In 2010, I started contributing to my retirement account. Back then, the maximum annual IRA contribution was just $5,000, and I’ve contributed the maximum every year since. Over the last approx 16 years, I’ve invested roughly $100,000 of my own money. Through patient, sensible investing, the account has grown to more than $1.1 million.
At first glance, that looks like roughly a 16% CAGR, but that’s actually a misleading way to measure the return because it assumes the entire $100,000 was invested on day one. In reality, money was added every year for 16 years, so much of the capital had less time to compound. The proper way to measure performance is with a money weighted return calculation (XIRR), which would be north of 25% CAGR. Calculating the exact figure requires the precise dates and amounts of every contribution, which I am too lazy to do.
For perspective, the S&P 500 compounded at roughly 14% annually over approximately the same period. The lesson isn’t that I beat the market, the lesson is that even a small annual edge, compounded over many years, can produce dramatically different outcomes.
To be fair, I was fortunate enough to own two investments that each became 30 baggers, and they accounted for a significant portion of the gains. Luck always plays a role in investing. But luck also tends to favor investors who stay in the game long enough to let exceptional businesses surprise them.
One of the biggest lessons I’ve learned is that investing isn’t about being right all the time. I made plenty of mistakes along the way. The overwhelming majority of my wealth came from just a handful of decisions. You don’t need to hit a home run every year. You need to avoid catastrophic mistakes, recognize exceptional businesses when you find them, and then have the discipline to hold them while they continue compounding.
The hardest part wasn’t finding those businesses. The hardest part was continuing to hold them after they doubled, tripled, went up 10x, and eventually more than 30x. Every step of the way there were reasons to sell. Every step of the way there were people saying they were overvalued. Looking back, my greatest enemy wasn’t inflation, recessions, or bear markets. It was the temptation to interrupt compounding.
I also realized something fascinating as the portfolio grew. In the beginning, my annual contributions were the primary driver of my wealth. Eventually, they became almost irrelevant. The businesses themselves began creating more wealth than I could contribute through new savings. That’s the moment you truly understand the power of ownership. The first $100,000 came from me. The next million increasingly came from the businesses I owned.
That’s why I believe the biggest secret in investing isn’t finding hundreds of great stocks. It’s finding a few truly exceptional businesses, buying them at sensible prices, and then having the patience to let time do what almost no investor is willing to allow it to do.
Compounding doesn’t look impressive in the beginning. It whispers for years before it roars (borrowed this quote is from my book 😉). The greatest fortunes are rarely built by constantly making brilliant decisions. They’re built by making a handful of great ones… and then having the discipline to get out of their way.
🌹
OPTIMUS JUST TOOK A $20-AN-HOUR COUNTER JOB IN LOS ANGELES
tesla's diner. real customers. real counter.
the robot stands right where you order - gold chest, black gloves, a row of blue water bottles lined up in front of it. not behind glass. not fenced off in a demo booth.
and people don't just walk up to order.
they line up to take selfies with the staff.
that's the actual play:
a counter worker in california costs $20/hour minimum. $41,600 a year. they hand you a drink.
this robot minds the drinks AND puts the diner on every feed on the internet. tesla isn't paying for that marketing - customers make it themselves, one selfie at a time.
front-of-house was supposed to be too social for robots.
turns out the robot is the most social thing in the building.
the honest catch nobody posts: it's minding drinks, not running the register. humans still take your order and cook your burger. today it's a mascot with hands. but mascots don't get software updates. this one does.
would you wait in line just to be served by a robot?
MercadoLibre $MELI Discussion
MELI is the first company which I have run through Buffett's 13-step evaluation where I do not feel comfortable forecasting future EPS growth. This is attributed to a: (i) highly complex business model spanning e-commerce and fintech operations; (ii) volatile EPS trajectory over the past ten years associated with major investment cycles eroding operating margins; and (ii) lack of forward EPS, revenue or margin guidance from company management.
NOTE: I am sure that management's current investment cycle will pay off, and that like 2021-2023 after the 2017-2020 investment cycle, MELI will see expanding operating margins and rapidly compounding EPS growth when its current investment cycle ends. What I cannot say is how long its current investment cycle will last and how extreme MELI's EPS growth will be when it concludes. If anyone can help me out with forecasting either of these variables it would be greatly appreciated, and I will revise my write-up accordingly.
Despite the absence of a price target, in this post I will still provide an overview of MELI's core businesses, its competitive landscape, its current investment cycle, the trajectory of EPS, revenue and operating margins over the last ten years, and my take on whether it should be valued on a multiple of its earnings or free cash flow.
Because MELI's business model is complex, I am re-arranging the steps to begin with an overview of MELI's core businesses before assessing its competitive positioning. I am also opting to include only Step One (Understanding Business), Two (Consumer Monopoly) and Four (Trailing Ten Year EPS Evaluation).
Step One - Do I understand MELI's business?
Not well enough to invest. As stated, MELI's business model is extremely complex. Below I attempt to provide an overview of the company's e-commerce and fintech operations.
E-Commerce
MELI operates the leading online commerce ecosystem in Latin America by gross merchandise volume, present in eighteen countries across South and Central America. The opportunity in this space is vast, as e-commerce penetration of total retail in Latin America lags the United States, the United Kingdom and China by a wide margin.
Within MELI's e-commerce ecosystem there are several business, including:
(1) The Mercado Libre Marketplace. This is a marketplace where third-party ("3P") sellers, who account for the majority of GMV, and MELI itself via its own inventory (confined to less than 10% of GMV), sell goods to buyers across Latin America. Categories span consumer electronics, apparel and beauty, home goods, automotive accessories, toys, books and entertainment, and consumer packaged goods, plus international supply from Chinese and U.S. sellers through cross-border trade.
(2) Mercado Libre Classifieds. A marketplace for higher value products, including vehicles, properties and services. Within this marketplace MELI charges placing fees only, not final value fees.
(3) Mercado Envíos. This is MELI's logistics arm, built around fulfilment centres, which account for more than half of shipments, plus cross-docking where MELI collects sold items directly from sellers or via MELI Places, a network of thousands of partner stores where sellers drop off items for injection into the network and buyers collect purchases and process returns. In addition to fulfillment centres and cross-docking, Mercado Envíos includes dedicated aircraft and trucks and thousands of last-mile delivery vans to transport goods sold on the Mercado Libre marketplace. The vast majority of MELI's transportation fleet is owned and operated by third-party carriers.
(4) Mercado Ads. MELI offers advertising on its marketplace to platform sellers and brands both on- and off-platform. Advertising revenues are bolstered by Mercado Play, a video-on-demand service launched by MELI in 2023 which licenses content from 3P studios under revenue-sharing agreements. Overall, advertising grew 73% YoY in Q1 2026, roughly four times the regional market, and MELI now ranks third in Latin American ad spend behind Google and Meta.
(5) Mi Página. The digital storefront solution letting sellers set up, manage and promote their own stores while running on MELI's logistics, advertising and payments rails.
Revenue within MELI's e-commerce segment has two drivers. The first is Service Sales, which comprise marketplace fees including: (a) final value fees; (b) flat fees on transactions below a certain merchandise value; (c) shipping fees net of third-party carrier costs where MELI acts as agent; (d) storage fees; (e) classifieds fees; (f) advertising fees; and (e) ancillary businesses. Second, Product Sales comprise first-party ("1P") merchandise sold from MELI's own inventory, plus related shipping.
Fintech
Mercado Pago operates in eight South American countries and is the leader in fintech monthly active users in Argentina, Chile and Mexico, and second in Brazil. It processes and settles every transaction on MELI's own marketplace in the countries where it operates, but the majority of its activity is now off-platform.
Mercado Pago began as an escrow mechanism for the MercadoLibre Marketplace. It then extended to online merchants outside of the marketplace, as management observed that individuals and micro, small and medium-sized enterprises in Latin America were underserved by incumbent banks and that a very large share of Latin American retail was still settling in cash. Today, Mercado Pago's digital accounts offer utility payments, mobile top-up, peer-to-peer transfers, pre-paid and debit cards, credit cards, merchant and consumer credit both on and off MELI's marketplace, insurance (extended warranties, personal accident, theft and damage), savings and investment products on idle balances, and a cryptocurrency buy-hold-sell feature in Brazil, Mexico and Chile.
Fintech revenue has three divers:
(1) Financial services and income. This stream includes commissions on off-platform payment processing and MELI's asset management product, instalment financing (whether MELI funds directly or sells the receivable), interest on cash and investments held as part of Mercado Pago activity including regulatory balances, net of interest passed through to Brazilian users on the asset management product, plus debit card commissions and insurtech fees.
(2) Credit revenues. Includes interest collected on merchant and consumer loans and advances, and interest and commissions on Mercado Pago credit card transactions.
(3) Fintech product sales. This includes mobile point-of-sale device hardware.
Step Two: Is MELI a Consumer Monopoly?
Under the taxonomy I set out in my framework write-up, the three consumer-monopoly categories are: (i) brand-name products; (ii) advertising agencies, publishers, social media and telecom networks; and (iii) businesses providing repetitive services that people and businesses are constantly in need of. MELI is a hybrid of the second and third.
Specifically, the company's e-commerce ecosystem belongs in category (ii). It is a toll bridge whereby Latin American sellers must pay MELI a fee on each sale made in the marketplace and in order to increase visibility on MELI's marketplace must pay an advertising fee.
Mercado Pago belongs in category (iii), given that payments, deposits and credit are needed repetitively and indefinitely. Additionally, switching costs to another financial provider, especially in the sparse setting of Latin America where individuals and businesses are underserved by traditional banks, are costly.
Switching costs are elevated by synergies between MELI's fintech and e-commerce ecosystems, as demonstrated by the company's Loyalty Programme. Specifically, this Loyalty Programme offers members free shipping and content from purchases made on the MercadoLibre Marketplace, along with better remuneration on their deposits, extra free installments on credit card purchases, and cashback on anything bought through Mercado Pago. The combination of benefits from MELI's e-commerce business and fintech ecosystem cannot be replicated by single-sided competitors, and encourages e-commerce members to become Mercado Pago customers. This phenomenon is supported by Brazilian credit card data cited by MELI management during its Q3 2025 Investor Presentation. Specifically, most of the 1.3 million cards issued in Brazil during the second quarter of 2025 went to existing marketplace buyers whom MELI had already observed for years.
Despite the competitive advantages associated with MELI's scale and synergies across its e-commerce and fintech ecosystems, the company's own risk disclosure states acknowledges that barriers to entry for large, well-established technology companies are relatively low, and that current and new competitors can launch sites at relatively low cost using commercially available software. Further, management has acknowledged that competition has intensified over recent years as local players grow and international players expand, mainly in Brazil and Mexico.
Rising competition has not yet weakened MELI's margins. Instead, margin erosion at MELI can be attributed to the company's current investment cycle, which is aimed at increasing market share in the e-commerce and credit card segments of Latin America. However, it is worth monitoring whether increasing competition results in margin erosion at MELI beyond its current investment cycle, as that would represent a more structural threat to MELI's EPS growth.
Step 4: MELI's Trailing 10-Year EPS Growth
As seen in the attached figure titled "MELI 2015 to Q1 2026 EPS Growth", MELI's EPS growth has compounded at a 32.3% rate over the past 10 years. This EPS growth has been entirely independent of share repurchases, with MELI's share count flat since increasing 15% between 2015 and 2022.
While strong, MELI's EPS growth has been volatile. A major investment cycle between 2017 and 2020 caused diluted EPS to turn negative before rapidly increasing to $37.69 in 2024. A new investment cycle initiated in 2024 caused MELI's EPS growth to grow only 4.5% in 2025 and on a trailing 12 month basis as of Q1 2026 MELI's EPS has contracted by 6.8%.
As seen in the attached figure titled "MELI Revenue and Operating Margins, 2021 to Q1 2026", MELI's operating margin has contracted each year since beginning its 2024 investment cycle. In 2024, operating margins contracted 190bps to 12.7%, followed by a 160bp contraction to 11.1% in 2025, and a 420bp contraction QoQ (~600bps YoY) in Q1 2026 to 6.9%. Management expects the company's current ~7% operating margin to remain relatively constant for the foreseeable future.
It is worth noting that MELI's investment cycle is driving high revenue growth, which has increased at a 38.3% CAGR since 2023. In Q1 2026, quarterly revenue was $8.85B, up 49.03% and representing MELI's fastest quarterly revenue growth since 2022.
The two central initiatives within MELI's ongoing investment cycle which have caused accelerating revenue growth and deteriorating operating margins are as follows:
(1) The Brazilian Free-Shipping Programme
In Q2 2025, MELI lowered its Brazilian free-shipping threshold from BRL 79 to BRL 19, representing the fourth cut from BRL 120 when free shipping launched in Brazil in 2017. Management identifies shipping costs as the largest friction suppressing e-commerce penetration in Latin America, and by removing this friction management is growing the entire Latin American e-commerce ecosystem along with MELI's share of it.
Every prior free-shipping threshold reduction produced an acceleration in GMV, total payment volume, purchase frequency and new users. This reduction has followed trend, with Brazilian item growth accelerating 34% in June 2025 (above Q2 2025 item growth of 26%) after the free-shipping reduction was implemented earlier that month. Group-wide in Q1 2026, unique buyer growth reached 32%, GMV reached 38% on a FX-neutral basis, and items sold increased 56%, more than double the pre-change rate.
Meanwhile, cost per shipment fell 17% in local currency as management converted products above BRL 19 to a two-to-four-day delivery timeframe rather than same- or next-day. CFO De los Santos explained that slow-ship items will act as a plug increasing transportation efficiencies for MELI's delivery fleet, such as a truck running from São Paulo to Rio for next-day delivery which must depart at 7 p.m. even if it at 75% capacity. Looking forward, management stated that the direction of shipping cost reduction should remain downwards but is not going to be linear.
(2) Credit Card Expansion
The second ambition of MELI's current investment cycle is to expand the number of individuals using a Mercado Pago credit card. CFO de los Santos refers to the benefits of credit card expansion as "principality", the phenomenon by which a Mercado Pago wallet is converted from a place customers transact into their primary financial relationship.
Management's credit card expansion efforts are working, as in Q1 2026 MELI's credit card TPV grew 90% YoY and its credit portfolio nearly doubled to $14.6 billion, bolstered by the issuance of 2.7 million credit cards in the quarter.
Increased credit card volume expands MELI revenues but comes with lower operating margins at the initial stage. Of the ~600bps of total YoY margin compression in Q1 2026, ~400bps was caused by credit provisioning alone. This is because expected-loss accounting registers the full life-of-loan reserve at origination, while interest income arrives over the following months, so a credit card book such as MELI's growing 87% YoY is constantly paying today for revenue it collects later. Additionally, MELI's extension of average personal loan duration in Brazil from five months to eight months means larger provisions and the assumption of greater early-repayment risk. Further, in order to incentivize new credit card users MELI is offering benefits such as lower spreads, reaching segments that are either more risky or require working with smaller spreads.
Conclusion
If MELI's current investment cycle continues to drive higher revenue growth and if operating margins expand once management slows its pace of investment, I expect MELI's EPS to grow rapidly. This is what occurred in the aftermath of MELI's 2017-2020 investment cycle. Specifically, operating margins increased from 6.2% in 2021 to 9.9% in 2022 to 14.6% in 2023 as the 2017 logistics cycle matured, and diluted EPS went from $1.67 to $9.53 to $19.46 across the same three years. The company's current investment cycle is based on a blueprint which MELI's management team has executed on before, which gives me confidence that it will eventually culminate in increasingly profitable growth.
However, due to: (i) a volatile EPS history; (ii) a complex business model; and (iii) the absence of management guidance on revenue growth or profitability, I cannot forecast when MELI's current investment cycle will end, or how quickly EPS growth will accelerate once it does.
At first glance, MELI's current valuation does not offer much cushion to fall back on if its current investment cycle continues for longer than anticipated or does not result in increased profitability.
At its current share price of roughly $1,800 and 50.7 million diluted shares, MELI's market cap is roughly $91B.
On trailing earnings of $37.89 per share that is a 47.5x P/E ratio, and about 44x forward. Against a multiline retail industry average P/E near 20x, the multiple assigned to MELI is pricing in a significant acceleration in EPS, which I am unable to forecast.
However, many analysts have pointed out that MELI should be valued on a cash-flow basis rather than on a multiple of company earnings. The most useful cash-flow-based valuation approach for MELI is a multiple of what Buffett refers to as Owner Cash — operating cash flow adjusted for the items that inflate it without belonging to shareholders.
As summarized by @Muzzlebuster in his article "The Single Greatest Mercado Libre Valuation Article Ever Written in the History of Human Kind", starting from an estimated $18.8B of 2026 operating cash flow, MELI's Owner Cash can be calculated by stripping out: (i) restricted customer funds held for regulatory reasons, roughly $7.5B, which MELI cannot lend, invest or distribute; (ii) card transaction settlement float, roughly $2.3B, held temporarily between collection and merchant payout; (iii) maintenance capital expenditure, proxied conservatively at full depreciation and amortization of roughly $1.1B; (iv) excess provisions, roughly $3.15B, being the gap between provisions booked and losses actually charged off, which exists only because the loan book is growing at close to 90%; and (v) a roughly $50m timing excess on the long-term retention plan.
That leaves approximately $4.7bn, or about $93 per share in Owner Cash, a multiple of 19.4x at $1,800. Owner Cash per share is rapidly expanding, up 37.4% between 2024 and 2025, and estimated to grow 47.6% this year.
19.4x Owner Cash is a far more reasonable multiple than ~45x earnings. I believe it is also a better way to examine a business that is deliberately suppressing earnings to fund growth, reflects the fact that MELI's reported free cash flow and EPS are both distorted outputs rather than measures of the business' earning power.
Given this attractive valuation and management's track record of exercising on investment cycles, I am tempted to allocate a small position to MELI even without being able to forecast its future EPS growth. I am sharing this post with the hope that the FinX community will be able to help me with forecasting future EPS growth and providing a timeline on MELI's investment cycle.
Thoughts on $META earnings
Very strong quarter despite the headlines. Revenue grew 28% to over $60b, impressions increased 14%, average price per ad increased 12%, and the advertising business looks healthy.
The issue wasn’t demand, it was spending. Costs jumped 55%, operating margin fell from 43% to 31%, that caused eps to miss by 14%, and capex guidance was raised again, albeit slightly, to an astonishing $140b
To me, this is the same story we’re seeing across its counterparts. The smartest people in the world are investing the largest amounts of capital in history, making enormous promises about AI. They may ultimately be right and create trillions of dollars of value. But they’re also asking investors to accept an extraordinary amount of uncertainty.
Everyone thinks they’re investing in AI, they’re really not. They’re investing in management’s ability to allocate hundreds of billions of dollars intelligently.
The market isn’t saying Zuckerberg is wrong. It’s saying, “show me the returns.” Spending money is easy, earning exceptional returns on that money is the hard part. History is full of companies that spent fortunes chasing the future and destroyed shareholder value along the way.
Every hyperscaler is spending because every hyperscaler feels like it has to spend. That doesn’t automatically mean every dollar invested will earn an exceptional return. Every dollar spent on AI is also a dollar that can’t be used for buybacks, acquisitions, dividends, or other opportunities. Capital always has an opportunity cost.
I’ve been thinking a lot about this so called AI arms race. If I had to bet, I’d probably bet it works for $GOOG, $AMZN, etc. But there’s a subtle difference between investing because the opportunity is extraordinary and investing because not investing may be even more dangerous. I think the hyperscalers are in the second camp. They may still earn phenomenal returns, but they don’t really have the luxury of saying no. Companies like $AAPL do. That’s a very different capital allocation decision and predicament to be in.
Ultimately, I don’t care what $meta spends this quarter. I care what these investments earn over the next decade. If this AI buildout generates extraordinary returns on capital, today’s spending will look brilliant. If it doesn’t, investors will eventually wonder why hundreds of billions of dollars were deployed in the first place and punish the stock.
The market doesn’t hate spending, it hates uncertainty. Until the returns become more visible, this is less a debate about AI and much more a debate about capital allocation. That’s what will ultimately determine shareholder value. 🌹
Mid-Year Portfolio Review
Typically share this with subs but it’s been awhile since i’ve discussed my portfolio and positioning in public so I thought I would give a quick update:
$SE: 18.1%
$DLO: 11.9%
$MELI: 10.4%
$AXON: 7.7%
$FICO: 6.7%
$GRAB: 6.3%
$RBRK: 5.7%
$TSM: 4.9%
$AMZN: 4.9%
$CRWV: 3.5%
$GLXY: 2.8%
$OUST: 2.6%
$NBIS: 1.8%
$HOOD: 1.0%
Cash: 11.7%
*Numbers are as of 30th June 2026
$SE:
I view Sea to be among the best asymmetric risk/reward plays in the market today. I truly believe this will be one of the most valuable businesses in SEA within the next decade and the path to a trillion is probable. Despite the narrative surrounding the business, the fundamentals remain incredibly strong and I see no reason to doubt the management team’s ability to execute.
$DLO:
Payments businesses have been hit badly in recent years and I think this might be one of the most misunderstood businesses in public markets. dLocal does the unglamorous work for global merchants. TPV is compounding well above 40% and while take rates continue to be the primary discussion factor, at roughly 16x forward earnings for a capital-light business earning 35%+ ROE, the market is pricing structural decline for a company still accelerating. This is my highest conviction on a risk/reward basis.
$MELI:
$90B for arguably the most important business in LatAm is cheap. The primary reason for the stock being in a 30% drawdown is largely due to margin concerns. However, the compression is entirely deliberate. Revenue is growing north of 40% and dominance has no signs of wavering.
$AXON:
This is one of the highest quality businesses in the world with a huge moat and installed base that enables it to cross-sell new products and services. The business has grown at 25% for over 10 years now and there are no signs of stopping with management guiding for 32% revenue growth this year.
$FICO:
This basically comes down to whether you think their moat is regulatory or network effects/switching cost. I believe it is the latter, and that with most things, nothing changes. VantageScore has been around for 19 years. Nothing will change unless the product is 10x better. Independent studies have shown that VantageScore 4.0 is only “marginally” better than 20-year-old Classic FICO at the low end, with ROC curves nearly overlapping. Glad to own this extremely high quality business.
$GRAB:
Been a very tough 8-10 months for the stock. Much of this has been out of their control. Oil prices have affected demand and supply of mobility/deliveries. The overhang of Indonesia’s regulatory tariffs and Gojek founder’s corruption case has been a net negative for the business. My thesis remains intact as a 20% compounder over the next decade. However, I will not be adding until the stock performs.
$RBRK
Seeing this get sold off into the $40s during the SaaSpocalypse was a massive gift. I made it a top 3 position in March and have since taken huge profits off it, the most profitable “trade” this year. I remain extremely bullish on the cyber resilience market and believe AI will expand the addressable market, not disrupt it.
$TSM:
Nothing much to add here, got in during March 2025 and sold nothing since. One of my easiest holds and clear beneficiary of the AI wave.
$CRWV:
My newest position. If you believe data centers are one of the key components of the next decade in AI, this is almost a lay-up at these prices.
$GLXY:
In my view, the most asymmetric risk/reward play in the data center HPC space due to the nature of its crypto business. Helios now operational and revenue generating.
$OUST:
Started a position in the high $30s and added in low $40s. Not a super high conviction investment but I think it’s one of the best physical AI plays. The ceiling here is supply not demand.
$NBIS:
Best % returner this year, got in at $80s in Jan this year, took profits in mid to high $200s and would consider adding again soon. Covered heavily on X so won’t comment much.
Meet "Tawan" (ตะวัน means “The sun” 🌞)
20-min Demo
AI Animated Film experiment (full feature is 90 mins)
With a $500 budget and 1.5 months of work, this story is based on my original idea from 2010... and today, the technology is finally ready to bring it to life.
AI showed me that solo creators can now manage an entire animation workflow. No need to pitch to big studios—you can fund yourself and bring your own stories to life. Without AI, this would still be stuck in my head.
(Of course, the quality is still far from high-budget films from big studios that have 300-600 people behind them, but I think the gap will close step by step in the future.)
This animated created by Seedance 2.0 on @dreamina_ai and @kinovi_ai and opening scene by @midjourney - thank you for watching.
Super excited to announce seven new world-class MAI models today. They represent what we consider a new era in AI designed to keep you in control and on the frontier.
First is our text foundation model, MAI-Thinking-1, exceptionally strong on reasoning and SWE tasks.
- It’s a 35B active parameter MoE with a 256K context window. Independent human raters on Surge prefer it for overall quality in blind side-by-sides versus Sonnet 4.6, and it’s achieved 97% on AIME 2025, the key measure of its general-purpose reasoning abilities.
- It's at 53% on SWE Bench Pro, placing it right alongside Opus 4.6 on one of the toughest coding benchmarks.
- And since we co-designed our models with our own silicon, MAI-Thinking-1 is optimized on our MAIA 200 chip. Benchmarking head-to-head against the GB200, we see 30% better performance per dollar as well as a 1.4x performance-per-watt gain when running our MAI models on the MAIA 200 end-to-end.
Next is MAI-Image-2.5 and its Flash variant. Two super strong models now at #2 on the leaderboards, surpassing the score of Nano Banana 2 on image editing.
Last for now is MAI-Code-1-Flash, our new inference efficient coding model, especially tuned for VS Code and GitHub Copilot CLI.
- Code-1-Flash achieves 51% on SWE Bench Pro, despite having just 5B parameters, putting it closer to Haiku in size but cheaper in cost.
All of this is the foundation for Microsoft Frontier Tuning. It lets you customize our models to create custom, company-specific agents that only you control. You can make our model, your model. Your data. Your agents. Your moat.
Early adopters are already seeing a difference. When we tuned our models for McKinsey’s tasks, MAI delivered the highest win rate, outperforming GPT-5.5 on quality, while being 10x lower on cost.
Also really excited to be collaborating with the amazing team at Mayo Clinic to jointly train a new frontier AI model for healthcare.
Our announcements today mark another milestone on the road to humanist superintelligence. You can learn more and about our other new models in our latest blog: https://t.co/v65eop5Ixq
Jeff Bezos: “I once asked Warren Buffett, why don’t more people copy your investment strategy? It’s not that difficult to understand in principle. And he said, ‘Oh, Jeff, that’s easy. My approach is a get-rich-slowly scheme.’ And people don’t like those.”
“If you can think in terms of seven years instead of three years, and you can defer gratification and think long term, that will give you a head start against all of your competitors, because most people can’t do that.”
Concerned about the wrong thing here by anchoring on a number like $375b because that’s not what determines whether this works or not. Let’s say $TSLA does “only” $300b instead. That sounds like a big miss on paper, but in reality it might not matter at all.
If these businesses actually get built and scaled, robotaxi, energy, AI, robotics, then the outcome is driven by the engine, not the estimate. Once something like that starts compounding, the difference between $300b and $375b becomes noise. What matters is that new profit pools exist and keep expanding over time.
This is where most investors get tripped up. They build very precise spreadsheets and think precision equals accuracy. In reality, it often just gives you a false sense of control over something that is inherently unpredictable. I like what Nick Sleep calls destination analysis, if things are moving in the right direction and the opportunity is vast and preferably keeps expanding, you don’t need to overthink every input.
The biggest winners are almost never fully modeled in advance. No one modeled AWS in $AMZN early on. No one modeled ads at $NFLX years ago, they were anti ads for over a decade. No one thought Mercado Pago would become what it is today when people first bought $MELI.
So the real question is not whether $TSLA hits a specific number in 2031. The real question is whether they are building something that keeps creating new opportunities that don’t even exist in your model today because that’s where the outsized returns come from.
A $75b “miss” sounds massive. But you’re really just debating the difference between two guesses about a future no one actually understands. That’s the illusion, it feels precise, but it’s not.
Buffett talked about this with See’s Candies. He almost walked away because he didn’t want to pay more than $25m. In hindsight, they could have paid $50m and it still would have been an incredible investment. The extra price felt huge at the time, but it was irrelevant compared to what the business became.
Time is what fixes that mistake. If a company can keep compounding, time turns what looked like overpaying into underpaying. But if the business doesn’t have that engine, no price is cheap enough to save you.
Most people think risk is missing the downside. In reality, the bigger risk is being right about the company and still missing the outcome because you anchored to a number that didn’t matter. That’s how you end up watching a great business run without you.
If $TSLA executes, no one will care whether your model said $300b or $375b. And if they don’t execute, your model won’t save you anyway.
🌹
FIRE is terrible. Don’t waste your 20s and 30s saving every single penny and never having any experiences.
Foster relationships now. Go out on dates. Go on road trips and camp outs. Try out a cruise, you may even enjoy it. Go have experiences with friends.
My oldest brother that I talk to frequently is a surgeon, believe me, there are no guarantees in life. You have no clue how your health will hold up over time.
It’s crazy to me that a group of people will throw away their 20s and 30s in hope that there’s some magical happy point in their 40s. It’s very misguided.
Thoughts on $TSLA
$TSLA is clearly no longer just a car company and this report makes that obvious. Musk is basically walking on water right now. The fact that he’s building autos, energy, AI, robotics, and a Robotaxi network all at once while still producing serious free cash flow is something you almost never see in business. And yeah, people are going to point to the PE and say it’s expensive, but that completely misses what’s actually going on here. Quite frankly, that argument is moronic and shows a lack of understanding of what’s being built.
Most companies struggle to do one thing well. $TSLA is trying to solve manufacturing, energy, autonomy, AI infrastructure, robotics, and fleet economics all at once. And somehow it’s still generating billions in free cash flow. That’s not normal, it’s not even comparable.
They’re scaling the core auto business, building energy, and at the same time pouring capital into AI and autonomy. This is an entirely different economic model. The bet is simple, give up near term margins to build something that looks more like software later.
Revenue was $22b, up 16%. That came from more deliveries, services up 42%, better pricing, and more FSD revenue. But not all of it was perfect. Energy was down and regulatory credits fell, so the growth was still good, but mixed.
Profitability improved, but let’s not pretend this looks like a software company yet. Operating income was about $900m with a 4.2% margin and $500m in net income. Margins are moving up, but they’re still low relative to what this business could become.
The reason is obvious. They’re spending heavily on AI, R&D, and infrastructure, and SBC is still elevated. That’s their strategy, and they’re deliberately compressing margins today to build something much bigger over time.
Cash flow is what matters most because they generated roughly $4b in operating cash flow and $1.4b in free cash flow. That’s very impressive, but they’re also spending $2.5b on capex and invested $2b into SpaceX. This is still a capital heavy business today, even if the long term goal is to become asset light.
The balance sheet is what makes all of this possible. Around $45b in cash and a massive asset base. That gives them time, and time is everything when you’re trying to build something this ambitious.
Every dollar $TSLA generates isn’t being returned, it’s being redeployed into AI compute, factories, and vertical integration. This isn’t a company optimizing earnings, it’s a company allocating capital into optionality. The question isn’t margins today, it’s what those dollars earn over the next decade.
This quarter, operations were a bit more mixed. Deliveries were 358k, up 6%, while production was 408k, up 13%. Inventory moved up to 27 days. That’s not alarming, but it tells you demand isn’t running away from them anymore. They are no longer supply constrained.
FSD is my favorite part of the report. Subscriptions hit 1.28m, up 51%. That’s a huge deal because it is an early signal of the shift from selling a car once to monetizing it over time. If that works, the entire model changes drastically.
At 1.28m subscribers, even modest pricing starts to matter big time. At $100 a month, that’s roughly $1.5b a year. Scale that across tens of millions of vehicles and the numbers start to get enormous. This is how a car company slowly becomes a software company.
Energy had a weak quarter with storage down 15%, but I wouldn’t overreact. This business is lumpy and new capacity is coming online. Longer term, it’s still a meaningful growth driver.
Meanwhile, the infrastructure keeps getting stronger. The Supercharger network is growing close to 20% with over 8,400 stations and roughly 80,000 connectors. People don’t talk about this enough, but it’s a huge advantage.
1/2 👇