Playing Without Seeing the Cards

What a 90-day round trip in Lenovo taught me about deciding with limited information, in markets and in life

At the end of July, I bought shares of Lenovo Group in Hong Kong at around HK$21. By mid-September they were trading near HK$38. By late October they were back near HK$22.¹

The whole round trip took less than 90 days.

The business did not reverse as dramatically as the share price did. Lenovo was still the world’s largest PC maker, and its August results showed a real improvement in server profitability. But those results changed what investors knew. Later, IDC’s preliminary figures showed Lenovo’s PC shipments down 22.6% year on year in the third quarter, while worries about AI financing grew. There was new evidence on both sides. I cannot put an exact percentage of the rise or fall down to fundamentals rather than expectations.²

That raised a question I find more interesting than the stock itself: when the evidence about a company changes, but its price swings even more violently, what do we actually know when we make a decision?

The more I thought about it, the more investing looked like Texas Hold’em. You never see your opponents’ cards. You don’t know what the next card on the table will be. You decide whether to call, how much to bet and when to fold, using only what’s in front of you. Most of the important decisions in life work the same way.

This essay uses my Lenovo trade as a case study for a broader question: how do you make good decisions when you can’t see all the cards?

Prices are bets on stories, not facts

A rough mental model for a share price—not a valuation formula—is:

Price ≈ today’s profits × confidence in the future ÷ the price of money

All three moved during those 90 days, but in very different ways.

Today’s profits did change, and that part was real. When Lenovo reported in mid-August, its infrastructure business reached a 9.1% operating margin, and the stock closed about 20% higher that day. The report gave investors a real reason to reprice the business. I cannot measure how much of the move from 21 to around 34 was earnings and how much was excitement about what might follow.²

Confidence in the future likely did much of the remaining work, in both directions. It helped carry the stock from 34 to 38 in September, then faded in October.

The price of money, meaning interest rates, also rose. The US 10-year Treasury yield went from about 4.5% at the start of July to about 5.3% at the end of September, although the path was not steady.³

Companies change slowly. Opinions about their next ten years can change overnight. And the more a stock’s value depends on the distant future, the more sensitive it is: shave a point off long-term growth or add a point to the cost of capital, and a valuation can drop by a quarter or more.

In 1981, Robert Shiller showed that stock prices swing far more than changes in dividends can justify under a simple present-value model. That result is a reason to question the model’s assumptions and the stability of expectations, not proof that any particular price move is irrational.⁴

So I’ve stopped thinking of a company’s “true value” as a fixed number waiting to be discovered. It depends on events that haven’t happened yet, which makes it a range of possibilities with a rough center. A model can’t reveal the truth. What it can do is show you which assumptions matter most, and how much. As the statistician George Box put it, all models are wrong, but some are useful.⁵

Grading my own hand

In Thinking in Bets, former professional poker player Annie Duke describes a trap she calls “resulting”: judging a decision by how it turned out. Win the pot and you think you played well. Lose it and you think you played badly.⁶

But good decisions lose all the time, and bad ones sometimes win. The right question after a hand isn’t “did I win?” It’s “given what I knew then, was that the right play?”

By that standard, the Lenovo trade comes down to three decisions.

Buying at HK$21: a good decision. The market was still pricing Lenovo as a PC company while its server business was taking off. The gap between perception and reality was wide, the downside looked limited and the upside large. That’s a hand with good odds.

Holding at HK$30 through earnings: also a good decision, for the right reasons. My original plan was to sell around 30. But I thought an earnings beat was likely, and the payoff looked lopsided. With the market this enthusiastic, a mediocre report probably wouldn’t sink the stock much, while a strong one could push it into the high 30s. It worked out. But I have to be careful here: the decision was good because of the reasoning, not because the stock went up. Had the results disappointed, the same call would still have been sensible.

Not selling at HK$36–38: the one that cost me. My plan going in was simple. Hold through this earnings season, then sell. I knew AI spending at this pace couldn’t last forever and would eventually hit a nonlinear ceiling. If the rally lasted past earnings, the trade would be a success.

Then the results came out. I listened to the earnings call and came away convinced the growth would continue, which was a reasonable view. Many brokers raised their targets into the 40s and 50s. Next to that, my own expectations felt conservative.

Looking back, the mistake wasn’t the forecast. Lenovo’s next quarter may well be strong too. The mistake was that I quietly rewrote my plan after new information arrived, without protecting myself against the possibility that the new view was wrong. The old plan said sell after earnings. The new view said hold. A sensible compromise was sitting right there: sell part, keep part.

In poker terms, staying in the hand after a good flop is fine. Sitting on a big win without taking a single chip off the table is not.

The warnings I missed, and the ones that came later

What I regret most is that at 37 or 38, I barely thought about what might stop AI spending. It wasn’t that the question had never crossed my mind. It just stayed in the “someday” category until bonds issued by CoreWeave, a heavily indebted company that rents out AI computing power, sold off sharply. The yield investors demanded on that debt rose, and “someday” became now.⁷

Looking back, some warning signs were visible before the peak; others, including the sharp move in CoreWeave’s bonds, became clear only afterward. I could not have known the October outcome in September. I could have been more alert to what might make my optimistic case fragile:

• AI stocks had already sold off once, in July. People were openly asking whether the spending would ever pay off. Lenovo was simply too strong at the time, and its earnings drowned out the doubt.
• Credit markets were sounding an alarm about AI financing. The yield on one of CoreWeave’s bonds climbed from about 11% in August to above 15% by late October. That is not a direct measure of Lenovo’s credit risk, and most of the move came after Lenovo’s September peak. It was still a reason to revisit how easily the AI buildout could be financed.⁷
• Rates had been rising all quarter. October didn’t come out of nowhere.
• The US market was historically expensive. The Shiller P/E hovered around 40 all summer, near levels associated with the dot-com era.⁸
• Other signals sat on the table too, from the upcoming US midterm elections to the possibility that customers were buying PCs early to get ahead of rising memory prices.²

So the problem wasn’t a lack of information. It was attention. In a rally, we naturally gravitate toward the facts that confirm what we already believe: booming orders, rising price targets, cloud companies saying they can’t build fast enough. The contrary evidence doesn’t disappear. It just gets filed as noise, until some event forces us to re-file it as signal.

There was an even simpler warning: the price itself. Later I built a deliberately pessimistic valuation of Lenovo. Even its most optimistic scenario came in well below 38. Had I run that model at the peak, I would have seen that the stock had already outrun my own best case. I didn’t need to predict CoreWeave’s troubles. I only needed to admit that the odds had flipped.

That’s the distinction I keep coming back to. Predicting when something will break is nearly impossible. Recognizing that a system has become fragile is not. You can’t say which grain of sand will collapse the pile. But you can see that the pile is getting steep: rising rates, concentrated debt, stretched valuations, and everyone agreeing with each other.

What the poker table teaches

I play Texas Hold’em now and then, as an amateur. But four ideas from the table map almost perfectly onto investing.

Pot odds: only play hands worth playing. In poker, you compare what it costs to call with what you stand to win. A hand that wins only 30% of the time is still worth playing if it pays five to one. I like asymmetric bets for the same reason. The goal isn’t to find sure things. It’s to find situations where you lose a little if you’re wrong and win a lot if you’re right. Buying Lenovo at 21 was that kind of hand.

Bet sizing: stay in the game. Even the best hand can lose. Professionals care less about winning any single pot than about never losing so much in one hand that they’re out of the game. In investing, that means matching position size to conviction, and always leaving room to be wrong.

Reading the table, not just your cards. Good players watch how others bet and whether the table is playing loose or tight. Markets are the same. My cards, Lenovo’s results, were excellent the whole time. What I didn’t read carefully was the table: rates, credit, valuations, the mood of everyone else. A strong hand doesn’t guarantee you win the round.

Folding: the hardest skill. Beginners hate to fold, especially after they’ve put a lot in. But folding isn’t surrender. It’s saving your chips for a better hand when the odds turn against you.

My decision to sell Lenovo gradually into any rebound is a fold. It isn’t because I’ve stopped believing in the company. It’s because my read of the whole table has changed, and I have another hand I’d rather play.

There’s a trap here worth naming, though. Folding is not the same as tilt, the poker term for chasing losses emotionally. The money is going into Gabrielle-Optech (戈碧迦), a small Chinese maker of optical glass. I’m treating it more like a venture investment than a trade: a multi-year bet that demand for optical materials will grow sharply as more technology runs on light. To tell a fold from tilt, I ask one question: would I make this new bet even if the Lenovo trade had never happened? If the answer is yes, it’s a reallocation, not a revenge trade.⁹

Two kinds of fog

I spent more than a decade in venture capital before shifting most of my attention to public markets in the last couple of years. Both involve deciding under uncertainty, but the fog is very different.

In private markets, you have little information but also little noise. Backing a startup, you often have a pitch deck, a handful of meetings and your own read of the industry. But you’re in for five to ten years, with no price flashing at you every day. So you focus on the business and the people: is the product real or just a story, who actually pays, what are the margins, can the founders execute?

In public markets, you have lots of information and lots of noise. Listed companies publish far more, and analyst notes are everywhere. But the price moves every day, and it carries emotion. The fundamentals can stay the same while the market swings from euphoria to despair. Public markets are also far more sensitive to the big picture: interest rates, liquidity and the economy show up in prices almost immediately.

The habits I brought from venture, focusing on businesses, people and the long term, served me well with Lenovo. But they also had a blind spot. In venture, macro swings and market moods tend to average out over a long holding period. In public markets, they can decide the outcome within weeks.

Put differently: venture tests whether you can pick the right company. Public markets test that too, and also whether you can read the whole table.

When everyone has the same information

I argued with AI the whole time I was writing this. I often have different models take opposite sides, one bullish and one bearish, let them attack each other’s case, and then make up my own mind.

After doing this for a while, I’ve reached a slightly uncomfortable conclusion: AI won’t make you money by itself. It gives everyone more information, faster. That also means consensus forms faster. An insight that once took months to spread now spreads in days. The window in which being early pays off keeps shrinking.

So what’s still scarce? I can think of two things.

The ability to see what others haven’t priced in yet. Nothing in this story was secret. Interest rates and the Shiller P/E were public throughout; CoreWeave’s bond selloff became visible later. What was scarce was asking early enough which part of the AI financing chain might fail my thesis.

Discipline. Between knowing and doing sits human nature: greed on the way up, fear on the way down. AI can’t change that. But oddly, it can help. You can write your rules down in advance and have a machine remind you, at your most excited moment, that by your own rules it’s time to sell some.

In a world where information is abundant, the edge comes less from knowing more and more from seeing earlier and actually acting on it.

From knowing to doing

The biggest lesson of this trade is that I didn’t lack insight. I knew AI spending had a ceiling. I knew rates were rising. What I lacked was a bridge from knowing to doing.

That bridge is a set of rules written down in advance. These are the ones I intend to keep:

1. Before buying, write down how you could be wrong. This is the “pre-mortem”: imagine it’s a year later and the investment has failed, then ask why. Turn the most likely answer into something you can watch. Had I written “AI financing costs rise” in July, I would have watched credit conditions throughout the rally instead of noticing them only after the selloff.
2. Watch the weakest link. Every cycle has a most fragile point. This time, heavily indebted companies renting out AI capacity deserved close attention. Their bonds were a useful warning signal. When stocks and credit markets disagree, I want to understand why.
3. Let price trigger action. If a stock moves above the best case in my own model, I start selling. It’s a mechanical rule, and it doesn’t care how good the story sounds.
4. You can change the plan, but protect yourself when you do. Updating your view on new information is right. But when you revise a plan, execute part of the old one first. If I had sold some shares after earnings and held the rest on my new view, this would be a very different essay.
5. Handle uncertainty with position size, not with forecasts. You’ll never know the whole truth. So size each bet to match your conviction, and make sure no single hand can knock you out.

None of these rules guarantees a win. They do one thing: over many decisions, they keep me on the right side of the odds more often.

Life is played the same way

As I wrote this, I realized none of these ideas belongs only to investing. Most of what I believe about markets, I believe about life.

Almost every big decision is made without seeing the cards: whether to change careers, move to a new city, or start something from scratch. When I left institutional investing to manage my own money, I didn’t know how it would turn out either. The same principles apply.

• Judge the decision, not the outcome. A careful choice can turn out badly, and a reckless one can turn out fine. Ask “was that right, given what I knew?” instead of “did it work?” That’s how experience actually accumulates, instead of luck masquerading as skill.
• Look for asymmetry. Trying something new often costs little, while what it could open up is large. Those hands are worth playing more often.
• Never let one hand knock you out. Health, family and basic financial security are the chips that keep you at the table. No opportunity is worth staking all of them.
• Fold when you should. Having invested a lot in something isn’t a reason to keep going. Sunk costs don’t justify the next bet. Only the odds do.
• Read the table. Effort matters, but so does timing. Even a great hand can be the wrong one to play at the wrong moment.

Limited information isn’t a problem to be solved. It’s the game itself. We’ll never see all the cards. What we can do is keep improving our models, so that each decision lands a little more often on the right side of the odds.

A last word

Making mistakes is fine. When you decide with limited information, mistakes are guaranteed, just as the best poker players lose plenty of hands. What matters is separating luck from judgment when you look back, and turning each lesson into a rule that makes the next decision a little better.

With Lenovo, I bought well, played the middle of the hand reasonably, and failed to take my chips off the table at the end. That was the most expensive lesson of the trade, and probably the most useful.

This is a personal reflection, not investment advice.

Sources and notes