We traded sleep for alpha, and alpha for scars.
That line haunts me every time I see a retail crowd chase a narrative without first asking: who is the exit liquidity?
South Korean retail investors just dropped $2.8 billion on Chinese AI assets in the first half of 2023. That figure is not a rounding error. It’s a coordinated, cross-border wave of capital flowing into stocks like Cambricon (the so-called “China’s NVIDIA”), Naura Technology (semiconductor equipment), SMIC (foundry), and even the AI startup MiniMax. At first glance, it looks like a vote of confidence in China’s ability to build a parallel AI stack—free from the grasp of US export controls.
But here’s the thing: I’ve been in this game long enough to know that when retail piles into a narrative with this much velocity, the smart money is usually on the other side of the trade. I was a junior quant during DeFi Summer. I saw the same pattern in 2020: yield farmers chasing absurd APYs while the LPs were already pricing in impermanent loss. The names change. The structural dynamics don’t.
Context — The Post-ETF, Post-Bitcoin World
The backdrop matters. By mid-2023, the Spot Bitcoin ETF approval had already shifted the crypto market into a Wall Street plaything. Satoshi’s vision of peer-to-peer electronic cash was effectively dead—replaced by institutional custody, OTC desks, and block trades. Bitcoin became a macro asset, not a monetary revolution. The same forces that turned BTC into a toy for grayscale and BlackRock are now reshaping AI investing.
Korean retail investors, long known for their risk appetite (remember the 2018 “kimchi premium” on BTC?), found a new narrative: “China’s AI self-sufficiency.” The US-China tech war created a supply chain vacuum, and retail investors filled it with conviction. They bought the equipment makers (Naura), the foundry (SMIC), the chip designer (Cambricon), and the application layer (MiniMax). This is not a diversified portfolio. It is a concentrated bet on a single thesis: China can bypass NVIDIA’s CUDA moat and build a homegrown AI infrastructure.
But is that thesis backed by technical reality? Or by hope dressed in the colors of a bull market?
Core — Tracing the Order Flow
Let’s dissect the asset-level flows. According to the available data, the $2.8B net buy was split across several channels: direct A-share purchases ($678M), Hong Kong-listed stocks and ETFs ($209M), and a large residual ($2.1B) likely consisting of other equities and derivatives. The concentrated bets were on names like Cambricon (AI chip), Naura (equipment), and SMIC (foundry). MiniMax, a generative AI startup, also appeared—a rare foray into unlisted or pre-IPO territory for retail.
Cambricon is the poster child. It trades at multiples that would make any value investor wince. In 2023, it was still deeply unprofitable, relying on government and state-owned enterprise contracts. Its AI chips are ASICs, not GPUs. That means they target specific inference workloads—not the general-purpose training that fuels the current AI boom. The “China’s NVIDIA” label is a marketing gimmick, not a technical equivalence. NVIDIA’s strength is its ecosystem (CUDA, TensorRT, cuDNN) and its ability to scale across training and inference. Cambricon has none of that.
Naura Technology and SMIC represent the manufacturing backbone. But SMIC is constrained by US export controls on advanced lithography machines. It can produce at 7nm via multiple patterning, but yields are lower and costs higher. The idea that SMIC can compete with TSMC on leading-edge nodes is a fantasy. Yet retail investors bought SMIC stock as if it were a direct substitute.
MiniMax adds another layer of speculation. Buying an AI startup through secondary markets implies a belief that Chinese LLMs can thrive under limited compute. That is possible, but the evidence is thin. Most Chinese AI labs rely on open-source models (e.g., Llama, Mistral) and struggle to monetize. MiniMax’s valuation in 2023 was already elevated, driven by the narrative that “China needs its own OpenAI.”
I spent 13 years watching capital flow into narratives that looked compelling on the surface but crumbled under forensic scrutiny. This is one of those moments. The price action tells you what retail wants to believe. The order flow tells you who is selling them the dream.
Chaos is just a pattern waiting for a label.
Contrarian — The Blind Spots
Let me give you three reasons why this move feels like a trap.
First, retail is the counterparty, not the catalyst. When Korean retail buys $2.8B of Chinese AI stocks in six months, you have to ask: who is selling? The answer is often institutions, arbitrageurs, and early insiders. They use retail enthusiasm to offload positions at inflated valuations. The data shows that Korean retail was a net buyer, but I haven’t seen evidence of institutional accumulation. In fact, the ETF flows—passive vehicles that aggregate retail sentiment—suggest the opposite.
Second, the valuation detachment is extreme. Cambricon’s price-to-sales ratio in mid-2023 was over 50x. NVIDIA’s was around 40x. The difference? NVIDIA actually generates $60B+ in data center revenue. Cambricon’s revenue was barely $100M, and most of that came from state-backed projects. This is not a discount; it’s a premium for unproven execution.
Third, the geopolitical tailwind is a double-edged sword. If the US expands export controls (e.g., banning the sale of EDA tools needed for ASIC design), Cambricon’s supply chain breaks. If China retaliates by restricting capital outflows, Korean investors could face a liquidity lock. This is not a bet on technology; it’s a bet on a specific regulatory outcome. And regulators are unpredictable.
The yield was real; the trust was phantom.
Takeaway — The Scar Trade
So where does that leave us? I’m not saying Korean retail is wrong. In crypto, being early and being wrong are the same thing until they aren’t. But the probability of this trade working out is low, and the asymmetry is dangerous.
The real signal here is not about Chinese AI. It’s about the global appetite for narrative-driven speculation. In a bear market, capital flees risk. But when a high-conviction narrative emerges—like “China’s AI independence”—retail will chase it, even with leverage. We saw this with Terra, with FTX, with the NFT mania. The pattern is universal.
My advice: watch the derivatives data. Look at Korean retail’s margin debt levels. Monitor the ETF premium. When those start to contract, the exits will get crowded. Because in the end, hope is a terrible hedge against a black swan.
I didn’t come here to tell you what to buy. I came to tell you what to question.
