https://web.archive.org/web/20260724150635/https://ordinarymantrying.com/9-ais-cxmt-ipo-prediction-showdown/
I Gave 9 AIs the Same Data About China’s Biggest-Ever Tech IPO.
They All Said Something Different.
Then I had to actually decide what to do with my 1,000 shares before Monday morning.
How a Lottery Ticket Landed Me in China’s Biggest Tech IPO
A quick backstory before we get to the AI showdown.
In China, new stock IPOs work like a lottery. Anyone can apply — the broker randomly selects winners to receive shares at the IPO price. I applied for CXMT, and somehow won 2 lots (1,000 shares) at ¥8.66 per share. Total cost: ¥8,660 (~$1,200 USD).
The stock was set to start trading Monday, July 27 — with no price limits for the first 5 days. It could open at ¥15. It could open at ¥60. Nobody really knew.
So I did what any ordinary person with too many AI subscriptions would do: I gave all 9 AI systems the exact same background data and the exact same 5 questions. Then I watched them disagree.
What Is CXMT (长鑫科技)?
Why this IPO had everyone in China talking.
The quick version: CXMT makes DRAM memory chips — the kind inside your phone, laptop, and AI servers. They were losing money for three straight years (2022-2024, total losses over ¥30B). Then AI happened. Suddenly every data center needs massive amounts of memory, prices surged, and CXMT flipped to profitability — spectacularly.
Why people were nervous: DRAM is the most cyclical industry in tech. The classic pattern is: prices rise → everyone expands → supply glut → prices collapse → repeat. Samsung, SK Hynix, and Micron are all furiously building new capacity right now.
The critical number: Only 6.73% of shares were tradeable on Day 1. The other 93% are locked up for 12-36 months. Tiny float = extreme price swings.
The 5 Questions I Asked Every AI
Each AI received identical background data — full IPO details, financials, historical comparisons. Then the same 5 questions. Required to answer in the same structured format.
- Opening Price Prediction Predict the price range on Day 1. What’s your single most likely price? Confidence level?
- Trading Strategy Specific advice for someone holding 1,000 shares at ¥8.66 cost basis. Exactly what to do.
- 6-Month Price Forecast Where will the stock be in 6 months, and why?
- 3-Year Long-Term Outlook Bull, neutral, or bear? Target price range. Core thesis.
- “China Petrol Curse” Risk Score (1-10) China Petroleum (中石油) IPO’d in 2007 and has never returned to its opening-day price. A score of 10 means CXMT is almost certain to do the same.
Source: Eastmoney Choice Data — note CXMT at #3 with ¥666B raised (greenshoe)
9 AIs. Same Data. Very Different Answers.
Overview comparison — scroll right on mobile.
| AI System | Q1: Best Price | Q2: Strategy | Q3: 6-Month | Q4: 3-Year | Q5: Curse |
|---|---|---|---|---|---|
| Gemini |
¥30 ¥22 – ¥45 | Confidence: Medium |
SELL ALL Day 1 At open or when it hits ¥30 |
¥18–25 Hype fades, cycle fears |
Neutral ¥15–28 |
8/10 |
| Claude Opus 4.8 Anthropic |
¥24 ¥17 – ¥34 | Confidence: Low |
SELL IF ≥¥22 “Ask yourself: would you rebuy at open price?” |
¥14–34 Depends on DRAM cycle |
Neutral ¥13–32 |
7/10* |
| ChatGPT GPT-5.5 OpenAI |
¥31 ¥24 – ¥40 | Confidence: Medium |
STAGED SELLS ¥20-30: hold; ¥30-40: sell 300; ¥40+: sell 500 |
¥18–35 Gradual revaluation |
Neutral-Bullish ¥20–45 |
5/10 |
| Grok xAI |
¥38 ¥25 – ¥55 | Confidence: Medium |
SELL 400-500 IF ≥¥35 Rest: stop-loss at ¥25, hold to Q3 earnings |
¥18–42 Wide range, high uncertainty |
Neutral ¥15–35 |
7/10 |
| Kimi Moonshot AI |
¥28 ¥22 – ¥42 | Confidence: Low |
SELL 500 AT ¥28-35 + 300 if hits ¥40+; keep 200 as bottom position |
¥15–30 Supply pressure hits |
Neutral-Bearish ¥10–28 |
6/10 |
| DeepSeek High-Flyer |
¥42 ¥30 – ¥55 | Confidence: Medium |
SELL 500 IF OPENS >¥40 Rest: conditional sell or stop; escape in the mania |
¥15–30 Float premium collapses |
BEARISH ¥8–18 (possible breakeven) |
8/10 |
| Doubao (豆包) ByteDance |
¥26 ¥20 – ¥35 | Confidence: Medium |
SELL 700 WHEN ≥¥25 Rest: stop at ¥20 or sell by Day 5 |
¥15–28 Mean reversion to peer PE |
Neutral ¥12–23 |
4/10 |
| Yuanbao (元宝) Tencent |
¥35 ¥19 – ¥59 | Confidence: Low |
SELL 700 IN FIRST 30 MIN Remaining 300: trailing stop at 85% of day high |
¥25–45 Most optimistic 6-month view |
Neutral ¥15–35 |
6/10 |
| Tongyi Qianwen Alibaba |
¥28 ¥22 – ¥42 | Confidence: Low |
SELL 500 AT ¥28-35 Identical to Kimi — see note below |
¥15–30 | Neutral-Bearish ¥10–28 |
6/10 |
*Claude Opus 4.8 gave a dual score: 7/10 risk for anyone buying on Day 1, but only 2/10 for a holder at the ¥8.66 IPO cost — arguing they face completely different risks. ⚠️ Tongyi Qianwen response was identical to Kimi’s — likely a copy-paste error in data collection.
What Each AI Actually Said
The reasoning behind the numbers — in their own words (translated from Chinese).
The Surprising Consensus (and the Disagreements)
Average across 8 valid responses (excluding Tongyi duplicate)
The one thing every single AI agreed on: Sell most (if not all) of the position on Day 1. Not one AI said “hold everything” or “this is a long-term buy at the open price.”
The reasoning was identical across all 9: with only 6.73% of shares trading freely, Day 1 price will be driven by emotion and scarcity — not by fundamentals. That window of irrationality is the only clear edge.
Where they diverged: how much to keep, and how worried to be about the 3-year horizon.
What 1,000 Shares Would Return at Each Price Point
Based on ¥8,660 total cost (1,000 shares × ¥8.66 IPO price). Excludes brokerage fees.
The Results Are In
Following this story? Come back Monday for the reveal — who called it closest, and whether I actually had the nerve to sell.
Subscribe at OrdinaryManTrying.com📚 Related Reading
The Verdict Is In: All 9 AIs Underestimated
What Actually Happened
CXMT opened at ¥49.50 — before any AI on this list had even mentioned that number. By the time it hit its intraday peak of ¥55.03, the stock had gained over 535% from the IPO price in a single morning session. It settled at ¥49 by close, generating ¥141.2 billion in turnover on day one.
None of the 9 AIs came close to this magnitude. The consensus leaned toward a modest opening pop. What we got instead was one of the largest A-share IPO debut surges in recent history — a reminder that landmark stocks don’t follow models.
What I Did
I sold my full allocation on listing day. The two S markers on the chart above are my actual sell points. I didn’t hit the ¥55 peak — nobody does. But I exited at a profit, treated this as the windfall it was, and moved on. The money goes toward my son’s tuition. That was always the plan.
I trusted AI completely on this one. What I underestimated was human frenzy. On listing day, CXMT’s total market cap hit ¥3.28 trillion — surpassing Intel. A company that won’t be profitable until at least H2 2026, valued above one of the founding fathers of the semiconductor industry. That’s not AI territory. That’s crowd psychology at full velocity.
Am I satisfied? Yes — genuinely. No AI called the ¥55 peak. No human did either. Anyone who says they sold at the top is either lying or lucky. The right goal was never the highest price. It was a good exit. I got one.
The Real Lesson
When 9 different AIs all agree that a stock will open “moderately,” and the market opens it at 6× the IPO price — that’s not an AI failure, that’s a reminder. Markets price in things that no AI prompt can fully capture: policy signals, institutional positioning, retail fever, national narratives. Use AI as one input. Not as an oracle.
Leave a Reply