Sections 4-14 are for paid subscribers. If you'd like to read about SoFi credit market information, Rubrik's Agent Cloud Release, a brief Lemonade vs. Root comparison, Trade Desk competition news and more, upgrade below. Next week, we will be publishing more than 10 earnings reviews on SoFi, Meta, Apple, Starbucks, Microsoft, PayPal, and more.

Upgrade

In case you missed it:

Anthropic and Google agreed to a new cloud deal. It includes up to 1 million Google AI chips (Tensor Processing Units; TPUs) deployed and over a gigawatt of compute by 2026. In total, it’s worth “tens of billions.” There are going to be many more cloud deals from this AI darling. I expect most of them to go to Alphabet and Amazon – the two providers with large equity stakes. Anthropic will not use a single vendor. That would be ill-advised. Multi-cloud adoption inherently creates a layer of resilience that modern companies flock to. So while I think this is a positive for Google, I don't think it's  super surprising. We all saw Anthropic’s $23B revenue guide for next year. That will require a lot more capacity from Google and Amazon coming online. These deals are all but inevitable, and I don’t see one of them going to Google over Amazon as a negative for AWS. They will both continue to get their fair share of Anthropic workload demand. That's why they each spent so much money acquiring a stake (and boosting it thereafter).

This past week, as already announced and expected, OpenAI released their web browser. This, in my mind, should be treated as noise. I worry a lot more about OpenAI taking search market share than browser-level market share. Meanwhile, Gemini, AI Overviews and even core search market share trends all look very good for Google.

There’s also an Alphabet rumor circulating of a $120B-$125B CapEx budget for 2026. That's a very big expense compared to a large $85B CapEx number this year. Alphabet doesn’t spend on AI CapEx unless they see near-term demand for more capacity  – such as the new Anthropic cloud agreement. If this rumor is true, it indicates their belief that demand will remain massive for their cloud and AI services in 2026. That will weigh materially on FCF generation in the near-term, but I view this as money well spent. If they overbuild, there are so many other Google products that can easily use the compute and allow them to avoid waste. And over the long haul, I expect ownership of this scaled infrastructure to be convincingly additive to overall growth and profitability. 

Uber is participating in a $375M AVRide investment. Nebius (parent company) is also participating. Uber and AVRide have been close partners since last year, with a Dallas AV deployment planned for this year. They already have delivery robots in 3 cities as well. This will allow them to accelerate expansion and fleet buildouts while giving Uber steady access to another source of supply. The mobility giant did something similar with Lucid and I expect more of these deals. Uber can use its balance sheet to accelerate this industry’s maturation. And it can help a lot more players secure needed funding to get to the finish line. This not only gives Uber guaranteed access to more supply, but helps drive more competition. That’s great news for the demand aggregator in this equation.

Uber has told us they’re confident in monetizing these investments down the road when their capital is not so pressingly needed. For now, this is the right decision and doesn’t prevent them from rapidly buying back shares or expanding the product suite. That’s the beauty of massive cash flows.

The PrimaLend bankruptcy is hitting lenders today. This is a subprime auto lender with a (per Bloomberg) buy here, pay here auto loan product. Subprime auto is where we've seen all of the credit issues thus far. SoFi is on the opposite side of the credit spectrum. Like Tricolor, this company (many believe) had a large book of business with undocumented immigrants. Tighter regulation there directly and specifically hurts these lenders. I do not expect this to translate into SoFi credit weakness, considering their niche, recent company commentary (at the Goldman conference we covered & more recently), and Big Bank credit data.

Lending Club's credit data and commentary bode well for SoFi's quarterly results coming next week.  They beat their guidance for originations and pre-provision revenue. Revenue beat sell-side estimates by 4%, while EPS beat $0.31 estimates by $0.06 or 21%. Q4 estimates did fall a bit following the guidance, but annual estimates rose thanks to Q3 strength more than making up for that modest weakness.

Fair value markings were strong. Borrower demand was solid. Capital market credit appetite was robust to a point of materially improving pricing dynamics for LC. That had a lot to do with insatiable demand from insurance agencies. Delinquencies remained healthy, which should mean capital market demand will as well. That forward-looking indicator was joined by strength in the lagging indicator -- net charge-offs (NCO). NCO rate was 2.9% vs. 5.4% Y/Y and 3.0% Q/Q. Encouraging.

Digging a bit deeper, credit loss allowance did rise, but that was related to a day-one provision based on a decision to retain more held-for-investment loans. There's some weird nuance in terms of it shifting to a structured certificates program and different held-for-investment accounting formats (amortized cost instead of fair value), which is why this happened while overall fair value fell. Not concerning as a read on overall credit health. Just an ongoing accounting shift. Despite this headwind, gross provisions for credit losses still fell Y/Y due to strong credit performance.

"Credit performance remains excellent. We continue to outperform the industry with delinquency and charge-off metrics in line with or better than our expectations... we continue to see strong performance across our vintages. I would highlight that the net charge-off ratio also continues to benefit from the more recent vintages we've added to the balance sheet. We expect the charge-off ratio to revert upwards to more normalized levels as these vintages mature. These anticipated dynamics are already factored into our provision." – CFO Andrew LaBenne

"

As we look ahead, the business enters the fourth quarter with significant momentum. Loan investor demand remains strong. Loan sales pricing continues to trend higher, and our product and marketing initiatives are driving high-quality volume growth

." -- CFO Andrew LaBenne

While we can't look at these results and conclude SoFi's will also be strong, this is still a relevant and positive hint. And furthermore, the performance divergence between the two that we've seen in the past has been LC showing signs of weakness while SoFi, with its ultra-prime and hyper-conservative credit niche, has fared better. So? If LC is doing this well, there's a great chance next week's report will be excellent. That's my expectation. Who knows how the stock responds after a year of fantastic gains. 

UBS lowered their price target from $500 to $450 today and maintained a buy rating. They're citing the same things other bearish notes have. According to their data (and many other sources), daily active user (DAU) growth for Q3 will likely miss guidance (37% Y/Y growth UBS forecast vs. 40% Duolingo guide). This is due to the social media blowback that is taking some time to normalize in terms of viral content frequency. Edgy content is how they grow. They've had to pause that while recovering from negative sentiment stemming from von Ahn's AI-first announcement. UBS also discussed Zaria Parvez leaving as a concern and AI competition as well.

None of this changes my view of the company (or theirs considering the buy rating). Q3 could likely come in weak, but I expect things to be business as usual thereafter as they do lean back into their bread-and-butter social marketing engine. As I've said before, this company is far bigger than any single talented marketer. And we've already seen other sell-siders call out social data improvements in recent weeks.

A weak Q3 is probably coming. That will not not lead me to grow bearish on this company -- especially if the Q4 guide and 2026 commentary are as resilient as I expect. I think language learning demand will remain healthy for the foreseeable future. Chatbots do not replace the core use cases for learning a language as  we so frequently discuss. Duolingo is unmatched in combining productive, competitive and fun learning.

I believe Duolingo will continue rounding out its product suite and become the leading digital education gamifier (definitely not just for language). Expansion for every other category is going well and that should continue. Succeeding here would unlock a $5T+ global education TAM that DUOL is positioned to win its fair share of. It would mean language learning is just one of many large financial drivers for the firm. Winning that at a $14B market cap leaves ample room left for a stock already trading near 30x forward FCF. There does not need to be more multiple contraction here for a long time if they execute. The stock can (I think) enjoy all of the 30%+ profit compounding coming its way.

Just keep executing. Just keep split testing like mad men. Just keep delivering a better product that means you don't have to spend a fortune on marketing. Just keep successfully entering new subjects. Just get back to your typical social media ways. And? Sentiment (I think) will take care of itself. I envision that happening as we move into 2026, again with Q3 probably being a tad underwhelming in the meantime.

Hello prediction markets. DraftKings seemingly got the green light from regulators to finally act here. They're buying Railbird to formally enter the space. They will launch "DraftKings Predictions" in the coming months, with "event contracts on finance, culture and entertainment." I love this decision. It gives them a foot in the door to this quickly-growing space without upsetting regulators. Importantly, threading that needle required them to forgo including sporting contracts in the initial launch. That means these won’t be offered in the states where DKNG doesn’t have an operating sportsbook, which would have been the largest source of incremental growth from this release. The lack of sporting event options today doesn’t mean this will be permanent. For now, it will mean this product is competing with a hand tied behind its back. Better than nothing, and a first step towards opening a full menu of prediction markets in states where it doesn’t have a sports betting license. Interestingly, Polymarket will serve as the main clearinghouse for this product, with DKNG boasting “flexibility with connections to multiple exchanges to enable one of the broadest suites of markets.” Just no sports (yet).

Regulatory certainty is what we want here. That will mean this is all allowed and DKNG can expand accordingly. Or? It will mean prediction markets can’t offer sports, which will eliminate that competitive threat while still allowing DKNG to expand into these prediction market categories. The bad outcome is this stretching out for years without anyone knowing what is permissible. Under that outcome, prediction markets will continue the “ask for forgiveness not permission” mantra and DraftKings will be forced to wait. They can’t ruin the relationships they’ve spent years building with state regulators.

Bad week for volume in New York. Growth came in at 0% Y/Y. This is heavily influenced by weekly event timing, but still not a positive. With the prediction market debate raging on, this will be fixated on and will add to short-term negative sentiment. This is not concerning in my mind unless it becomes a several-week trend, considering how good data has been for the last few months. If that trend convincingly turns negative, my bullishness will wane.

Mercado Libre announced an interesting new partnership today with Casas Bahia in Brazil. This is a specialty retailer with ~1,000 stores in the nation, a scaled fulfillment network and a few percent of Brazilian e-commerce market share as well.

While MELI is a pure generalist in terms of assortment focus, Casas Bahia has a tight focus on home appliances, furniture and electronics (including very large items like TVs and freezers). MELI has been pushing hard to improve assortment in these three categories, with their Brazil market share just 25% of what it is for their strong segments. This partnership should help a lot.

One of the main holdups for MELI accelerating growth within these often bulky categories was delivery complexity. And now?  MELI will be able to tap into Casas Bahia’s network while gaining access to a lot more assortment and avoiding the headache.  

I could see MELI wanting to own this business outright. It’s a 70-year-old retailer that’s growing slowly and could enjoy a vast acceleration in demand if under MELI’s wing. MELI would gain new fulfillment capabilities, strength in under-indexing categories and a large base of locations to handle last-mile delivery for a plethora of goods. To them, if this goes well enough, that may be worth the challenge of operating this part of their logistics network, as MELI has not shied away from doing hard things in the past. I’m speculating.

On Monday, AWS was down for a few hours. That’s obviously not ideal. The issue was fixed very quickly & I don’t expect this to materially impact results. We saw CrowdStrike shut parts of the entire world down for days last year and came out fine. This is not remotely as bad as that and AWS is a sticky platform too.

Amazon is reportedly looking to forgo adding 600,000 fulfillment-related jobs over the next 8 years. That equates to 40% of their entire current workforce and is thanks to aggressive robotics implementation.

This will be positive for margins. Its non-AWS revenue last quarter boasted a 6.6% operating margin with a base of nearly $140B in revenue. Every single point of margin expansion is worth $1.4B in quarterly incremental EBIT, which is 7% of their total EBIT generation last quarter. Massive base of revenue… with razor-thin margins… shedding a large portion of a gigantic cost.

That’s a wonderful setup for this segment over the next several years. While Amazon is considered an AI laggard today (until Anthropic capacity starts coming online), robotics and AI are tied at the hip… and Amazon is a clear leader in leveraging this disruptive technology.

The rapid source of leverage will leave Amazon with an enviable decision to make: They can choose to let that added profit flow down the income statement directly. Or? They can use this to elongate their e-commerce marginal cost lead, amplify their economy of scale, juice their volume growth, take more market share, establish even more dominance and harvest more margin down the road.

Rubrik launched their Agent Cloud offering this week. It’s built on Rubrik Security Cloud (RSC) and (I think) has a great chance to resonate with clients. The purpose of the debut is clear. Agents are exploding in size and creating a massive new autonomous machine asset class. These machines race around the digital ether to collect information for task completion and openly reason on how best to complete those tasks. That’s very exciting. It can vastly accelerate productivity while cutting costs and eliminating tedious work from day-to-day schedules. But? It also creates a large new security risk. These autonomous assets aren’t perfect. They make mistakes and reason incorrectly just like people do. And furthermore, adversaries are openly targeting them, knowing a successful hack can lead to a much larger, more persuasive and more damaging breach. These agents move very quickly and touch many parts of the online world. That’s immensely valuable for productivity, but also means corporations must concede a bigger security risk.

Agent Cloud is meant to eliminate that security concession. It’s a way for companies to confidently embrace this disruptive asset class, while enjoying its profound positives and minimizing the negatives.

Specifically, Agent Cloud includes three products. First, Agent Monitor flags and organizes infrastructure-based and platform-based agents across public cloud environments. This observability product provides a bird’s-eye view of all agentic assets in circulation, fostering a more complete idea of agentic operations. Knowledge is power… but knowledge with actionable work stemming from it is even more powerful. That’s where the other two products help.

Secondly, Agent Govern is what actually tracks agent performance. It uncovers which parts of prompts are potentially leading to suboptimal work, with real-time tools to tweak those prompts and broader policies. This is the agentic performance optimizer, which tightly integrates with popular identity directories. That’s so important, as it means companies can implement the same human-based access guardrails they have for machine-based assets. They can use that experience rather than having to learn brand new ways to set ID policies. All of this again minimizes the tradeoff between agentic adoption and security headache.

The third product in this release is perhaps the most exciting. It’s called Agent Remediate. Many companies have great observability and configuration tools for cloud and agent-based assets. It’s far more rare for them to have modules that can actively undo agent-based blunders. Rubrik leverages the “Agent Rewind” capability that it debuted in August (thanks to its Predibase purchase) to make this possible. With it, RBRK can rapidly determine which pieces of a company are vulnerable to an agentic error (called a “blast radius”). It can then use its immutable data backups to revert to the most recent, pre-error environment. I love seeing them applying their bread-and-butter product to new categories. Agent Remediate expedites time to normalization and diminishes the amount of a digital estate that’s affected. Both of these things lower customer costs – especially considering this happens without any “downtime or data loss.” Rubrik can proactively find and fix agentic errors before they become a catastrophe. 

Going back to Predibase for a moment. The entire point of that purchase was to help Rubrik accelerate enterprise AI adoption. They call this acceleration process “AI enablement.” indirectly benefits from this enablement via the large boost in overall company assets that need protecting. Their product suite was already well-positioned for that before this launch. With this release and Predibase’s low-code agent and model customization tools, it is quickly rounding out exciting capabilities to directly harvest financial value from AI enablement.

I got the “why Lemonade over Root” question in the Discord during the week and wanted to share my answer here.

Root is my second favorite name in the next-gen insurance space. I think it’s a good company and think positively of it. It’s way more mature than LMND in terms of margins. Already at a 7% EBITDA margin and LMND won't get to positive EBITDA until late next year. At the same time, LMND's gross margin is already a full 7 points better than Root. And that’s related to the same reasons that explain why I prefer LMND. It’s not because LMND underwriting is more mature. It isn’t. Their loss ratios are higher than Root’s right now because Lemonade is expanding into new product categories so successfully. Newer product loss rates take time to mature and improve. This GPM advantage is because of how much more success they’re having in cross-selling plans to customers and expanding to new product categories. Those customers come with $0 in CAC, so they’re a powerful tailwind for gross margin. That’s why their LTV/CAC is over 3x and rising. I’m more confident in Lemonade’s continued cross-selling than I am in Root’s ability to do so. Lemonade has been convincingly better in this regard so far.

That’s why Lemonade’s revenue CAGR is expected to accelerate over 30%+ Y/Y (even when ignoring the help from ceding a lower share of premiums). Root’s growth rate beyond 2025 is expected to be around 10% Y/Y. Furthermore, Lemonade’s source of more efficient growth is also why Lemonade’s EBITDA margin is currently expected to surpass Root’s some time in 2028 (always uncertain with multi-year forecasts). All of that makes me more eager to pay a cheap gross profit multiple for Lemonade than an extremely cheap gross profit multiple for Root.

And finally, while there’s nothing wrong with Root leadership, I think Lemonade is run by stars. They’ve met or exceeded their promises for 3 years. They’ve stuck to (or moved up) their path to profitability despite being a young company where multi-year targets are a crapshoot. I talk about it in almost every review – they are somehow a margin puppet master despite being a decade old. They could be profitable right now if they wanted to be. They’ve made me very comfortable with trusting their words. They’ve shown a clear ability to flatline fixed costs while revenue accelerates, which is why leverage is currently so strong and expected to remain that way. They responsibly pulled back amid high inflation and accepted lower growth while they caught up on regulatory filings. That was the right decision and now it is time for sharp, margin-accretive acceleration to kick into overdrive.

Another noisy article from Ad Week right as we were getting ready for our weekends. Amazon is offering 4-6 week ad platform trials to clients of competing vendors including The Trade Desk. I don't see this as a reason to panic about this company. I don’t think a few weeks of a free sample will materially disrupt the humungous multi-year joint business partnerships (JBPs) that power a large and growing piece of TTD’s financial engine. Those relationships are a lot deeper than that, as they’re based on long-term value creation, rather than a quick deal. They will go with the vendor that can durably deliver maximum return on ad spend (ROAS), not the vendor that has the lowest fees for a month. While it’s likely that the promotion entices some clients to experiment with Amazon, I am optimistic in TTD’s ability to flex its targeting muscles and sustainably compete.

I am confidently sticking with this name. Their platform is too valuable. Their demonstrated impact on targeting and ROAS is too sharp. Their conflict-of-interest-free niche is too compelling. Their opportunities across streaming, audio and retail are too sizable. I am not growing sour on a company because Ad Week published their 1000th negative article on it this quarter. We'll see how Q3 looks. That will go a long way in confirming whether or not I'm right. I was tempted to add today, but I've already done so recently, so I'm forcing myself to be patient.

As I've said over and over again... this is a massive industry where no one player will take the entire pie.  TTD taking a large slice in the years to come is a bet I’m comfortable making – regardless of a competitor running a promotion for a little while.

Meta is cutting 600 AI-related jobs. When they hired as aggressively as they did for several months, this became almost inevitable. There has been rapid change within that team in 2025. That rapid change will invariably mean the job is no longer a good fit for some people.

Lemonade co-CEO Daniel Schreiber gave a short interview this week. I listened, but it really just functioned to go over the structural advantages we discuss constantly. Some shared that 50% of their car insurance sales are from existing customers, but we already knew that. If you’d like to listen to it, you can do so here.

Oracle is taking out close to $40B in fresh debt to fund data center buildouts. Most of this infrastructure has been funded by free cash and cash on hand to date. That seems to be changing. This should add fuel to the fire for now. It also adds a sense of urgency in terms of all these investments actually netting positive net present value.

UBS thinks Starbucks margins will remain challenged this quarter (along with everyone else). This is fully expected.

Markets finally started to show signs of cooling off this week, before some headlines and positive inflation data quickly heated things back up once more. I did not get a chance to do any dip buying and I am staying patient for now. Any kind of meaningful correction (especially for the high fliers with no fundamental support) would be healthy in my mind. It’s time for me to do nothing and see if that will play out. I remain 92% invested, which, to me, strikes a good balance between remaining flexible and able to take advantage of any future pain, while not getting too cute with market timing. It’s not time for me to deploy the cash pile I’ve built this year just because things up 200% year-to-date (based mainly on multiple expansion) fell 10%. I've done the risk management work via raising cash and moving exposure to more boring names. It’s time to be patient and demand better deals before accumulating more shares. I’m not seeing those deals yet, as the small doses of volatility we’ve seen have been mild and short-lived. That could always change in a hurry. We shall see. For now, I am doing nothing. We have been up and to the right since April. Multiples have exploded higher. I do not think it's prudent for me to immediately and aggressively buy dips. 

It was another very light week of economic data with the ongoing government shutdown. We did, however, get some important data on Friday. The consumer price index (CPI) for September came in at 0.3% M/M vs. 0.4% expected and 0.4% last month. Similarly, the 0.2% M/M core CPI for September was 10 bps cooler than expected and 10 bps cooler than last month. 

From an output perspective, things also looked good. The Manufacturing Purchasing Managers Index (PMI) for October was 52.2 vs. 51.9 expected and 52 last month. The Services PMI was 55.2 vs. 53.5 expected and 54.2 last month. Finally, existing home sales for September were 4.06M as expected.