Current Market Opportunities | Mercado Libre, Amazon, & Constellation Software (TIP808)

ELI5 / TLDR

Clay Finck’s farewell episode at TIP, with Daniel Mahncke walking through four names that have all sold off: Mercado Libre, Amazon, Constellation Software (plus its spinoffs Topicus and Lumine), and Hermès. The unifying thread is destination analysis — Nick Sleep’s habit of asking where a business will be in ten years rather than next quarter. Margins are compressed today because of deliberate reinvestment (MELI’s credit book, Amazon’s $200B AI capex, Lumine’s carve-out cleanup) or AI panic (Constellation, the SaaS complex). Daniel argues the panic mostly mistakes investment for weakness, and that the same scaled-economies-shared playbook that worked for Amazon for two decades is being repeated by MELI and, less obviously, by the CSI ecosystem.

The Full Story

Mercado Libre — the Amazon playbook in Spanish

The headline number is absurd: 28 consecutive quarters of >30% YoY revenue growth, with the latest print at 45%. The stock fell anyway because operating margins took a 5–6 percentage point hit. Three things drove the compression: a doubling of the credit portfolio, lower free shipping thresholds in Brazil, and a scale-up of the cross-border and first-party businesses.

Daniel’s frame is borrowed wholesale from Nick Sleep’s old letters on Amazon — the market is essentially telling MELI’s management “your gross spending has no value, turn yourself into a cash cow.” Sleep called this absurd for a business growing 20%+. MELI is growing 45%.

The destination thesis rests on two structural gaps. E-commerce penetration in Latin America sits at 14–15% versus ~25% in the US, ~30% in the UK, and >30% in China. MELI owns 30–35% share in Brazil, its biggest market. The fintech arm rides the same wave — Latin Americans aren’t going back to cash and informal lending.

The credit book is the spiciest piece. Loans average under four months, credit cards under three months, and the data comes straight off the marketplace, which is the first place a struggling consumer cuts back. That’s a much faster feedback loop than a generalist bank gets. Daniel still prefers MELI to Nubank because of that data advantage.

The competitive moat in Latin America is geographic commitment. Amazon has a dozen other places to deploy capital. Shopee (parent: Sea Limited) is primarily a Southeast Asia business — when Sea’s stock collapsed 90% post-COVID, Shopee shut down most of its South American footprint and now only operates in Brazil. MELI has no other continent to retreat to. That focus is itself a moat.

Amazon — robots, ads, and an under-loved marketplace

Daniel and his co-host Shawn skipped Amazon when they first looked at it, then bought below $200 after the AI capex panic. The thesis hinges on three flywheels.

First, AWS. Andy Jassy now expects AWS to hit $600B in revenue by 2036, double the prior internal estimate, implying a 15% CAGR over a decade. AWS revenue grew 24% in the last quarter and is accelerating. There are only three credible hyperscalers — Amazon, Microsoft, Google — and demand still outstrips supply. The $200B capex cycle may create overcapacity for one or two years, but the destination question is whether AWS grows into that capacity over five. Daniel thinks obviously yes.

Second, robotics. Amazon employs over a million warehouse workers globally. The new generation of robots can handle picking and packing, not just shelf-shuffling. They store inventory ~75% faster and cut order processing by up to 25%. Total fulfillment and shipping cost is around $90B per year — a conservative 10–15% reduction adds $9–14B to the bottom line, and Daniel thinks the majority of that $90B could come out within five years. North America retail margins have already gone from 5% in 2021 to 11% today.

Third, Amazon’s own AI consumption. Bedrock distributes hundreds of foundation models — Amazon’s Nova, Anthropic’s Claude (Amazon owns a substantial stake), and now OpenAI’s models via a recently announced $50B deal. The more interesting use case is internal: better search, better ad targeting, the Rufus shopping assistant, and the computer-vision systems behind warehouse robots. Amazon is sitting on more product-handling data than anyone, which feeds the robotics flywheel directly. The ads business alone is now >$70B in revenue at very high margins.

Constellation Software — the AI panic and the bear case worth taking seriously

Constellation owns 1,000+ niche vertical market software (VMS) businesses — cemetery management software, court administration systems, golf course tee-time platforms. For two decades the market treated this as undisruptable and gave it 40x cash flow. AI changed the mood.

The bear case has two prongs. First, AI makes building competing software cheap, so new entrants flood in and the existing portfolio gets chipped away. Second, the same cheapness shrinks the M&A pipeline.

Daniel’s counter is built on three observations.

The 10x rule. Switching software in regulated industries is brutal. Court records have a legal chain of custody — every migration has to prove timestamps, access logs, and permissions stayed intact, and any AI tool involved has to be tested and certified. Anything new has to be roughly 10x better to justify that risk.

The cost of switching isn’t the software. Over three quarters of CSI’s revenue is maintenance — ongoing support, updates, the human team that keeps things running. AI lowers the cost of building the software but not of supporting it. CSI’s competitors would still need that team. Meanwhile the software is rarely a meaningful line item — a $2M/year golf course pays maybe $10,000 for software that handles tee times, dues, payments, and tournaments. Retention runs above 90%. The course manager isn’t going to vibe-code their own replacement.

Distribution and switching costs in unregulated verticals. Even if you build a better gym-management product, you have to reach gym owners and persuade them to switch. The trigger event is usually a payment-data hiccup that asks members to re-enter card details — and gyms make most of their profit from members who never show up. Re-entering the card is when those members cancel. The cost of a botched switch dwarfs the savings.

That said, Daniel is honest about one risk worth thinking about: the agentic future. If LLM agents do the navigating, the underlying software loses its training-cost moat and becomes a commodity database. CSI’s decentralization — usually a feature — turns into a bug here, because each subsidiary’s data is siloed and a horizontal AI built on consolidated vertical data could produce better models than CSI can build subsidiary-by-subsidiary. The ROI of building an agent for a $5–10M EBITDA subsidiary doesn’t pencil. ServiceNow can build one workflow engine for a unified platform. CSI cannot.

The growth engine is also evolving. Constellation does ~100 acquisitions/year at $5–10M apiece. The law of large numbers makes more difficult, so they’ve launched PMS — Permanently Engaged Minority Shareholders — taking stakes in larger public companies and pushing for board seats. The first big one is Sabre (SABR), down 90% over five years, where Mark Leonard himself is reportedly leading the charge. Sabre’s management threatened a poison pill but didn’t pull the trigger. Private market valuations haven’t moved much; public markets have, which is why CSI is fishing there.

Topicus and Lumine — younger, smaller, possibly easier to compound

Topicus was spun off in 2021 and operates the European VMS market via TSS (Total Specific Solutions). Mark Leonard sits on the board, CSI owns ~30%. Europe is harder to consolidate than North America because of language and regulatory fragmentation, but that fragmentation is also the moat — Topicus is one of very few players with the scale to do it. The European succession problem (founders aging out without successors) is even more acute than in North America.

The incentive structure is a small but telling detail: TSS CEOs are measured on total revenue growth including acquisitions, while the original Topicus arm’s CEO is measured only on organic growth. Different jobs, different bonuses.

Lumine is geography-agnostic but vertical-specific (media and communications), spun off in 2023. It went from above $50 to below $20 in months. Lumine specializes in carve-outs — orphan units inside larger companies. Less competitive bidding, more cleanup work, larger deal sizes. Carve-outs explain why Lumine’s organic growth can swing negative in any quarter — they’re cancelling unprofitable contracts and rebuilding from scratch. Two to four deals a year can move the needle on a 34-company base, versus 100+ for the parent.

Mark Leonard’s line in the spinoff press release: “I hope my grandkids are still holding Lumine shares 50 years from now.” Daniel and Clay both treat that as a real signal given Leonard’s allergy to overstatement.

Hermès — the inversion bet

Clay’s frame here is from Linde plc — instead of asking what wins from AI, ask what’s almost certain not to fail. Hermès is down >40% from its all-time high on Middle East narrative noise.

The case rests on customer concentration at the very top — top 0.1%, arguably top 0.01% — which is the fastest-growing luxury cohort (10% CAGR vs ~1% for aspirational buyers) and least exposed to macro. Family-run since 1837, sixth generation, every heir starts as a production apprentice and spends a decade there before they’re allowed an executive role. Zero risk of the shareholder waking up to find production moved to Asia.

The risk Daniel flags is China — both the social-media flaunt-shaming policy and a longer-run shift toward domestic luxury brands. He thinks both hit LVMH harder than Hermès, since the top 0.01% are global citizens and don’t post their handbags on Weibo.

The valuation discipline matters. Hermès still trades at ~40x cash flow. Ferrari was at similar multiples not long ago and is now at 25x. A drawdown from 40x doesn’t mean you’ve found a bargain.

Key Takeaways

MELI: 28 consecutive quarters of >30% revenue growth, latest at 45%. Margin compression is deliberate reinvestment in credit, shipping, and 1P. LatAm e-commerce penetration is half of the developed world; MELI owns 30–35% of Brazil. 90%+ of GMV is third-party, structurally higher margin than Amazon’s blended mix.
Amazon: AWS is now expected to hit $600B by 2036 (15% CAGR). $90B annual fulfillment cost is the robotics prize — a 10–15% cut equals $9–14B straight to the bottom line. NA retail margin already 5% → 11% (2021 → today). Ads business >$70B at high margins.
Constellation: Bear case is real but mostly wrong. The moat is maintenance revenue (75%+ of total), regulatory switching costs, and trivial software spend as a share of customer revenue. Real risk is agentic AI commoditizing the underlying database — CSI’s decentralization makes building unified vertical agents harder than for centralized SaaS.
CSI’s PMS strategy (minority public-market stakes) is a response to the law of large numbers, not the AI panic. Private-market multiples haven’t moved; public ones have.
Topicus and Lumine offer the same operating playbook from a much smaller base. Lumine is a carve-out specialist with lumpy organic growth — the cleanup work suppresses headline numbers in the short term.
Hermès is the inversion pick — top 0.1% customer base, sixth-generation family ownership, structural insulation from macro and aspirational-luxury cycles. Still expensive at 40x cash flow.

Claude’s Take

Daniel is a careful thinker and the destination-analysis framing does most of the heavy lifting in this episode. The MELI section is the strongest — the Amazon playbook analogy isn’t lazy because the structural differences (3P-first, no AWS subsidy, geographic captivity) are spelled out. The credit-data argument is also sharper than the usual “they have data” hand-wave because the loan duration is genuinely short enough to matter.

The Amazon section is fine but mostly recapitulates the consensus AWS-plus-robotics story. The interesting datum is the 5% → 11% North America retail margin inflection, which suggests the operating leverage is already showing up — not waiting on robots that haven’t been deployed.

The Constellation section is the most intellectually honest part. Daniel doesn’t pretend the AI bear case is silly. The decentralization-as-disadvantage point is the kind of thing you don’t usually hear from a holder, and it’s correct — CSI’s structure is genuinely worse for amortizing AI infrastructure than ServiceNow’s. The counter (maintenance revenue, switching costs, distribution) is well-argued but doesn’t dispose of the long-tail agent risk. If you’re long CSI for ten years, that’s the question worth losing sleep over, not whether some kid vibe-codes a tee-time app.

The Hermès section is short and reasonable. The 40x-cash-flow caveat at the end is the right note — drawdowns from premium multiples are not the same as bargains.

Two weaknesses. First, almost no quantitative valuation work — this is qualitative thesis-walking, not a model. Useful as a frame, less useful as a buy signal. Second, Daniel and Clay are both holders of every name discussed. The episode is a defense of an existing portfolio more than a fresh search. That’s fine, but worth flagging.

Score: 7/10. Solid, well-organized, honest about risks. Not novel — most of these arguments exist elsewhere — but cleanly assembled in one place. The CSI section alone is worth the runtime.