You bought NVIDIA, it went up 30% and you sold, thinking you were quite clever.
Then it went up another 120%. You stared at the K-line chart for five minutes, getting angrier the more you thought about it.
This is actually what most retail investors look like.
In 2022, someone did the exact opposite of the crowd. He didn't buy NVIDIA; he bought a company that supplies NVIDIA—a niche stock that 90% of people had never heard of with a market cap of $700 million, $AXTI. At the time, the stock price was $12, but it later rose to over $70.
This person is Serenity, who has become an explosive sensation in investment circles both at home and abroad this year. Out of 35 publicly disclosed positions, 31 rose, yielding a 225x return. Even Bloomberg and Reuters have been following and reporting on his tweets.
After scrolling through dozens of his posts, I found that what he does is different from start to finish compared to what the public does when buying NVIDIA.
What did the public do when buying NVIDIA? They looked at the PE ratio, the financial report growth rate, news saying AI was about to explode, saw institutional funds buying, and then placed an order. I could do that sequence with my eyes closed.
What did Serenity do before buying $AXTI?
He started with NVIDIA's GPUs and drew a supply chain map. For GPUs to run, you need data centers; for data centers to transmit data, you need optical modules; the core component in optical modules is the laser; and the raw material for lasers is called indium phosphide. Then he did something I never even thought of: he checked the global production capacity distribution of indium phosphide.
There are only two companies in the world capable of mass-producing indium phosphide substrates. $AXTI accounts for one-quarter to one-third of that.
In other words, as long as AI chips continue to be built in the future, all optical module factories must buy raw materials from these two companies. Furthermore, this landscape won't change in the short term because the cycle from factory construction to mass production is measured in years.
Then he went through the company's patent documents, customer lists, capacity limits, and upstream mineral sources. Only after checking everything did he place the order.
The Perilla Leaf Theory
He named this strategy the "Perilla Leaf Theory."
He says if you go to a high-end sushi restaurant, everyone is staring at the fatty tuna (Otoro). But what the kitchen truly cannot afford to run out of is the perilla leaf. Without tuna, the menu loses a few dishes; without the perilla leaf, the whole shop might as well close.
In the AI industry chain, NVIDIA, Microsoft, and OpenAI are the fatty tuna. The perilla leaf is a material with a name that's hard to pronounce—a niche company with a market cap of a billion or so and no analyst coverage, a component with only two suppliers globally where the entire chain stops without it.
This theory breaks down into three steps.
Step 1: Start from the top-level demand and trace down layer by layer
AI explosion → GPU demand surges → GPU manufacturing needs lithography machines → The core component of lithography is the lens → Who makes the lenses globally? Zeiss, and only them. Keep going. Who supplies the special glass used in Zeiss lenses? Perhaps another small Japanese factory.
At each layer, ask the same question: For this layer to work, what thing in the next layer is irreplaceable by others?
Most people might stop at the second layer and start discussing whether NVIDIA's PE is too expensive. But Serenity deconstructs down to the fifth or sixth layer, until he reaches the company with the smallest market cap and the most unfamiliar name.
Step 2: Count the players in that specific segment globally
More than three? Pass! Because there is sufficient competition and no pricing power.
Two? Keep an eye on them.
One, or a substantial monopoly? That's the one.
The logic is simple. As AI expands, more money flows upstream along the supply chain. A rising tide lifts all boats.
But if there is only one boat in a certain segment, it doesn't just rise with the tide; it can turn around and blackmail the downstream: you can only use me, and I call the shots.
Step 3: Before buying, post your analysis and wait for people to attack it
You don't need to find people who agree; we want to do the opposite and specifically wait for experts to refute us.
If someone points out that a step in his logic jumped too fast, he goes back and re-evaluates. If someone tells him he missed a supplier, he goes back to fill the gap. Until all the holes are plugged and no one can criticize it anymore, then he orders ✅
He has said himself that ChatGPT won't argue with you. If you feed your analysis to an AI, it will always agree and say it makes sense. Therefore, you must give it to real people.
$SIVE: The second verification of the same method
He has used this method more than once.
In 2025, he took a heavy position in $SIVE, a Swedish semiconductor company that makes lasers. Its market cap was over a billion dollars, priced in Swedish Krona, and basically not covered by US analysts.
Why did he target it? Because the next-generation optical communication architecture for data centers is called CPO. CPO has a physical flaw: silicon cannot emit light. How do you transmit data if you can't emit light? You must plug in an independent external laser module. $SIVE makes high-power continuous-wave lasers, which are the external light sources for CPO.
He walked through that chain again: NVIDIA GPU → Data center expansion → CPO optical interconnect → External light source demand explosion → $SIVE is one of the few global suppliers.
After buying in, this stock rose nearly twenty-fold.
RPI: The new demand Wall Street missed
Then there's Raspberry Pi, which he posted about in February 2026.
RPI is a British company that makes cheap microcomputers—a board sells for $35, used by kids to learn programming. Wall Street analysts' consensus: 14% annual revenue growth.
He wrote a different number: 55%.
How did he calculate it? He went through developer communities. A massive number of AI developers on GitHub started using Raspberry Pi to deploy AI Agents, and the growth curve for related repositories was almost vertical.
He tallied procurement discussions on various forums and developer growth trends, then reverse-engineered it. Wall Street's revenue model completely failed to account for this new demand, missing at least 40 percentage points.
Within two days of the tweet, RPI's stock price rose nearly 90%. Two months later, the financial report was released, showing actual growth of 58%. The Wall Street consensus had been 14%.
In all three cases—$AXTI, $SIVE, and RPI—the underlying logic is exactly the same.
Find a position on the industry chain where the pricing isn't yet in place, but the demand is already locked in.
The three most common traps for retail investors
Speaking of this, I want to talk about the three most common traps retail investors fall into and what problems this method actually solves.
Trap 1: Chasing highs and selling lows, always being the bag holder
Seeing a sector get hot, a stock rise, or the news talking about it, and then rushing in. As a result, it starts to fall right after you enter. You can't hold on, so you cut your losses. After you cut, it rises again, leaving you confused...
Why? Because by the time something is so hot that even someone like me, who doesn't research deeply, has heard of it, its pricing has long been fully baked in by the market.
First and second-layer companies like NVIDIA, Microsoft, and TSMC are being watched by analysts worldwide and bought by global capital. Why do I think I'm faster than everyone else?
Serenity's method is to dig down. Dig below the third layer, where analysts don't cover, institutions haven't built positions, and the market cap is too small or liquidity too poor for them to enter. In these places, there are plenty of pricing loopholes to catch.
This solves one problem: you don't have to compete on speed with the smartest people in the world; you can slowly flip through a track where no one is competing with you.
Trap 2: Buying in, then panicking when it falls and panicking when it rises
There is only one reason for panic: you don't know exactly how much the thing you bought is worth.
You might have bought it because it looked like it was going to rise. So, if it looks like it's going to fall, do you sell? Without your own anchor for judgment, your mindset will collapse as the market moves.
Serenity's method forces you to answer three questions before ordering: Is this segment irreplaceable? How many suppliers are there globally? Is downstream demand rising or falling?
Once these three questions are answered, you have a logical anchor independent of the stock price. You don't need to panic if it falls in the short term unless the answers to those three questions change.
This solves one problem: rising and falling don't depend on faith, but on the supply chain map in your hand.
Trap 3: Everyone is looking at the exact same information
PE, ROE, financial growth, northbound capital, top trader lists. You open any trading software, and the interface everyone sees is the same. This information is already fully reflected in the stock price.
What does Serenity look at? Patent databases, supplier directories, customs export data, developer discussion volume in industry forums, and technical route comparisons in academic papers.
These things are free and public, but most retail investors have never opened them in their lives.
This solves one problem: only when your information source is different from others can your judgment possibly be different from others.
There are also things this method cannot solve
Of course, this method is not a panacea; there are several things it cannot solve.
- First, Serenity himself has hit landmines. UPWK -35%, HIMS -50%, CRCL -45%. Having the right method doesn't mean every trade is right. No matter how accurately the supply chain is drawn, if company management has issues, the industry direction suddenly turns, or macro policies change, things can still go wrong.
- Second, he is anonymous. Former AI scientist, Nature paper author, turned down an NVIDIA offer—these titles are all self-described, and no one can verify them. He buys micro-cap stocks; he can pump a stock just by sending a tweet, and followers rushing in can push it further. He has never disclosed when he sells.
- Third, his entire portfolio is bet on two predictions: that CPO will become the sole technical route for data centers, and that humanoid robots will be deployed to the level of a billion units. If either is overturned, the logic for many of his picks collapses.
His method tells you that this position is a bottleneck, but whether the neck itself will always be there requires your own judgment.
What if we use it to look at Crypto?
I tried using this method to look at the Crypto sector.
Start deconstructing from Meme launchpads. Platforms like Pump.fun → what do they rely on? Market-making protocols → where does the liquidity for market-making protocols come from? Keep tracing down.
A summary in plain English of what this method can give you: when you get used to digging down from the top, you won't just rush in when you see something that has risen. You will instinctively ask: what layer is it on? Who is its upstream? Who is the upstream of the upstream? Where is that segment that only one or two companies are doing and that everyone else cannot bypass?
I thought about it for a long time and reviewed it for a long time, and I suddenly felt enlightened.
It's not that I found some wealth code, but that I suddenly realized many people's every trade, whether in stocks or crypto, is entirely spinning around the first layer of the industry chain. They buy whatever is hot, and all their decision-making radii do not exceed the watchlist page of their trading software.
Therefore, the strategy of the "Stock God" Serenity is not to tell you directly what to buy, let alone to tell you to follow him in FOMO.
What we need to do is something else: pull yourself out of the first layer where everyone is crowded together and let yourself see that the pricing of the fifth layer has not yet been discovered by the market.
