI chip sales to China (including design software), the entire narrative collapses. Korean retail has no protection. They are long a binary event with no stop-loss.
Fourth, the "ETF passivity" effect. Over $400 million flowed into Hong Kong-listed China chip ETFs. When the tide turns, redemption mechanisms will force selling of underlying stocks regardless of price. This is the same liquidity cascade we saw in crypto’s 2022 unwinding—the death spiral is baked into the instrument.
Arbitrage opportunities don’t last long when the crowd is piling in. This one is already closing.
Takeaway: What to Watch Next
The $2.8 billion inflow is a textbook case of narrative-driven price discovery untethered from fundamentals. It will end in one of two ways: either the Chinese AI agents (the technology, not the companies) actually deliver a breakthrough that justifies the multiples, or the hype deflates as reality sets in.
The first signal to watch is the next quarterly earnings for Cambricon and NAURA. If revenues disappoint—and they likely will—the story loses its anchor. The second is any escalation in U.S. chip export rules. The third is a shift in Korean retail net flows back to U.S. markets.
Hype is a trap; data is the only map I trust. Right now, the map shows a retail cohort chasing a mirage across a desert of unrealistic expectations. The smart money is already positioning for the exit.
Final question: when Korean retail finally realizes they’re holding bags of unprofitable Chinese tech with no moat—who will they sell to?
