App Economy Insights — This Week in Visuals: BRK.B TCEHY CSCO AMAT · SE NU JD ADYEY · ONON HIMS GLBE MNDY STNE DLO

Berkshire is Warren Buffett's conglomerate — it owns whole businesses (railroads, insurers, manufacturers) and a giant portfolio of stocks. For years its defining feature was a cash pile so large it became a running joke: nearly $400 billion sitting idle because nothing looked cheap enough to buy.

This quarter that changed. Under new CEO Greg Abel, Berkshire bought $23 billion of stocks while selling only $3 billion — its first quarter as a net buyer in over three years — including about $10 billion of Alphabet. It also bought back $4.5 billion of its own shares (and an estimated $3.4 billion more in July) and paid $6.8 billion for homebuilder Taylor Morrison. Cash fell to $365.5 billion. The underlying businesses were mixed: manufacturing profit +24%, but insurance underwriting fell 13% and GEICO's underwriting earnings dropped 45%.

App Economy's framing is the reason this is neutral rather than positive: the money is finally moving, but that only replaces one question with a harder one. "It is whether Abel can earn Buffett-like returns on it" — and that takes years to judge.

Tencent runs WeChat — China's do-everything messaging app, with about 1.4 billion users — plus the world's largest video-game business and a cloud arm. Revenue grew 11% and, importantly, sped up from the prior quarter, with advertising up 22% (AI is making its ad targeting better) and Chinese gaming up 17%.

The headline of the section is the cost. Capital spending — the money laid out on AI chips and data centres — jumped 176% versus a year ago to about $7.8 billion in a single quarter. That flipped free cash flow (the cash left after all that spending) from a positive ¥56.7 billion to a negative ¥13.8 billion. Tencent's defence is that if its own products don't use all that computing power, it can rent the surplus out through Tencent Cloud. App Economy's two-word reaction — "Sounds familiar?" — is a nod to Amazon, Alphabet and Meta making exactly the same argument about exactly the same problem.

Why still positive: unlike a pure infrastructure bet, Tencent already owns the distribution that could pay for it. A WeChat AI agent that books your ride is being tested, WorkBuddy is China's most-used office AI agent, and usage of its Hy3 model is up twenty-fold since April. The stance rests on that distribution turning into revenue; the risk is that the bill keeps arriving before it does.

Cisco makes the switches and routers that move data around inside networks — the plumbing of the internet and of every corporate data centre. The AI boom needs enormous amounts of that plumbing, and this quarter Cisco booked $4 billion of AI orders from the big cloud companies alone, taking its full-year AI orders to $9.3 billion against a roughly $9 billion target.

The most interesting number is a subtraction. Total product orders grew 35% — but 25% even if you strip out the hyperscalers entirely. That says ordinary enterprises are upgrading their networks too, in preparation for AI, so this is a broad refresh cycle rather than a handful of giant cloud contracts.

The stock still fell about 4%, because Cisco only guided $7.5 billion of AI revenue for next year and investors thought that looked small against the order book. That is the tell App Economy points at: the argument has moved on from "is the demand real" to "how fast can Cisco convert orders into revenue without the hardware-heavy mix eating its margins" — gross margin already slipped two points to 66% on memory costs.

Applied Materials doesn't make chips — it makes the machines that make chips. When memory manufacturers race to build more HBM (the high-speed memory that sits next to AI accelerators), they buy Applied's equipment first. This quarter revenue hit a record $9.1 billion, up 25%, with a record 34% operating margin, and DRAM memory now accounts for 26% of its semiconductor-systems sales.

The striking detail is the shape of the constraint. Customers give Applied eight quarters of forward forecasts and are "pushing Applied to deliver equipment faster" — the company has nearly doubled its factory space and plans to double how many systems it ships per quarter by 2028. In other words the bottleneck has moved from demand to Applied's own ability to build.

So why did the stock slip on a record quarter with guidance far above expectations? Because it had nearly doubled this year already. That is the caveat App Economy attaches to the positive read: the business is getting better, but the good news is already in the price, so the stock can be right and still not go up.

Sea runs three businesses across South-East Asia: Shopee (online marketplace), Monee (consumer lending) and Garena (games, best known for Free Fire). The number that matters this quarter is the gap between two growth rates: the value of goods sold through Shopee rose 28%, but Shopee's own revenue rose 49%.

That gap is the "take rate" — the slice of every transaction Sea keeps, which climbed from 12.6% to 14.6% via higher seller fees and advertising growing more than 70%. Marketplaces are usually valued on how much of the volume they can eventually capture, so proving the take rate can rise without killing volume is the single most valuable thing Shopee can demonstrate.

The offsetting worry: profit isn't following. Group EBITDA (a rough proxy for operating profit) rose only 11% and the margin fell to about 12% from 16%, partly because the fast-growing lending arm carries heavy marketing and loan-loss provisions. So Sea is proving it can monetize while not yet proving it can turn that into operating leverage — profits growing faster than revenue as scale kicks in.

Nu is the digital bank that started in Brazil and now serves 139 million customers across Latin America. The perennial bear case on any fast-growing lender is simple: growth in the loan book today becomes bad debt tomorrow. This quarter is the direct rebuttal.

Nu grew its credit portfolio 37% to $39.4 billion while the cost of that credit actually fell 9% quarter-over-quarter. "Risk-adjusted NIM" — the interest margin it earns after subtracting expected loan losses, i.e. what the lending is really worth — rebounded to 12.4% from 9.5%, and early-stage missed payments improved. Net income passed $1 billion in a quarter for the first time at a 33% return on equity, which is exceptional for a bank of any kind.

The forward story is Mexico, described as "the Brazil playbook on fast-forward": 15.8 million customers, now operating as a full bank, and already earning $12.3 per active customer versus $5.6 for Brazil at the same stage of development. It is monetizing new customers roughly twice as fast as it did the first time around. Set this beside StoneCo in the same issue — same country, opposite credit outcome — and the divergence is the point.

JD.com is one of China's two big online retailers, and unlike a pure marketplace it buys and warehouses much of what it sells. This quarter revenue actually fell 3% — its first quarterly decline since listing — partly because last year's government appliance-subsidy boom made for an impossible comparison, and partly because Chinese consumers are spending cautiously.

Underneath, though, profitability improved sharply. The lower-margin business of selling goods shrank 5% while the higher-margin service business (letting third parties sell on JD, plus advertising) grew 7%, and gross margin expanded for a seventeenth straight quarter. The bigger swing was food delivery: JD had been burning cash subsidizing that war, and cutting the subsidies halved the losses, more than doubling group EBITDA.

Neutral is the honest read because the two trends point opposite ways. Margins and profits are recovering fast, but the top line has gone backwards, and a retailer that fixes profitability by shrinking hasn't fixed much. "The next test is whether JD can restore top-line growth without giving those profitability gains back."

Adyen is the Dutch payments company that processes card and online transactions for large merchants — the machinery behind checkout for the likes of Spotify, Uber and eBay. It processed €804 billion of volume in the half, up 24%, with net revenue up 19% and an EBITDA margin of 49%, and growth accelerated as the half went on.

The strategic shift is that Adyen is no longer content to just move money. In July it made its first two acquisitions ever — Talon.One (loyalty programmes and promotions) and Orb (billing for subscription and usage-based pricing) — with the goal of becoming a "financial operating system" for merchants: the system that handles payments, billing, promotions and money movement together. It also signed OpenAI as a customer, which matters as a signal that AI-native companies and "agentic commerce" (software agents buying things on your behalf) will need payment rails.

The cost is honest and disclosed: the two deals will shave about a percentage point off this year's margin, and Adyen is pulling forward data-centre spending to roughly 7% of revenue. So the platform gets more valuable while the execution bar rises, against a long-term margin ambition of 55%+.

On is the Swiss premium running-shoe brand. Its shares fell nearly 20% because revenue missed and management cut this year's growth guidance from "at least 23%" down to the low 20s — a real disappointment for a stock priced as a compounder.

But the reason behind the cut is unusual, and it's why this isn't simply negative. On's direct-to-consumer business (its own stores and website) grew 26% and is now 46% of sales. The weakness was wholesale — selling to retailers — and On deliberately held back shipments in the Americas rather than push shoes into stores that would only discount them. Its CFO said plainly that the company was "willing to sacrifice volume to protect full-price selling and retailer inventory health."

That discipline shows: a record 65% gross margin, EBITDA up 24%, apparel growing 56%, and gross-margin guidance raised again. The unresolved question is which story this is — a brand choosing short-term pain to protect its premium positioning, or the first visible crack in a growth story that was priced for perfection. App Economy leaves it open, and so does this rating.

Hims & Hers is a telehealth company: you fill in a questionnaire online, a clinician prescribes, and medication arrives by subscription — originally for hair loss and erectile dysfunction, now increasingly for weight loss. Revenue grew 38% to $753 million and subscribers reached 2.9 million.

The trade-off in the headline is a margin one. Selling branded GLP-1 weight-loss drugs (the Ozempic/Wegovy class) means buying them from a pharmaceutical company at a real cost, unlike the cheap generics Hims built its business on. That pushed gross margin down from 76% a year ago to 64%, and management says the lower margin is here to stay. Free cash flow swung to −$68 million.

So the business is bigger and growing faster — guidance was raised to $3.1–$3.3 billion, international is now material after buying Australia's Eucalyptus, and testosterone is scaling quickly — but it is becoming a different, structurally less profitable company. App Economy's line captures the neutral stance: Hims is "increasingly looking like a lower-gross-margin global healthcare platform rather than the exceptionally high-margin telehealth model investors were used to." Investors have to decide which multiple that deserves.

Global-E lets an online store sell internationally without dealing with the mess of doing so — local currencies, duties, taxes, shipping and returns are handled for the merchant. It is deeply tied to Shopify, whose merchants are its main customer base. Revenue grew 39% and the value of goods sold through it accelerated to +44%.

Two things drive the positive read. First, "Managed Markets" — a lighter-touch version of the product where Global-E doesn't take legal ownership of the goods — finished migrating to a new platform (V2), expanded into Canada and the UK, and got simple enough that a merchant can switch it on in a single sitting. That is what makes it scalable, and App Economy calls it the first real evidence it could become "a meaningful Shopify growth lever."

Second, buying Passport — an asset-light cross-border logistics platform — extends the addressable market to merchants who never fit the original "Merchant of Record" model. And this is growth that pays: EBITDA rose 62% with margin up three points to 21%, and guidance rose again on both volume and revenue.

Monday.com sells work-management software — the boards and workflows teams use to run projects, plus CRM and IT service products built on the same platform. Revenue grew 22% and it beat handily on profit, with operating margin expanding to 17%.

The encouraging part is AI adoption showing up as money rather than talk: recurring revenue from AI products doubled quarter-over-quarter and made up 17% of all new recurring revenue, up from 10%. Its pricing model gives customers a bundle of AI credits with their seats, and they are consuming past the bundle — the clearest sign the features are actually used.

The problem is everything else is slowing. Despite beating this quarter, management left full-year guidance unchanged and guided next quarter to 16–17% growth — a real step down from 22%. It also cut roughly a fifth of its workforce to concentrate on AI, CRM and service management. So the AI lever is working; it just isn't yet big enough to offset the deceleration in the core, which is exactly why this reads neutral.

StoneCo provides card-payment processing to small and medium Brazilian merchants, and has been pushing to become their main bank — taking deposits and, crucially, lending to them. The lending part is growing spectacularly: the loan book more than doubled in a year and credit revenue rose 153%.

That is also the problem. Provisions — money set aside for loans expected to go bad — jumped 128%, and loans more than 90 days overdue rose to 8.6% of the book from 4.7% a year ago. Nearly doubling your bad-debt rate while nearly doubling your loan book is the classic sequence in which a lender discovers it grew too fast. Management blames specific weaker loan "vintages" (batches originated in late 2025 and early 2026) plus a few large troubled borrowers.

Meanwhile the underlying payments business is barely growing — revenue +3%, payment volume +4% — and reported earnings per share only rose because buybacks shrank the share count. Guidance was kept but steered toward the low end because Brazil's very high interest rates are not letting up. The negative read is about sequencing: the credit deterioration is visible now, while proof that those vintages were an anomaly is quarters away.

dLocal handles payments in emerging markets for global companies — if Netflix or Amazon wants to collect money in Nigeria, Egypt or Argentina using whatever local payment method people actually use, dLocal makes that work. The volume it processes grew 92%, its fastest in four years, to $17.7 billion.

But gross profit grew only 29%, because dLocal earns much less per dollar it moves than it used to: 0.72 cents per dollar of volume, down from 1.07 a year ago. Two reasons — lower-margin "local-to-local" payments (money moving within one country, simpler and cheaper than cross-border) surged 141% and are now 61% of volume, and several big merchants crossed volume thresholds that automatically earn them cheaper pricing. App Economy's summary is blunt: "dLocal is processing vastly more money but earning less on each dollar."

That is a real risk — a business whose unit economics keep compressing has no natural floor. But two things pull the rating back toward positive. Volume is growing so fast that total gross profit still beats expectations, so guidance went up. And costs are behaving: operating profit rose to 50% of gross profit from 44%, with more improvement expected as AI automation lands and front-loaded marketing rolls off. The open question is whether the take rate ever stabilizes.