This feels like a good time to remind everyone that none of this is financial advice.

Subscribe now

I did eventually buy some shares myself, then sold a couple of months later at a slight loss because I wanted fewer positions and more money in my best ideas. I’d love to tell you I saw this coming, but I just got lucky with the timing.

Anyway, I’ve kept following VNET, and the slightly awkward thing about revisiting it is that I still like the business. It builds and operates data centers in China, providing the space, power, cooling and connectivity that other companies need to run their servers.

I still think China is going to need a lot more compute capacity. VNET is winning orders, its wholesale business is growing, and customers are filling the data centers. There is quite a bit here that has gone the way we hoped.

So why has owning the stock been such a miserable experience?

Well, building all of this costs an absurd amount of money, and I’m increasingly focused on how much of the eventual upside actually reaches shareholders. You can be right about the demand and still underestimate what it costs to serve it.

I still think there’s a case for owning VNET. I just think this one requires a bit more work than I originally thought.

Demand still looks pretty good

The wholesale business did pretty much what we wanted it to do.

Wholesale revenue grew 58.1% in the first quarter and another 29.3% in the second. Adjusted EBITDA grew 30.6% and 25.4%, respectively. Wholesale also passed retail revenue for the first time, while Q2 adjusted EBITDA margin reached 33.0%, up from 30.1% last year.

Taken at face value, those are very decent numbers… The whole point of the original thesis was that wholesale would become large enough to change the earnings profile of the company. That is happening.

The order book is even crazier. VNET signed 862MW of wholesale orders in the first half and disclosed another 355MW of reservations. Add those together and VNET booked or reserved more capacity in six months than it currently has in service. Is demand the problem here? I really don’t see how.

But I am now more focused on the gap between committed and utilized capacity. At the end of June, VNET had 1,007MW in service and 970MW committed, but only 744MW utilized. That leaves roughly 226MW already built and committed where the customer has not fully moved in yet.

This is where I struggle a bit... Committed capacity is great. Reservations are great. Giant orders are obviously great. But VNET already spent the money to build a lot of this capacity, and I want to see customers actually move in and start paying for it.

Mature wholesale facilities were 92.5% utilized, while the newer ramp-up capacity was only 36.6%. That tells me most of the low utilization is coming from recently delivered projects. It also explains why Q2 gross margin fell from 22.5% to 18.2% even as adjusted EBITDA grew 25%. Depreciation starts when the facility is ready. Revenue shows up as the customer installs equipment.

Could move-ins take longer than expected? Sure. Could some reservations fail to convert? Also possible. I just don’t see much evidence that customers are disappearing. VNET already won the damn orders. Now I want those orders to turn into revenue and cash, and sooner rather than later.

I’m offering 30% off annual subscriptions for the first year. That includes the rest of this VNET update, the paid archive, and all new paid posts throughout the year.

Get 30% off your first year

Read more

The stock is trading at roughly $36 at the time of writing. I’m used to seeing it somewhere around $25. That’s basically where Dropbox has traded for most of the past five years.

I have followed the company on and off since 2020 and owned it twice during that stretch. Both times I sold because I found something I liked better. Nothing really broke at Dropbox. It was just easy to move on from a stock that never seemed to do much.

I last wrote about Dropbox in October 2025. Growth had basically stopped by then, but customers were sticking around and the company was still generating a lot of cash.

The thesis was pretty boring... Dropbox had a sticky product, almost no growth, strong margins, a lot of free cash flow, and management was using a ridiculous amount of that cash to buy back stock.

The buyback was really the whole reason I liked it. Dropbox didn’t really need to grow all that much if the core business stayed stable and management kept making the share count disappear. It was becoming a genuine share cannibal, and the stock sitting in the same range year after year was actually helping.

At the end of that post, I said I wanted to see a few more quarters of flat revenue, strong cash flow, and continued buybacks before starting a small position.

Dropbox gave me almost exactly that. I still never bought it.

Now the stock is near the highest level it has traded in years, so I wanted to go back through the original thesis and see what actually happened.

So did the thesis work?

For the most part, yes.

The core business looks pretty much the same as it did when I last wrote about it. Revenue has stayed around $2.5 billion and the business excluding FormSwift has returned to slight growth. Dropbox has now added paying users for three straight quarters, while average revenue per user is still sitting around $140.

That is really all I wanted to see. I do not care whether core revenue grows 1% or 2% in a given quarter. The question was whether Dropbox could hold onto its users and pricing while Microsoft, Google, and Apple continued bundling storage into their much larger ecosystems. So far, it has.

There are now 18.19 million paying users, up 96,000 sequentially in the latest quarter. ARPU was $139.68, compared with $138.32 a year ago. Management expects some pressure as its cheaper Simple plan becomes a larger part of the mix, which is fine if that plan attracts and retains more users. I would only become concerned if the user growth is coming entirely from cheaper plans while the higher-value customers keep leaving.

Cash flow has also held up. Dropbox generated a little over $1 billion in unlevered free cash flow last year and now expects at least $1.07 billion in 2026. The company is still producing roughly $1 billion of cash from a revenue base that barely grows.

More importantly, it is turning that cash into much better results on a per-share basis.

Dropbox repurchased 60.4 million shares in 2025 and another 26.9 million during the first half of this year. Diluted shares were down roughly 18% year over year in Q2, which helped drive a 25% increase in unlevered free cash flow per share.

That is basically the entire thesis.

Dropbox does not need to grow much when every remaining share represents a meaningfully larger piece of the business each year. I originally figured the company might reduce its share count by 6% to 8% annually. It has been moving much faster than that. The diluted share count has fallen from roughly 419 million in 2020 to about 227 million in the latest quarter.

That is nearly half the company gone in six years.

The board also added another $900 million to the buyback authorization in June, leaving about $1.4 billion available at the end of Q2. Against a market cap of roughly $8 billion, that is still a massive amount of firepower.

Read more

I spent most of Friday and Saturday morning writing this with no idea why it sold off. Then Saturday afternoon word started going around that Alibaba is raising HK$80 billion, roughly $10.2 billion, through a share placement to fund the AI buildout. I don’t know if that leaked Friday. It would explain a lot if it did.

The dilution isn’t ideal, but the raise itself I’m fine with. I trust this management team more than I did a year ago, and given what cloud is doing I understand why they want to keep spending. Would’ve been nice to know before I added on Friday, but I probably would’ve bought anyway.

Quick background for anyone who hasn’t been following the name closely. Alibaba is still an enormous Chinese e-commerce business that barely grows, and for the past year it’s been eating the cost of an expensive instant-delivery fight with Meituan that wrecked group margins and did most of the damage to earnings. Those losses are coming down quickly now.

Cloud has been going the other direction the whole time. Nine straight quarters of acceleration, and this was the first one where the margin finally followed.

The quarter itself was good, and I liked the business more after reading the report than I did going in. Almost all of that is cloud, so that’s where I’ll start.

Cloud

AI Cloud and Compute Services revenue came in at RMB 48.4 billion, up 45%. Total segment revenue and external customer revenue both grew 45%.

That 45% was also pretty close to what analysts were already using. Alibaba does pre-earnings calls with the sell side, and the previews I saw in early July were around 45% growth with the segment margin moving from roughly 9% to 11%. Alibaba printed 45% and 11.6%. I thought it was a great quarter, but the headline numbers themselves were basically in line.

Still, this was the ninth straight quarter of acceleration, from a 3% decline in the March 2023 quarter to 45% growth now. Eddie Wu said it was the fastest cloud growth in 22 quarters, which takes you back to late 2020.

AI product revenue was RMB 12.4 billion, the 12th straight quarter of triple-digit growth. That annualizes to RMB 49.5 billion, or about $7.3 billion, and it is now 35% of external cloud revenue. A year ago it was a little over 20%.

Management thinks that annualized number will be close to $10 billion next quarter. It also said the model and application services business, including model-as-a-service, had crossed RMB 16 billion of ARR in August. There is some overlap in those figures, so I wouldn’t add them together. AI is 35% of external cloud revenue now and is moving the whole segment.

I’ve written about the revenue acceleration for long enough. The margin was the number I cared most about this quarter.

Read more

Baidu reports Tuesday morning, Kingsoft Cloud follows Wednesday, and Alibaba reports Thursday. PDD is currently expected next Monday, although the company still has not officially confirmed the date, which is very helpful.

Here is the order:

• Tuesday: Baidu ($BIDU)

• Wednesday: Kingsoft Cloud ($KC)

• Thursday: Alibaba ($BABA)

• Monday, August 24: PDD ($PDD), expected but not officially confirmed

Each report gives us a slightly different look at what is happening in China tech. Baidu and Kingsoft Cloud should tell us more about AI and cloud demand. Alibaba gives us cloud, e-commerce and quick commerce in one report. PDD should help fill in the remaining pieces around domestic e-commerce market share.

I cover Baidu and Alibaba much more closely, so that is where most of my attention will be.

We already got a few early clues from Tencent and JD last week.

Tencent’s advertising revenue grew 22%, AI demand remained strong and capex jumped again. JD’s report may have given us an even more useful signal for Alibaba.

JD is still the third and smallest player in quick commerce behind Meituan ($MPNGY) and Alibaba ($BABA), so I do not want to read too much into one quarter. But profit jumped and food delivery losses narrowed quite a bit, which suggests the price war across e-commerce and quick commerce has cooled.

That lines up with Alibaba’s pre-earnings guidance. Management has already told the Street that quick commerce losses are falling faster than expected and e-commerce margins are improving.

Put those together and the early read is that the price war has finally started to cool quite a bit.

That should help margins across the entire e-commerce and quick commerce market, including Alibaba, Meituan and JD. Alibaba may have the most room to improve given how much it has been spending, but the signal is good for the whole group.

Baidu

Baidu reports Tuesday morning.

The main things I am watching are Cloud growth, guidance and margins. More importantly for the stock, I want details on the Kunlunxin chip spinoff.

Last quarter, Baidu’s AI-powered business generated RMB 13.6 billion in revenue, up 49%, and crossed 50% of General Business revenue for the first time.

AI Cloud Infrastructure grew 79%. GPU Cloud revenue grew 184%.

Those are massive numbers, although I obviously do not expect Cloud to grow 79% forever. What I want to hear Tuesday is whether demand remains strong, what the pipeline looks like and how management is thinking about growth through the rest of the year.

The margin commentary will be important too. Baidu is spending heavily to build capacity, so I want some evidence that the economics are improving as the business scales. Another big Cloud number would be great, but guidance and margins will probably tell us more than the reported beat or miss.

Meanwhile, online marketing revenue fell 22% last quarter.

Baidu does not need search advertising to suddenly return to strong growth. It does need the decline to become less ugly while AI takes over more of the business. Another 20% drop would leave Cloud doing even more of the work.

Then there is Kunlunxin.

I think this is probably the biggest piece for the stock right now. In January, Baidu officially announced⁠ that Kunlunxin had confidentially submitted an application for a proposed spinoff and separate listing in Hong Kong. Baidu expects Kunlunxin to remain a subsidiary after the listing.

Since then, The Information reported that Kunlunxin is targeting a $50 billion IPO valuation. Reuters could not independently verify that figure⁠, and Baidu has not confirmed it, so I would treat it as a target.

Baidu reportedly owns around 58% of Kunlunxin. The basic math would value that stake at roughly $29 billion if Kunlunxin actually lists at $50 billion. That calculation ignores dilution from the IPO and any holding-company discount, but it would still represent a huge portion of Baidu’s current market value.

For context, Kunlunxin’s December funding round valued the company at only about $3.1 billion. Going from $3 billion to $50 billion within a year would be fairly ridiculous, so I am not plugging the full $50 billion into my valuation and calling it a day.

The official filing also says that the offering size, structure and Baidu’s ownership after the deal have not been finalized. The listing still requires regulatory approvals, and Baidu has given no firm timetable.

That is why management’s commentary could be so important Tuesday. I want any update they can give on timing, the potential offering size and how much of Kunlunxin Baidu expects to own afterward. Even if the final valuation comes in well below $50 billion, a completed listing would finally give the market a cleaner way to value an asset that is mostly buried inside Baidu today.

Kingsoft Cloud

Kingsoft Cloud reports Wednesday morning.

I do not follow the company closely enough to pretend I have some big opinion on the stock, but the report should give us another useful look at AI infrastructure demand in China.

Q1 revenue grew 37%, public cloud revenue increased 48%, and gross billings from AI customers grew 90%. AI represented more than half of public cloud revenue for the first time.

The weak part was gross margin, which fell from 16.2% to 12.8% as server costs and investment increased.

Demand clearly has not been the problem. I want to see whether margins begin stabilizing as all of this new capacity comes online.

For me, Kingsoft Cloud is mostly a read-through for Baidu and Alibaba. Strong AI demand would be another sign that China’s cloud buildout still has plenty of room to run.

Alibaba

Alibaba reports Thursday morning, and this is the one I care about most.

The two biggest numbers for me are Cloud growth and Cloud adjusted EBITA margin.

I think Cloud revenue grows more than 45%. Parts of the sell side are already around the same number, so this is no longer some wild estimate that nobody expects.

Cloud growth has accelerated from 29% to 34%, then 36%, and most recently 38%. AI-related products accounted for 30% of external Cloud revenue last quarter, and management expects that figure to exceed 50% within roughly a year.

I think the bigger surprise could come from margins.

Alibaba Cloud’s adjusted EBITA margin has been stuck in the high single digits while the company invests aggressively in capacity. Last quarter, it was around 9%.

Management said margins should improve over the next one or two quarters. I think we could see Cloud EBITA margin move back into double digits for the first time in years.

That would be huge.

Cloud has been growing quickly for a while, but the market has been reluctant to give Alibaba much credit because margins were barely moving. That concern has been fair. Revenue growth only goes so far if the company has to consume every dollar building more capacity.

A 45%+ Cloud print with double-digit EBITA margins would show that some operating leverage is finally showing up. I think that combination could force investors to value the Cloud business very differently.

If Cloud grows quickly and margins remain stuck around 9%, I will want to understand why. This is the quarter where we should begin seeing some of the improvement management has been talking about.

Quick commerce is the other big piece.

I wrote in July that management’s pre-earnings calls were noticeably more positive than the calls held before the previous quarter. Quick commerce losses were apparently falling faster than expected, and management wanted the Street to understand that the June quarter should look better than feared.

Alibaba gave analysts the opposite message back in January. Management told them to lower estimates across Cloud, e-commerce and margins, and the quarter eventually came in pretty close to those warnings.

So I take the change in tone seriously.

The core e-commerce business still needs to hold up, but I think the stock reaction will mostly come down to Cloud growth, Cloud margins and how quickly quick commerce losses are falling.

Expectations have moved higher. The stock has bounced sharply from its June lows, 45% Cloud growth is now being discussed openly and investors already know quick commerce is improving.

A 40% Cloud print would normally look great. This time it could be viewed as a disappointment.

But if Alibaba reports 45%+ Cloud growth, moves Cloud EBITA margins into double digits and shows a real reduction in quick commerce losses, I think the stock goes quite a bit higher.

I am very bullish heading into the print.

PDD

PDD is currently expected to report next Monday, although we are still waiting on an official date.

The main thing I want from this report is a better idea of where e-commerce market share sits between PDD, JD and Alibaba.

All three companies have been spending heavily. Alibaba attacked quick commerce, JD moved into food delivery and PDD is increasing investment in its supply chain and first-party brand business.

Last quarter, PDD’s revenue grew 11% and transaction services revenue increased 20%, but net income fell 15%.

Management made it clear that investment will remain high, even if near-term profits take a hit.

PDD does not make market share easy to calculate, so we will have to piece it together using its results alongside what JD and Alibaba report. If PDD’s domestic growth slows while Alibaba’s GMV and customer management revenue improve, that would be a pretty good sign that share is moving.

If all three report strong domestic demand, then maybe the Chinese consumer is doing better than people think.

Temu will get plenty of attention too, especially around tariffs and its international expansion. For me, the more useful part will be what PDD tells us about competition inside China.

By the time PDD reports, we should have a much cleaner picture of Cloud demand, AI spending and e-commerce market share across China.

For Alibaba specifically, this could be the quarter where Cloud finally earns a different valuation. Margin expansion has been the missing proof point for me and probably for the market too.

We will find out Thursday.

Disc. Long &

Disclaimer: This post is for informational purposes only and should not be considered financial advice. I am not a financial advisor. Please do your own research and make your own investment decisions.

The original thesis was pretty straightforward: a payments business growing around 30% a year, showing real operating leverage, with a roadmap to roughly $1 billion in free cash flow by 2027 and a valuation that looked way too cheap if management could get anywhere close.

A lot has happened since then.

Global Blue has become a much bigger part of the story, the stock has gotten smoked, the old $1 billion FCF target disappeared, management has had some pretty obvious issues communicating guidance, and the company has since started shrinking the share count pretty aggressively.

If you’re new to the name, my original writeup is probably still the best place to start.

Anyway, Q2 gave us another pretty good example of why I’m still here.

The actual quarter was strong. Really strong, honestly.

Volume came in at $61 billion, up 22%. Gross revenue less network fees was $624 million, up 51%, or 11% organically. Payments-based GRLNF grew 27%. Adjusted EBITDA was $284 million, up 39%, with margins improving to 46%. Non-GAAP EPS came in at $1.32.

And they beat the quarterly guide they gave us in May.

Management had been looking for roughly $615 million of GRLNF, $278 million of adjusted EBITDA and only about $10 million of adjusted free cash flow. They finished at $624 million, $284 million and $21 million respectively.

So operationally, I really don’t have much to bitch about.

The international payments business continues to fly. Payments-based GRLNF outside the Americas is up 52% YTD versus the high-20s growth management expected coming into the year. The Americas business is up 17%, basically right where it should be, and Shift4 One is now live in 12 European countries with management still targeting at least 15 by year-end.

The Global Blue integration also looks fine from what I can tell.

Remember, this was one of the big concerns when I wrote about FOUR earlier this year. Global Blue was going to temporarily hurt EBITDA margins and free cash flow conversion while Shift4 integrated the business and started pushing payments through the network.

That was expected.

What I wanted to see was the underlying payments business continue growing while the integration worked its way through the numbers.

So far, that’s pretty much what is happening.

What’s Up With the Guide?

Shift4 lowered full-year GRLNF guidance from $2.50-$2.60 billion to $2.48-$2.53 billion.

Adjusted EBITDA moved from $1.165-$1.215 billion to $1.15-$1.18 billion.

Adjusted free cash flow went from $490-$510 million to $465-$475 million, and EPS moved from $5.50-$5.70 to $5.15-$5.35. Volume guidance was unchanged.

Obviously, cutting guidance isn’t good.

But I think the reason behind the cut is worth thinking through.

Management said roughly $25 million of the GRLNF reduction comes from assuming continued Middle East-related travel disruption in Q3, with another $20 million from FX translation. The updated FCF and EPS numbers also include the impact from the company’s recent financing.

The strange part is that the Middle East headwind was actually a little better than expected in Q2.

Those aren’t my words either. Taylor Lauber specifically said international performance was better than they originally anticipated during the quarter, while simultaneously lowering the full-year outlook because they are assuming the travel disruption continues into Q3.

“The Middle East conflict remained a headwind and weighed on inbound travel to Europe and across several Gulf countries. However, the overall impact on our Q2 results was slightly better than we had forecast.”

That’s where I’m struggling a little.

If Q2 was better than expected and the worst of the travel disruption appears to have been during the quarter, why assume a fairly meaningful hit in Q3?

Maybe they know something I don’t.

Or maybe management finally learned how to guide.

Back in March, I spent a good chunk of my last Shift4 post complaining about exactly this.

The business kept executing, but management had developed a pretty impressive ability to shoot itself in the foot every time it talked about the future. The $1 billion FCF target disappeared, EPS guidance came in well below the Street, the messaging around Global Blue was clumsy and investors stopped trusting the numbers.

I wrote at the time that they needed to learn the oldest trick in the book: underpromise and overdeliver.

Well… maybe this is them doing it?

I may be giving management too much credit. I own a lot of the stock and I’m obviously looking at this through that lens.

But I’d like to think Taylor and the new team learned something from watching the stock get repeatedly murdered over aggressive targets and messy communication.

Read more

Well, it was.

The margin recovery will take longer than I expected. But the customer behavior looks much healthier than the headline numbers suggest, and I can live with the first if the second keeps holding.

So yeah, there’s plenty here to complain about.

But I actually came away from the quarter feeling better about the part of the thesis I care about most.

The breach happened last November, and Q2 was really the first quarter where most of the fallout showed up at once. The fine was booked. Marketing spend stayed elevated as they tried to win customers back. The logistics network was built for demand that temporarily disappeared. And you still have a chunk of customers who stopped buying after the breach and haven’t returned.

Against all of that, the underlying business held up better than I expected.

Revenue

Revenue was $8.86 billion, up 4%, which came in about $170 million light of consensus. That number is close to useless.

The won hit its weakest level against the dollar in more than fifteen years during the quarter. Translating Korean revenue back into dollars took $548 million off the top line. In constant currency the business grew 10%. Product commerce grew 8% in constant currency, up from 5% in Q1.

Nothing to do about that. I own a Korean business that reports in dollars, so some quarters the currency helps and some quarters it doesn’t. What I care about is whether Koreans are buying more stuff from Coupang, and in local terms they are, at a faster clip than three months ago.

Gross profit was $2.49 billion, down 3% reported and up 3% in constant currency, with margin down 188 basis points to 28.2%.

Customers

The thing that actually got my attention was buried in Bom Kim’s prepared remarks.

If you strip out the specific group of customers who stopped buying during the breach and still haven’t come back, spending across everyone else grew about ~16% year over year. That is roughly what product commerce was compounding at before any of this happened.

“The vast majority of our customer spend never moved. That group is spending at the highest levels in our history and compounding similarly to before last year’s data incident.” — Bom Kim

He also put a number on the older cohorts that I hadn’t seen before. They now spend nearly ten times what they spent in their first year, and they’re still increasing. That’s the part of this business that never shows up in a quarterly print and is most of the reason to own it.

The customers who left and came back apparently return at their full prior spend and grow from there. New customers are signing up faster than they were before the incident. WOW membership is above where it was pre-breach with additions accelerating, though new members start at the bottom of the spend curve, so that shows up in revenue later rather than now.

I’m aware of what this metric is. When a company hands you a growth number that excludes the customers who stopped buying, you hold it at arm’s length. Every business on earth looks healthy once you remove the people who left.

But it does answer the one question I actually care about, which is whether the breach broke anything. If the base that stayed were spending less, or ordering less often, or if WOW were leaking members, that’s where it would show up first. Instead the people who stayed are spending more than they ever have, and the ones who left return at full strength whenever they decide to return.

To me, the missing customers still look angry about the breach. The behavior of the ones who returned doesn’t suggest they found a better service somewhere else. Coupang still has work to do rebuilding trust, but I don’t see much evidence that its actual customer proposition got worse.

Margins

Product commerce adjusted EBITDA margin was 5.1%, down 390 basis points year over year. Segment EBITDA fell $281 million to $382 million. Consolidated margin was 1.8%.

Management points at three things. Supply chain got dislocated when demand dropped. Marketing is elevated because they’re paying to reacquire customers. And the fixed cost base was sized for a demand curve that stopped existing in November, so they’re running a network with slack in it.

That last one was a choice.

“We could cut them significantly, we’ve chosen not to because the right long-term decision is to grow into the capacity and support our customer experience that has always been our North Star.” — Bom Kim

Read more

My last portfolio update was back in February, and the portfolio looks very different today. Some positions have gotten much larger, several names are gone, and a few new ones have worked their way in.

Performance has also been a bit of a roller coaster.

For most of the year, returns were honestly pretty subpar. I wasn’t happy with how the portfolio was performing, and up until this past month there wasn’t much worth bragging about.

A few timely call positions in Alibaba and KWEB changed that pretty quickly. Those trades have worked extremely well and gave the portfolio a meaningful boost over the last several weeks.

As of July 31, the portfolio is up 15.84% YTD versus 10.13% for SPY. Since inception, the portfolio is up 17.92% compared to 8.79% for SPY.

That looks good today, but I’m not going to pretend it has been smooth. A large portion of the outperformance came recently, and it came from taking more risk than I normally would through those options positions.

Below is the full portfolio, along with a quick look at what changed, what drove the recent rebound, and where I currently see the biggest opportunities.

Read more

I opened that post complaining the stock had run 40% in the two weeks it took me to write it.

This one’s been worse. I drafted most of it Tuesday with the stock at $72. Wednesday it closed at $65.30. Thursday it closed at $78.65 and traded up to $81.74 after hours. I’ve rebuilt the math three times in four days and at this point I’ve just accepted that some of these numbers are going to be wrong by the time you read them. Sorry in advance.

Anyways…

Stock closed June at $126.89. Bottomed Wednesday at $65.30, which is about a ~50% drawdown. Back in the high $70s now. Still roughly ~40% off the high, still up roughly ~100% on the year.

What happened

Short version, none of it was stock specific.

The AI trade came into question this month. Nvidia laid out AI infrastructure commitments north of $750 billion and instead of taking that as good news everyone started asking whether the demand behind it is real or just a bunch of companies financing each other in a circle. Everything with a chip in it started coming down.

Korea has been the center of gravity for all of it. The KOSPI’s been getting taken apart all month, SK Hynix and Samsung are the epicenter, and because Seoul trades hours ahead of New York it’s basically been setting the tone for every semi open over here. SK Hynix printed a record quarter and got sold anyway. When Korea has a bad night, everything with a chip in it opens red here.

Then it stopped being a normal selloff and turned into something much dumber.

Situational Awareness, lol

If you haven’t been following this one, buckle up, because it’s genuinely one of the great stories.

Leopold Aschenbrenner. 25 years old. Columbia valedictorian at 19, joined OpenAI, got fired in 2024 over what OpenAI called improper disclosure of internal information, wrote a 165-page essay about AGI that everyone in the Valley claimed to have read, and then started a hedge fund named after his own essay.

No prior trading experience. Zero.

The fund went from $225 million to as much as $45 billion in under two years and returned 439% through June. Long AI infrastructure, SK Hynix and CoreWeave and Bloom Energy and some crypto miners, short software. Reportedly levered up to 400%.

Then July showed up and took 67% of it.

On July 24 he wrote to his investors telling them the selloff was one of the best buying opportunities since early last year and inviting them to commit fresh capital starting August 1. Six days later he was liquidating the entire public book to Citadel at a discount. Fund’s down to something like $10 billion and most of what’s left is a private stake in Anthropic.

And the thing is, he wasn’t even wrong…

The AI infrastructure thesis has been right for two years. He saw the compute buildout coming earlier than almost anybody and made 439% being right about it. He’s probably still right, honestly. Nothing that happened in July says the buildout isn’t real.

He just borrowed so much money against being right that he didn’t get to stay in the trade long enough to collect on it. At 4x levered you don’t get to have an opinion during a drawdown. The margin clerk has the opinion. Once the prime brokers start calling it doesn’t matter what you think SK Hynix is worth in 2028, you’re selling it Thursday at whatever Citadel feels like paying that day.

Being right is only half of it. The other half is being able to sit there while you wait to be proven right, and leverage takes that part out of your hands.

There’s also a very good conspiracy floating around that Citadel Securities put out a note days before the Fed meeting arguing for a surprise rate hike nobody was expecting, the AI trade cracked, Leopold got margin called, and then Citadel bought his whole book at the lows.

The note is real, for what it’s worth. The rest is fintwit doing what fintwit does, and technically Citadel Securities and Citadel the hedge fund are separate entities. Doesn’t matter. Great meme.

The actual point, and the reason I’m spending this much space on a hedge fund that has nothing to do with ACM: what started as a correction turned into a deleveraging spiral. One of the biggest forced sellers in the market was dumping a levered book into a falling tape and every correlated thing on the screen went down with it. Then Citadel took the book off his hands and the same names ripped. Market’s up two days running now.

That’s it. That’s what happened to this stock.

ACM Research has not said a word since July 15, and that was a press release announcing the date of the Q2 call.

And they sold it on the bull case

Here’s the part I actually can’t get past.

Reporting on the memory rout was pretty specific about the trigger. The selling got set off by fears that China’s CXMT would accelerate its capacity expansion after its IPO, plus reports of Chinese progress on domestic DUV lithography.

Read that again.

People sold semis because a Chinese chipmaker is about to spend a pile of fresh IPO money building more capacity. And ACM went down with them.

ACM sells the tools CXMT buys.

Same thing with the lithography story. If China cracks DUV, the biggest bottleneck on domestic fab construction gets smaller and more fabs get built. ASML dropping on that makes sense, that’s ASML’s revenue at risk. ACM sits on the other end and went down with the group anyway, because it’s got a semi equipment label on it and in a week like that one nobody’s reading past the label.

The two things that took this from $127 to $65 were, on the merits, reasons to own it.

I’ve been adding. Full case below.

Read more

I just thought it would be useful, and honestly kind of fun, to think about what this business might trade at if it were based in the U.S. instead of China.

There isn’t a perfect comparison because Tencent is a weirdly broad company. It’s part social network, part gaming business, part payments network, part cloud provider and part investment company. There really isn’t another business quite like it.

Still, lining it up next to Meta, Google, Apple and a couple of the large gaming companies gives us a decent idea of how differently the market treats Tencent.

Meta is growing much faster than Tencent. Revenue is up 26% over the past twelve months and has compounded at nearly 20% annually over three years. It’s obviously a great business and probably deserves to trade at a slight premium, especially given that it’s U.S.-listed.

But is Meta really that much more dominant of a business than Tencent? I don’t think that’s obvious at all.

Google is growing faster too. Revenue is up 20% over the past year, helped by Cloud going absolutely nuts, and the business has compounded revenue at 12.5% annually over three years.

Tencent is at 10.7%.

That is a difference, but it’s not a massive one. Google trades at 24x forward earnings while Tencent trades at just 12.6x.

Google deserves the higher valuation. I just struggle to believe the underlying businesses are so different that one dollar of Google earnings should be worth nearly twice as much.

Then there is Apple.

Tencent and Apple grew revenue at almost the exact same rate over the past twelve months. Tencent grew 12.84%. Apple grew 12.76%.

Over the last three years, Tencent has compounded revenue at nearly 11% annually. Apple is below 2%.

Apple trades at a whopping 36.5x forward earnings.

Obviously Apple deserves a premium. It has one of the greatest consumer ecosystems ever built, an enormous global brand and a customer base that will happily spend $1,200 replacing a phone that still works perfectly fine.

I own no issue with Apple trading well above Tencent.

Almost three times the multiple is a lot, though…

Some of that gap comes from the premium the U.S. market awards its best technology companies. Some comes from the discount attached to anything Chinese. Put those two together and you end up with a pretty absurd spread.

Then there’s gaming

EA and Take-Two are not perfect comparisons either. Gaming is essentially the entire business for both of them. For Tencent, gaming is one piece of a much larger company.

Still, Tencent owns the largest gaming business in the world, so they’re worth including.

EA has barely grown revenue over the past three years and trades at 23.6x forward earnings.

Take-Two has grown faster recently, mostly as investors look forward to Grand Theft Auto VI and the earnings the company should produce once that arrives. The stock trades at 33.7x forward earnings.

I understand the argument for both.

What feels odd is that investors are willing to look several years into the future for Take-Two while giving Tencent very little credit for what it already owns and earns today.

Tencent already has an enormous and highly profitable gaming business. Honour of Kings, Peacekeeper Elite, League of Legends, Clash of Clans, Brawl Stars, Delta Force and a pile of other franchises spread across China and the rest of the world.

Then, on top of that, you get WeChat, advertising, cloud, payments, fintech and the investment portfolio.

The entire company trades at roughly half EA’s multiple and a little over one-third of Take-Two’s.

China deserves a discount. I don’t think it explains all of that.

The other thing that’s hard to capture in a comparison like this is just how good the business actually is.

Revenue growth and earnings multiples are useful, but they leave out the thing I care about most.

Tencent is a superb business.

WeChat is probably the best example of a true super app anywhere in the world. More than a billion people use it daily for messaging, payments, shopping, food delivery, transportation, entertainment, business and basically everything else.

I would put the moat around WeChat up against almost anything.

People are not leaving. Merchants are not leaving. The mini-program ecosystem is already there. The payments network is already there. Businesses have spent years building their operations around it.

I don’t know what would realistically disrupt that position at this point…

A better messaging app would not be enough. A competitor would need to recreate the social graph, payments layer, merchant network, mini-program ecosystem and all the little habits that have built up over more than a decade.

Good luck lol

That is what makes the valuation so hard for me to understand. Tencent is literally one of the best businesses in the world. It owns one of the most dominant consumer platforms ever created.

And then you have to consider management.

Pony Ma founded Tencent in 1998 and is still running it. Over that period, the company has built WeChat, become the largest gaming company in the world, created a giant payments and fintech business, invested in many of China’s best internet companies and returned enormous amounts of capital to shareholders.

Tencent’s management team belongs in the same conversation as the best teams running any major technology company.

They have repeatedly shown that they know when to invest, when to pull back, when to monetize outside holdings and when to buy their own stock.

The record is not perfect. Nobody’s is. But over nearly three decades, very few management teams have allocated capital this well across this many different businesses.

That deserves a premium too, or at least it should prevent the stock from trading like a questionable melting ice cube.

What would this trade at in the U.S.?

There is no way to know, obviously.

But I have a very hard time believing an American Tencent would trade anywhere close to its current valuation.

Imagine the pitch.

You have one of the most dominant consumer platforms in the world. More than a billion daily users. The largest gaming business. A growing advertising operation. Cloud, payments and fintech. A huge investment portfolio. Net cash. Double-digit revenue growth. Profit growing faster than revenue. A shrinking share count. One of the best management teams in technology.

Then add AI.

Tencent already has the users, merchants, advertisers, cloud customers, payments network and distribution. It does not need to build an audience for AI from scratch. It can push the technology into products people and businesses already use every day.

If that exact collection of assets were headquartered in California and listed on Nasdaq, investors would probably argue about whether it deserved 25x or 35x earnings.

I doubt anyone would suggest 12x was reasonable…

I’m not saying Tencent deserves Apple’s multiple. I’m not even sure it deserves Google’s.

But there is an enormous amount of room between 12.6x and the valuations the market gives these American companies.

At 16x earnings, Tencent would have roughly 27% upside from the multiple alone.

At 18x, the upside would be around 43%.

Even at 18x, Tencent would still trade below Google, Apple, EA and Take-Two. Investors would still be getting compensated for China. The valuation would just begin to resemble the quality of the business.

A Discount, Sure. This Much?

U.S. companies trade at a premium. They have deeper capital markets, more consistent foreign demand and far less political baggage.

Tencent operates in China, trades in Hong Kong and comes with risks that Meta, Google and Apple do not have. Some valuation gap is completely reasonable.

I just think the current gap has become far too wide.

Tencent is still growing. AI should create more opportunities across advertising, gaming, cloud and WeChat. The share count is coming down. The balance sheet is loaded. Management has proven itself over and over again.

Most importantly, the core business is about as difficult to disrupt as anything I have ever seen.

That combination should not trade like an average company.

Maybe Hong Kong leaves it cheap for another year. Wouldn’t exactly be shocking. But I have a hard time believing one of the best businesses and management teams in the world will permanently trade at such a massive discount to companies with similar, and in some cases worse, growth.

If Tencent traded in the U.S., the market would almost certainly value it like the superb business it is.

For now, we get the Hong Kong price.

I’m fine with that.

This article is for informational purposes only and does not constitute investment advice. I own shares of Tencent, and my own money is at risk. Always do your own work before making investment decisions.

Management had just put out 2026 guidance that came in well below what the Street expected. The old $1 billion free cash flow target disappeared from the presentation. Global Blue was dragging on margins and cash conversion. And after years of questionable communication, the market basically decided it was done giving this company the benefit of the doubt.

I understood the frustration. I was frustrated too.

Then the stock went on to fall another 20%.

FOUR eventually traded down to $34 in June before bouncing back above $50. From the low, the stock is now up nearly ~50% in a little over a month. After watching it get pummeled for most of the year, I will gladly take it.

But whenever a stock rallies this quickly, you start hearing the same question: did we miss it?

I don’t think so.

At $39, which is where I built the model below, Shift4 was trading at roughly 4.2x my estimate of 2027 free cash flow. At today’s price around $50, the multiple has increased to roughly 5.3x.

I want to be very clear about the denominator here because this is where the math can get sloppy.

The ~5x figure is not based on the old $1 billion free cash flow target management quietly removed earlier this year. I’m using approximately $740 million of adjusted free cash flow in 2027, with roughly 78 million diluted shares.

That works out to around $9.49 of free cash flow per share. The stock is trading at just over five times that number.

My estimate is actually more conservative than the $800 million to $900 million year-end run-rate I discussed in March. The $740 million represents free cash flow earned across the full year. Shift4 could exit 2027 at a higher run rate without producing $800 million or $900 million during the calendar year itself.

Either way, I do not need the abandoned $1 billion target for the valuation to work.

The Cash Flow Ramp

The whole thesis now comes down to whether cash conversion recovers as the Global Blue integration matures.

Shift4 generated approximately $500 million of adjusted free cash flow in 2025. I have them producing roughly the same amount in 2026, followed by $740 million in 2027, $920 million in 2028 and just over $1 billion in 2029.

That sounds like a large jump, and it is. But the assumptions underneath it aren’t especially wild.

For 2027, I’m estimating:

• $3.15 billion of gross revenue less network fees

• $1.56 billion of adjusted EBITDA

• A 49.5% adjusted EBITDA margin

• 47.4% free cash flow conversion

• $740 million of adjusted free cash flow

The conversion assumption is the most important piece. Shift4 converted roughly 51.5% of adjusted EBITDA into free cash flow in 2025. Management is guiding to only 42% in 2026 as the company works through Global Blue integration spending, heavier interest expense and some ugly working-capital timing.

My 2027 estimate assumes conversion recovers to 47.4%. That would still be below the 2025 level.

Read more

Management had been holding pre-earnings calls with the Street, and notes from those conversations started making the rounds. The message was that quick commerce losses are coming down much faster than people expected and that the June quarter, which Alibaba reports next month, should look better than feared.

That was basically it. The money-losing business is apparently losing less money.

Alibaba did the same thing back in January. Those calls leaked too, and I spent a weekend writing up what management was telling people.

Except in January, the message was the exact opposite. And the stock is now about a third lower than it was when those calls happened.

That’s the part worth sitting with for a minute because it has been a rough few months to own this thing. The stock went from $193 to $140, then slowly bled from around $112 into the mid-$90s during June. I’ve been long the whole way, written about it the whole way, and none of it was fun.

I also put out a post last week saying China tech had become one of the most ignored corners of the market because every dollar chasing AI was getting sucked into the semiconductor supply chain.

Then Alibaba ripped 48 hours later.

I’ll take it, but I’m not going to pretend I saw that coming.

January’s Message

The January message was pretty negative.

Take your cloud numbers down. December-quarter e-commerce was slowing materially and CMR kept decelerating. Quick commerce losses weren’t narrowing because Alibaba was still spending heavily to push its market share above 50%. The buyback was non existent, with management rightly prioritizing investment in AI and cloud growth instead, and there was no suggestion it would pick back up anytime soon.

Somebody even asked directly whether e-commerce growth could fall from 9% to 3% or 4%, and management wouldn’t answer.

I thought they were sandbagging. They mostly weren’t. The quarter came in around where management had pointed, and the stock still sold off.

The interesting part is that management clearly knows these calls move expectations. If they were worried about the quarter they’re about to report, I doubt they’d be going out of their way to tee up good news seven weeks early.

The details inside the notes are worth paying attention to, but the direction of the message probably tells you more. In January, management wanted estimates coming down. This time, they went out of their way to tell people things were improving.

The funny thing is, almost none of what came out of the call should have been a surprise.

Read more

This time it’s two names, calls on both, together a little over ~10% of the portfolio.

I genuinely can’t think of a more hated corner of the market right now. Actually, hated might be generous. Hated implies people are paying attention. These names are just ignored, for now.

Every dollar in the world is chasing the AI bottleneck trade, memory, power, anything with a supply constraint attached to it, and that trade is sucking capital out of everything else. This group sits on the wrong end of that rotation. Money has been walking out the door for months, and a buyer base that supported these names for years recently flipped to net selling for the first time in years.

The index that tracks this group is down more than ~30% from its highs and some of the biggest names in it just traded at their lowest levels since early 2025.

The funny part is that the two biggest names in this corner are about as clear AI beneficiaries as you'll find anywhere. Actual AI revenue growing fast, with real usage behind it, and the market refuses to pay for any of it for reasons that have nothing to do with the businesses themselves.

I've watched enough of these rotations to know how they usually end. The crowded trade gets more crowded, the ignored stuff gets cheaper, and it feels permanent the whole time. Then the valuation gap gets stupid enough that people start looking around. I think that point is very very close.

Now obviously “cheap” alone didn't get me back in the trade. It was cheap a few months ago too and the market did not care. What changed is the buying. There's a very specific buyer showing up in these names every single trading day right now, and the pace picked up noticeably in the last couple weeks.

I got burned last time, and part of that was on me. I didn't buy enough time, so when the trade needed a few more months to work, I didn't have them.

Though in fairness, I'm not sure anyone saw the kind of route these names went on from there coming…

Read more

I’ll be honest, my first reaction was relief. Which is a strange thing to feel when a company you own just caught the biggest fine of its kind ever handed out, but if you’ve followed this saga you probably felt it too. Since last fall, every conversation about Coupang has really been a conversation about the breach, and now that’s finally over. There’s a number, and the number is far better than feared.

The market seemed to feel the same way, because the stock rallied the day the news hit.

Remember what people were tossing around this winter? A possible business suspension, even a forced breakup, with fine estimates running into the billions. What actually showed up works out to about just ~1.4% of revenue. Coupang is obviously appealing, and either way it will land in the Q2 numbers, but this is a bill they pay without breaking stride.

If you're new here, fair warning: I've written about Coupang a lot. There's the original deep dive from last November, the buying opportunity post from January when the breach had everyone panicking, and the Q1 earnings review in May that dealt with all the voucher noise. At this point Coupang is less a stock I cover and more a recurring character on this newsletter. So I won't rehash any of it. The short version is the business came through fine, and if you want the full saga, it's all there.

What’s been bugging me lately is something else entirely.

The breach is mostly behind us now, but the stock still hasn’t really moved on from it. That would be annoying on its own. What makes it harder to dismiss is what’s happening around it: Korea has suddenly become one of the hottest markets in the world, and Coupang has basically sat out the entire move.

Maybe the market is right and Coupang doesn’t deserve to be part of the Korea trade. I get the objection. This isn’t SK Hynix. It doesn’t make memory chips, and nobody is buying Coupang because AI capex is exploding.

But that’s also why it’s interesting to me. If the AI trade is making Korean households richer, if brokerage accounts are up, bonuses are getting paid, and people are feeling a little more comfortable spending, then Coupang should eventually feel some of that?

Maybe not immediately. Maybe not in some clean one-for-one way. But it’s hard for me to believe the company showing up at everyone’s door every day gets zero benefit from a country suddenly having more money.

I publish this kind of thing most weeks. Subscribing is free, but paid unlocks the full archive of nearly 200 posts, plus every new writeup and research report going forward.

My recent Tencent writeup below is probably the best example of what paid readers are getting right now.

Subscribe now

Here’s the past twelve months performance for context. Korean market (KOSPI) in green, Coupang in blue…

At first, the underperformance mostly just annoyed me because I own the stock and, well, that’s how underperformance works. But now I’m less annoyed and more curious. Coupang is still near the bottom of its own valuation range while the country around it is having one of the strangest/largest wealth booms ever.

I know that sounds dramatic, but I don’t really think it is…

The KOSPI has roughly doubled this year alone, its best half-year on record, on top of a ~75% gain in 2025 that was already its best full year since 1999. SK Hynix is up something like ~340% and just passed Samsung as the most valuable company in Korea for the first time since 2000, a ranking Samsung is disputing, which tells you how close it is. Samsung itself is up around ~200%, and the two of them together now make up more than half of the entire index.

The whole country is levered to AI and the trade is working.

All of that money is going somewhere, and where it’s going is Korean brokerage accounts. Stock ownership in Korea went from about 6 million people in 2019 to over 14.5 million now, and JPMorgan estimates household stock wealth has grown by more than 1,000 trillion won this year. That’s a quadrillion won of gains in a country of 51 million people.

Then there’s the part that almost sounds made up. SK Hynix agreed to pay employees ~10% of operating profit as bonuses with no cap, which on this year’s expected profits comes out to roughly 700 million won per employee, about $460,000 on average.

Samsung’s chip workers spent months threatening strikes until they got a deal of their own, an average payout somewhere around $340,000.

It’s reached the point where the Korean press reports SK Hynix engineers have become the most sought-after bachelors in the country’s matchmaking market, suddenly ranked alongside doctors and lawyers, and the Bank of Korea is warning about bonus checks stoking inflation because card spending near the chip campuses in Gyeonggi is running visibly hotter than the rest of the country.

And apparently the matchmaking is working, because Korea is having babies again.

And again, this is Korea. Basically the global example everyone uses when they wants to talk about collapsing birth rates and long-term consumption problems. Births have now risen year over year for 21 straight months. March was up 19.4%, the biggest jump in over three decades, and the monthly fertility rate touched 0.99 in January, nearly back to a whole baby per woman!

Marriages are also at their highest level since 2018, and in Korea marriage is basically a two-to-three year leading indicator of births, so this probably isn’t done.

To be fair, the rebound started in mid-2024, before the rally, and demographers mostly chalk it up to post-COVID catch-up weddings and a big cohort hitting their early thirties, not the KOSPI. Fine. But a country full of newly minted millionaires can’t be bad for weddings, and every wedding and baby is a future Coupang household.

Overnight diapers were literally the original Rocket delivery pitch. The oldest knock on Coupang has always been that it’s the best retailer in a shrinking country, and for the first time in nearly a decade that curve is bending the other way.

So you’ve got millions of newly flush households, tens of thousands of workers getting life-changing bonuses, and now more weddings and babies on top of it. Somebody is going to sell those people more stuff.

My money, literally, is on the company that already sells Korea most of its stuff.

Now the pushback, because there’s a real one. Koreans historically don’t spend stock gains the way Americans do. The Bank of Korea’s own research says only about 1.3% of equity gains become consumption there, versus 3 to 4% in the US, because Korean wealth traditionally gets parked in apartments rather than spent at the mall.

And the spending that has shown up so far skews rich: department stores up 15%, luxury watches flying, jewelry up triple digits at stores near the chip campuses. The old knock, and I’ve leaned on it myself, is that Coupang sells diapers and bottled water, not Rolexes, so a luxury-led boom mostly passes it by.

Except that line is a couple of years out of date.

Coupang bought Farfetch for $500 million in early 2024, back when everyone treated the deal as a strange distressed detour, and has since dragged it to roughly breakeven. Then came R.Lux, the luxury vertical that started with high-end beauty in late 2024 and folded in Farfetch’s entire catalog about a year ago. Today it carries something like 1,400 luxury labels, watches and fine jewelry included, with import taxes baked into the price and WOW perks attached.

The timing worked out better than anyone could have planned, because Korea’s homegrown luxury platforms spent the past two years collapsing, Balaan into court protection, Trenbe and Must’It with revenue cut in half, right as the biggest per-capita luxury spenders on earth minted a fresh class of rich people. I won’t pretend R.Lux is about to rival the marble halls at Shinsegae, since most of this boom’s luxury money is still walking through department store doors. But Coupang built a net for high-end spending right before the fish showed up, which is not nothing.

Beyond the luxury angle, two more things keep me from letting the bear case kill the idea.

First, even the stingy 1.3% applied to a quadrillion won is about $28 billion of new spending, which happens to be JPMorgan’s estimate as well. Coupang’s Korean revenue runs around $29 to 30 billion, so it doesn’t need to catch much of that flow for the numbers to move. If it captures even a modest slice, roughly in line with its share of overall Korean consumption, you’re talking about a billion or two of extra revenue landing on a network that’s already built and paid for.

Incremental volume at Coupang shouldn’t cost what average volume costs. The network is already built, and more volume running through that network should come with better economics than the average order. On a company that earned about $200 million in net income last year, even a billion or two of extra high-quality revenue would change how the business looks pretty quickly. It’s the difference between the market yawning and the market repricing the stock.

Second, the 1.3% figure was measured in a world where every Korean windfall went into real estate, and the government is actively closing that door. The lending curbs and cooling measures rolling out right now exist specifically to keep this boom from becoming another housing bubble, and if the traditional destination for windfalls is blocked, more of the money leaks into daily life.

It’s also worth noting that the people making this money skew under 50, which is exactly the demographic that already runs its entire life through a WOW membership.

Is there any evidence yet, or is this all just me connecting dots? A little, and I’d call it early evidence with the emphasis on early.

Coupang’s app user base held basically flat through the worst months of the breach and then made a new high in May at about 33.7 million monthly users. Web visits jumped from around 200 million in February to the 240 to 250 million range this spring. And Google search interest for 쿠팡 (Coupang) has been climbing for years but just spiked to a five-year high in the latest reading.

I’m not going to pretend a few months of third-party panel data proves a macro thesis, and some of that recovery is surely just people drifting back after the breach scare. But if Korean consumers were pulling back, or if the breach had done lasting damage, engagement is exactly where the cracks would show up first, and instead everything is pointed the same direction.

It’s not proof. It’s just what you’d hope to see if you were right, and so far it’s what we’re seeing.

To be clear about what I’m claiming here: the wealth effect is not my Coupang thesis.

The thesis is what it’s always been, an untouchable logistics network compounding year after year, now trading near the bottom of its range because of a scandal that just formally ended. The wealth effect is the free option sitting on top. If Korean consumption inflects, Coupang is the single most obvious beneficiary and nobody is modeling it, and if it doesn’t, I still own the best retailer in Korea at a price I like.

But it is a strange thing to watch. A country mints a quadrillion won in stock gains, the central bank starts worrying about bonus checks stoking inflation, and the dominant consumption platform in that country still trades like the breach never ended…

One of those two things is wrong. We’ll find out which sooner than later.

Share

Disclaimer: This article is for informational purposes only and does not constitute investment advice. Always do your own due diligence before making investment decisions.

It runs WeChat, and calling it a messaging app really undersells what it is. People message on it, pay with it, hail rides, order food, book doctor appointments, run entire small businesses, all without ever leaving it. More than a billion people open it every single day.

“WeChat has one of the widest moats I’ve ever seen.”
— , Saber Capital

There’s no real western equivalent and there isn’t a second app in China that comes close. That kind of position doesn’t get competed away. People have been trying to dislodge it for fifteen years and nobody’s gotten anywhere.

On top of that you’ve got the largest gaming business on the planet, an advertising machine that’s accelerating, a cloud and fintech arm, and a giant portfolio of stakes in other companies. The whole thing throws off enormous amounts of cash and earns very high returns on the capital it puts to work. This is about as high quality as large-cap tech gets, anywhere in the world.

And the stock is sitting at 52-week lows. Down almost 40% from its October highs. Trading around ~12x forward earnings and ~9x EV/EBITDA.

I keep staring at this and I struggle to make it make sense.

Why is it so cheap?

I think it starts with the market Tencent trades in, but I want to be careful with how I frame this, because it’s not the same as 2022 or 2023.

Back then the whole China complex was basically radioactive. Nobody wanted anything. Mainland stocks, Hong Kong stocks, ADRs, internet platforms, consumer names, property-adjacent names, all of it got thrown into the same uninvestable bucket.

This time is a little different.

Mainland China stocks have actually been acting pretty well. A-shares have rallied, AI names have ripped, and there’s plenty of speculative energy in parts of the domestic market. Some of the mainland AI stuff honestly looks quite frothy at the moment. Money is chasing semiconductors, robotics, AI infrastructure, domestic replacement, and anything that smells like policy support.

Hong Kong, on the other hand, has been left for dead.

Tencent, Alibaba, Meituan, Kuaishou, JD, and the rest of the group have been trading like investors want nothing to do with them. Some of that is macro. China’s economy has been sluggish and the consumer recovery hasn’t been nearly as strong as people hoped, myself included.

Property is still a drag, confidence is weak, and consumers remain very cautious.

I do think China eventually comes out of this, though more stimulus is probably needed. Right now the economy feels stuck in this odd half-recovery where certain parts are fine, certain parts are even hot, and the consumer still looks tired.

I publish this kind of thing most weeks. Subscribing is free, but paid unlocks the full archive of nearly 200 posts, plus every new writeup and research report going forward.

Subscribe now

Then there’s the flow problem.

Money that might normally find its way into Hong Kong internet is getting pulled toward mainland AI instead. A domestic investor can look at a semiconductor name ripping in Shanghai or Shenzhen, then look at Tencent grinding lower in Hong Kong, and the short-term choice becomes pretty obvious.

Why sit in a cheap mega-cap internet stock when the party is happening somewhere else?

I get it.

But these rotations always feel permanent while they’re happening. The hated group gets cheaper, the crowded group gets more crowded. Then at some point the easy money in the hot stuff gets made, the valuation gap gets stupid, and people start looking around again.

Tencent is on the wrong side of that rotation right now. That’s painful if you own it. It’s also why the opportunity exists.

The other piece is the AI perception thing.

There’s been a narrative for a while that Tencent is behind in AI. And to be fair, a year ago that had some truth to it. Tencent was slow and cautious at the start of the AI boom while Alibaba, ByteDance, DeepSeek, Baidu, Moonshot, and a bunch of others were moving as fast as possible to cement their place as AI leaders

I don’t think that’s a fair read anymore.

Tencent hired Yao Shunyu, a former OpenAI researcher, as its Chief AI Scientist last year, and reorganized the whole AI effort around Hunyuan. Then they launched the Hy3 preview model in April, and management has been clear that it’s now good enough to use across real products.

It’s been the top-ranked model on OpenRouter by token usage since late April, which is about as real-world a measure of developer adoption as you’ll find.

I’m not going to pretend I can tell you which Chinese model will be the best two years from now. But I also don’t think that’s really the Tencent story.

The advantage for Tencent is where the AI can go. They already own the place where people spend their time. That’s the part I care about.

It reminds me a little of Apple. Everyone has spent the last couple of years yelling that Apple is behind in AI, and maybe that’s true. But Apple has the device in your pocket, the operating system, the default apps, the payments layer, and the customer relationship. Being late matters a lot less when you already control the surface area.

Tencent has a version of that in China through WeChat.

Read more

But that kind of euphoria always creates ignored corners somewhere else. While everybody is piling into the same handful of trades, real businesses with real cash flow keep grinding away at multi-year lows because they don’t fit the theme of the moment.

So I want to do the unfashionable thing and write up five names I think are too cheap right now. Some of these I've written about before. Some haven't gotten as much airtime lately. What they all have in common is the stock has been getting worse while the actual business has been doing fine. Or in a couple of cases, very very well.

None of them are AI plays, at least not the kind people are chasing right now. That's not an accident.

That does not mean these are easy. Cheap stocks are usually cheap for a reason. Sometimes the reason is stupid. Sometimes it is very real. The job is figuring out which one you’re dealing with.

In no particular order…

Markel (MKL)

Markel is the lowest blood pressure name on this list.

No one is going to confuse this with a rocket ship. It’s specialty insurance, an investment portfolio, and a bunch of operating businesses inside Markel Ventures. That mix has always made the “mini-Berkshire” comparison easy, even if the comparison has been beaten to death at this point.

The stock has finally gotten interesting again. At around $1,850, you’re paying roughly ~1.3x book. That’s not a screaming 50-cent dollar. But for Markel, which has historically gotten more respect than this, it’s a reasonable starting point.

The insurance business also looks better than the headline numbers suggest. Q1 GAAP operating results were messy because of equity portfolio marks, but adjusted operating income was up 4%. The insurance combined ratio improved to 93%, and Markel Insurance adjusted operating income rose 31% to $369 million. That’s the stuff I care about.

Markel also bought back $134 million of stock in Q1, while adjusted underwriting gross premium volume still grew 10% excluding the impact of Global Reinsurance and Hagerty-related changes. I’m not looking for fireworks here. I’m looking for quiet compounding and intelligent capital allocation, and that’s what they keep delivering.

You don't need perfection from Markel. You need underwriting to stay disciplined, Gayner to keep allocating capital decently, and the share count to shrink when the stock is cheap.

This is not the name I expect to double overnight.

It is the one I expect to wake up ten years from now and be glad I owned.

The other four names are for paid subscribers.

Paid gets you the full archive, nearly 200 posts, plus every new writeup from here. It’s $24/month or $220/year.

I also opened a lifetime option to the email list yesterday. It’s $600 one time, no renewals ever. The Stripe link is capped at 10 total spots, and 8 are left as of right now.

Get Lifetime Access

Read more

I was listening to The Compound and Friends podcast a few weeks ago and Batnick brought up this Paul Graham idea about how experts can be wrong specifically because they’re experts on an earlier version of the world.

He also talked about the 1950s, how investors who’d been scarred by the Depression literally could not believe that stocks yielding less than bonds was a real thing. They had decades of experience telling them it was wrong. The experience was the problem. The regime changed and the people with the best historical pattern matching were the most confused by it.

I haven’t been able to stop thinking about it because I’m pretty sure I’m that guy.

Subscribe now

I’ve spent a lot of time over the years reading investing books and trying to study the greats. Buffett, Druckenmiller, Marks, all the usual suspects. And when you go down that road long enough you inevitably end up deep in bubble and crash history too. Tulips, 1929, Nifty Fifty, Japan, dotcom. I know the signs and patterns. Concentration at the top of the index, retail euphoria, everyone convinced the old rules don’t apply anymore.

And all of that reading did make me a better investor (I think). It gave me a real framework for risk. Kept me from doing dumb stuff when things got heated.

It also probably cost me a bunch of money over the last few years but let’s not dwell on that. (We’re going to dwell on that.)

What happened is I’ve been too skeptical of the US market for a while now. Not just big tech, kind of the whole thing. I’d look at valuations/market conditions, convince myself the whole thing was about to fall apart, and either pass, keep it small, or sit on cash waiting for the pullback that was always six months away. I found plenty of stuff to do overseas and in smaller value/GARP names and some of that worked well, but the broad US market? I was mostly just watching from the sidelines telling myself I was being disciplined.

Disciplined is a generous word for it. Scared is probably closer.

And the frustrating part is the market didn’t go up on nothing. It’s not some situation where stocks were floating on hopes and dreams. Earnings have been legitimately great. There’s a massive AI investment cycle underway where companies are spending absurd amounts building out real infrastructure, funded from their own cash flows. Whether all of that spending turns out to be perfectly allocated I have no idea, but it’s real economic activity backed by real money. That’s a different animal than what my brain keeps trying to compare it to.

I think what Paul Graham was describing is exactly what happened to me. I learned the historical patterns so well that anytime I see something that rhymes with a prior setup my brain just fills in the ending. Market’s concentrated? Must be a top. Multiples look elevated? Bubble. Everybody’s excited about new technology? Must be 1999. Except sometimes it isn’t 1999.

The historical record can only tell you about stuff that already happened, and sometimes the new stuff matters more than the rhymes.

Could it still end badly? Obviously. I’m not going full perma-bull here. There are parts of this market I still wouldn’t touch. But I do need to look at the scoreboard, and the scoreboard says I’ve been predicting nine of the last two crashes. Not exactly a great hit rate.

The Nifty Fifty is a good example/case study. Most folks know the cautionary version. Polaroid traded at something like 90x earnings at the peak, dropped over 90% in the crash, eventually went bankrupt. But if you actually go back and look at what happened to that group over the following decades, a bunch of those names worked out fine from those prices. McDonald’s. Coke. Philip Morris. You overpaid and the business bailed you out over time. The lesson everyone took away was never pay up for anything. The real lesson was more like… don’t pay up for the wrong businesses. Completely different takeaway. Funny how the second version barely gets mentioned.

When you look at who actually compounded real wealth over long periods, it’s never the people who called every top. Buffett didn’t make his money being the smartest macro guy in the room. He just owned good businesses for a really long time and got aggressive when everyone else panicked. He sat through some ugly drawdowns. Didn’t matter. The businesses kept compounding underneath.

It’s been the same for me. The wins came from doing the work on individual names and having the conviction to size them. Not from macro calls. Every time I’d look at forward estimates I’d convince myself there was no way companies could actually hit those numbers, and then they’d go and hit them. Over and over. That skepticism felt like discipline at the time. In hindsight it was mostly just wrong.

I still think parts of the AI trade look bubbly. I’m not abandoning all skepticism here. But the internet had bubble dynamics too and then went on to change basically everything about how the world works. Plenty of smart people called the dotcom mania correctly and then spent the next twenty years on the wrong side of the biggest wealth creation event in history. They were right about the bubble. Didn’t help.

I think at some point being the skeptic just became my default setting and I didn’t totally notice it happening. Caution always sounds smart. That’s what makes it tricky. You can spend basically any point in the last decade worrying about valuations and sound perfectly reasonable doing it. But sounding reasonable and making money aren’t really the same thing, and I’m trying to remember that going forward.

Anyway. No clean takeaway here, mostly just rambling. Back to regular stuff soon.

Now that Meituan, Alibaba, and JD have all reported, I wanted to try and get a clean read on where market share actually sits today. Not the perfect answer, because that doesn’t really exist here, but a reasonable estimate using the best public data we have.

The annoying part is that none of these companies make this easy. Meituan stopped giving clean food delivery GTV years ago. Alibaba gives the best disclosure, but even that revenue line is net of subsidies. JD basically buries the whole thing inside New Businesses.

So this is not me pretending I have the number nailed to the decimal. I don’t. Nobody does. But after going through the filings, earnings calls, management comments, and sell-side estimates, I do think the rough picture is pretty clear.

The short version: the last eighteen months have been brutal for Meituan, with Alibaba and JD both in full attack mode.

For newer readers, I spend a lot of time writing about China, China internet, and AI. A few related pieces that might be helpful:

Read more

The favor first, and it’s mostly for the paid crowd. I want to know what actually pulled you in. Was it a specific name/stock, the China stuff, or just the general way I think through things? I ask because I genuinely don’t know, and I’d rather not guess.

When you write for a while you build a story in your head about who’s reading and why, and that story is usually half wrong. If a bunch of you came in for one name and I’ve been off writing about something else, I’d like to know that before I spend another three months on the wrong thing.

I put it in a quick survey, it’s four questions, takes about a minute. No pressure, and obviously feel free to skip the ones that don’t apply. I promise it's not going into a void... your responses will actually change what I write and spend time on, so weigh in.

Start Survey

For what it's worth, the usual suspects aren't going anywhere. Asia/China stays front and center, and so do my largest positions. The survey is really about everything past that. I tend to start more things than I finish, and knowing what you'd actually want to read would help me close a few of those tabs.

Last thing, and then I’m done asking you for stuff. Through end of day Sunday, May 31, I’m running 30% OFF the annual plan. If you've been on the fence, this is the cheap window. And if you're already paid, ignore this part and just go yell at me in the survey.

Get 30% off forever

Thanks for reading, as always. It’s a strange and great thing that any of you do.

But I still think it’s worth writing up because nothing about the actual opportunity/thesis has changed. The stock is just less cheap than when I started.

The company is a semiconductor equipment maker. Sells the tools Chinese fabs need to build chips domestically. Cleaning, plating, furnaces, deposition, packaging. I actually first came across it through on X, he honestly knows this name and the broader semi space way better than I do. If you’re not following him you absolutely should be.

A few years ago this was basically a single cleaning-tool business. Now it ships across six or seven product categories, with the newest lines growing triple digits. And even after the recent run, it trades at a roughly ~45% discount to the public market value of its own subsidiary.

The stock’s entire market cap is less than half of what its stake in the subsidiary is worth… oh and there’s about $1 billion in net cash sitting on the balance sheet.

I believe domestic chip production in China is going to accelerate significantly over the next several years, driven by AI demand and hardened by export controls. If that view is even roughly correct, the companies selling tools into that buildout are going to do very very well. This one is already starting to do well, the market just hasn’t fully caught up to it yet IMO.

I really do think this is one of the more interesting risk/reward setups I’ve come across in a while.

Read more

At the time, Baidu had something like $30 billion in market cap with $27 billion in net cash and equity stakes sitting on the balance sheet. You were basically getting paid to take the operating business for free.

I wrote it up, laid out a sum-of-the-parts, and the conclusion was very, very simple: even if you were deeply skeptical about AI, robotaxis, and everything else management was pitching on conference calls, the math didn’t require any of it to work…

The stock has roughly doubled since then. The original cheap-stock pitch has mostly played out and the “easy” money has been made.

But the business has also gotten more interesting, which is why I’m writing about it again. And most of that comes down to Kunlunxin.

Kunlunxin is Baidu’s chip subsidiary. I mentioned it last year but honestly I didn’t spend much time on it. It was making AI chips for Baidu’s own data centers, which was fine, but it wasn’t a real business yet.

It didn't stay that way for very long. Kunlunxin is now pursuing a dual-track IPO in Hong Kong and on Shanghai’s STAR Market with a reported target valuation of at least $14.7 billion. Jefferies has the range at $16-23 billion.

I completely missed this, and I think it’s a good example of why buying stocks trading around net cash (even in China) tends to work out. You were sitting on free call options all over the business. Kunlunxin was one I didn’t even know I had.

So I want to spend most of this post trying to actually think through what the chip business could be worth and what that means for Baidu’s valuation today.

They just reported Q1 earnings though so I guess we should start there.

Q1 2026 Results

Baidu reported on May 18th. Revenue was RMB 32.1 billion, and Baidu General Business (which strips out iQiyi) came in at RMB 26 billion, up 2% year-over-year. First quarter of positive growth in a while.

The AI-powered business crossed ~50% of General Business revenue for the first time. RMB 13.6 billion, up 49%. AI Cloud Infra did RMB 8.8 billion, up 79%. GPU cloud revenue grew 184%. Impressive huh?

Online marketing revenue was RMB 12.6 billion, down 22%. Search ads are shrinking, and honestly faster than I expected. Although part of me wonders if AI is just growing so fast that it’s making everything else look worse by comparison. A year ago search was 63% of General Business revenue. Now it’s 48%. Either way the business is moving.

Non-GAAP net income was RMB 4.3 billion at a 14% margin, adjusted EBITDA was RMB 6 billion at 19%, and they beat EPS estimates by a few percent.

The balance sheet has roughly $40.5 billion in total cash and investments against about $12.6 billion in debt and debt-like financing, so call it roughly $28 billion of net cash and investments.

That number is still in the same ballpark as when I first wrote about Baidu in 2025, helped by slightly lower debt.

The quarter confirmed what we already knew, which is that AI is becoming the business while search slowly fades into the background.

The Chips

So, back to Kunlunxin.

The quarter is mostly context. Baidu has been moving this way for a while now. AI is becoming a bigger piece of the business, and each quarter has made that a little harder to ignore.

But Kunlunxin is the part I think still needs more attention.

The company went from roughly RMB 2 billion in revenue in 2024 to an estimated RMB 3.5 billion in 2025, and it reportedly hit breakeven, with external customers (meaning not Baidu) crossing 50% of revenue for the first time.

The customer list is pretty hard to argue with: Tencent, Vivo, China Mobile (which gave them a contract worth over RMB 1 billion), with JD, Meituan, ByteDance, State Grid, and China Merchants Bank all in various stages of testing or deployment. By unit volume, Kunlunxin is the #2 domestic AI chip shipper in China behind Huawei’s Ascend.

Two years ago this was an internal cost center. Now it has a legitimate third-party customer base and it’s filing for a dual-track IPO.

The reason all of this happened is pretty obvious, at least in hindsight. Nvidia can’t sell advanced chips into China anymore, and no major Chinese company is going to build its AI future assuming that changes. Whether the restrictions ease or tighten, the pull toward domestic compute was always going to accelerate once it started. Kunlunxin happened to be in the right place with working silicon when the demand showed up.

Read more