Table of Contents:

My latest portfolio & performance vs. the S&P 500 can be found here. Earnings season coverage gets rolling next week with big banks.

Google launched a product called Gemini Enterprise this week. The release builds on Alphabet’s positioning as another highly capable full-stack AI player. It has world-class models… world-class distribution… wonderfully efficient chips and data centers... elite research talent… AND an army of organized, customizable agents poised to extract more automation-inspired efficiency across an entire enterprise. And intuitively, as another way to leverage its broader product suite, these agents will be powered by Gemini and equipped with conversational querying for simple access.

The value creation will reach every facet of a company's day-to-day operations to make sure their AI investments are creating enough value and time savings to power real ROI. The suite includes a "workbench," which seamlessly builds complex agentic workflows in a no-code manner and a large library of agent templates for companies not working to build their own. All of this will be under one roof, ensuring superior interoperability, broad integration flexibility for customers.  It also provides rapid product improvement informed by a larger sum of relevant data than others. Remember when these guys were falling behind? Me neither.

As we talk about a lot, there are 3 layers of the AI opportunity -- infrastructure, models and apps/agents. Alphabet's TPU/GPU hybrid infrastructure (they say) is the most efficient and productive on the planet. Their Gemini models rank at the top of leaderboards with a new batch coming soon. They have 6 existing apps with 2B+ active users. This is how they can use these great assets to ensure they're a central piece of the app/agent layer. Monetization within that layer has really only been reached by a few companies such as Palantir, Microsoft and ServiceNow. This will pin Alphabet in more direct competition with the agent-based products that those companies are selling. Gemini Enterprise is the search giant's most direct attempt yet to monetize this 3rd layer… and it’s hard to think Alphabet won't take its fair share here too. As a reminder, they were the late entrant in cloud computing and have built that into a fantastic business. I don't think the later debut will prevent it from winning yet again.

This will be offered for as low as $21/user/month for Gemini Business Plan subscribers.

Wells Fargo published some new search data that bodes well for both traditional Google Search and Gemini. The chart is not well put together, but the data in it is very useful. As you can see below, traditional search market share (which includes AI overviews but not Gemini) rose in July and August after falling consistently for more than a year. That share looks to have come from ChatGPT. In the other image, you can see Gemini beginning to take meaningful market share for the first time in September. Good news on both fronts.

Alphabet recently cut the number of search results that LLMs using Google data can get from 100 to 10. These chatbots rely heavily on Google data to power their own offerings, and Alphabet just 10Xed the revenue it can potentially get with the exact same search API usage from competitors.

PayPal is running a 5% cash-back BNPL promotion through the end of the holiday season. I’m a fan of this. BNPL is becoming extremely popular for younger consumers as they shy away from traditional credit cards. It’s also a higher margin product for PayPal vs. credit cards. Building consumer BNPL habits can make the 5% promo a durable source of structural transaction-related operating leverage.

And not only is BNPL good for transaction margin, it’s great for customer frequency, lifetime value and retention. BNPL users spend more on PayPal and lean on it more heavily for their financial lives. With PayPal’s vast product suite and larger customer data profiles than competitors, they can season their models better, know their customer better, and create more lucrative overall relationships with more resilient credit health.

The product is already doing quite well for PayPal as they have integrated it earlier on in merchant shopping flows. All of these advantages (along with a ubiquitous brand) make it clear why they’re leaning into this strength. I expect this to be a modest accelerant for user growth, especially considering the omni-channel nature of the relationship. They’ve also added pay monthly options to their in-store experience as they gear up for a busy season.

PayPal BNPL (especially with upstream presentment) amplifies conversion rates and revenue for merchants too… not just for PayPal. That makes this payments firm a more strategic vendor beyond a transactional button, which they’re trying very hard to do.

PayPal debuted the PayPal Ads Manager this week. The product greatly helps smaller businesses monetize their lucrative ad impressions. PayPal organizes that supply, plugs it into the firm’s vast base of advertising demand and matches merchant supply with the most valuable suitor for both parties. This maximizes cost per impression for its merchants and return on ad spend for its advertisers. Win, Win. They are the value-creating conduit in this situation. It’s wonderful what happens when a company like PayPal can connect high-intent shoppers to brands the company knows these shoppers will be interested in. Just another reason why PayPal’s vast customer data profiles on the consumer and merchant sides are so incredibly valuable. They’re finally taking advantage.

Ads Manager is similar to Amazon’s and Google’s and also somewhat similar to Trade Desk’s retail media product. It’s different from Trade Desk in that it focuses more on publishers (sell-side vs. buy-side) and on smaller customers rather than gigantic ones. Merchants can onboard in minutes and do not pay any upfront fees for access to this offering. Like previous ad offerings from the company, businesses simply compensate PayPal as they make money from automated impression placements. PayPal does all of the work for them, and offers complementary reporting and measurement products so an organization can act accordingly. This should be a popular offering for its small business customers, and a high-margin source of PayPal revenue.

Amazon announced “Quick Suite” as its own agentic enterprise AI platform to power more automation and better outcomes. This is going to sound a lot like the Gemini Enterprise section. It unifies agentic work across every facet of a company’s operations. That important feature ensures data engineers, developers and their autonomous agents can, in one place, access whatever information they need as they work through detailed tasks.

The launch includes a few pieces – starting with Quick Index. This is the data repository that provides an organized, low latency environment to store needed data. It integrates with leading vendors like Snowflake and Microsoft and handles a diverse range of data formats, including a seamless ability to ingest and make sense of unstructured chaos (1st & 3rd-party). And, because this product and data retrieval spans every company department, it can powerfully collect unified information and improve query response quality. That’s a lot more appealing than forcing humans to manually go find everything they need. It’s a “single, intelligent knowledge base” that offers superior data availability that enhances productivity and uplevels accuracy. This is the foundation of Quick Suite.

Quick Research is the next machine-based agent to discuss. This is what plans, proposes and conducts lengthy, automated research processes. It pulls heavily from Quick Index to comb through the mountains of tagged data, inform complex answers and provide keen insights on advancing day-to-day projects. It shrinks hours upon hours of research down to minutes. Users merely need to conversationally input a prompt and the agent can use that to formulate an intricate, actionable research proposal. 

Clients can seamlessly edit any part of this plan – again in a no-code manner – until it’s ready to be deployed. After that, the agent progresses behind-the-scenes to complete the work on an employee’s behalf. Research findings are backed by primary sources and equipped with easy-to-follow “reasoning paths” that make it obvious where answers came from. This, along with lower model hallucination rates thanks to Quick Index, should give enterprises the confidence needed to rely on these findings.

Tying closely to Quick Research is Quick Sight Business Intelligence. This makes a person’s job even easier. Not only does Amazon do the research for them, but it can also create valuable recommendations and summaries to improve routines and decisions within an enterprise. These improvements can be approved and unleashed with the click of a button. Again, this spans every company department and leverages data from all of those departments as well, meaning insights are informed by valuable information across an entire company – rather than a siloed segment. Quick Sight unlocks follow-up questions, ensuring a user completely understands the findings and implications from this tool. And finally, it offers “what-if” scenarios for software engineers to easily experiment in a no-risk environment and actually observe what works pre-deployment.

Quick Flows Automation uses the deep research and proposed multi-step tasks from Quick Sight Business Intelligence. It takes this knowledge to directly handle the automation of work. This is sort of like the puppet master, as it combines various tasks across an enterprise and strings them together in an orchestrated, ordered way. Users can easily customize and tweak various parts of the proposed agentic tasks,  making them more relevant for desired objectives. That’s done via “Chat Agents,” which is a unifying chatbot that sits on top of Quick Suite to permit seamless querying.

Quick Automate is very similar to Quick Flows Automation, except it’s for more elaborate jobs with added department involvement and more steps to completion. This also provides deeper code debugging, testing and monitoring tools to keep these autonomous agents in check. And additionally, these agents are customizable with a firm’s own lucrative data to make workflows even more granular and expansive for certain use cases.

Why does all of this matter? It is making Amazon more of an end-to-end AI partner and vendor. Offering more products under one roof should be great for lifetime value, retention, cross-selling and Amazon’s overall results. The e-commerce and cloud giant has a massive library of 1st and 3rd-party models, great model usage and optimization offerings like Bedrock and SageMaker, quickly improving training chips and now an aggregated suite of software-based agentic tools to accelerate enterprise AI adoption. It combines all of these capabilities with world-class distribution, which means world-class data access to expedite model and agent improvement. The approach here compared to Alphabet is quite similar. 

The GPU build-out is going to slow down at some point. Timing uncertain. Whenever that happens, monetizing the app layer of this AI opportunity will be paramount for extending growth runways and avoiding sensational flameouts. Launches like these are how Amazon (and also Google this week) can take their fair share of the far more structural growth opportunity.

Amazon is adding unmanned pharmacy kiosks at its One Medical offices. These can field and fulfill requests right from a physical location, adding another layer of convenience for the budding Amazon Pharmacy offering.

Every time I consider adding some exposure to the AI trade, I determine I'd rather have that exposure through Amazon and Google. These companies will benefit mightily from the AI runway extending longer and longer. Whether that's via GPU renting, selling their own chips, rising model demand, enhanced cloud workload demand etc. they will continue to reap the rewards. This also doesn't even get into the enterprise AI software releases both announced this week or the apps they're building for future monetization... or internal cost controls... or existing app improvements.

And if the cycle slows? They have world-class businesses entirely unrelated to AI infrastructure. Those businesses will buffer financial weakness stemming from AI tailwinds fading. I think that will mean corrections are a lot less severe vs. chip companies and pure-play data center vendors. That, to me, is very comforting as the world wrestles with how long the AI boom persists. As I've been saying recently, markets right now (for me) are about shielding myself from some potential downside amid any correction. That is more important than maximizing short-term outperformance. That's why the cash pile has grown and some exposure has shifted from high beta to lower beta.

What happens when GPU and related demand finally begin to slow? CapEx growth dwindles and FCF for hyperscalers explodes. Additionally, compute disinflation will accelerate as the supply/demand mismatch fades. That will also be fantastic for software company margins, as their costs associated with AI-inspired upgrades fall.

So again... tempting to add exposure to the high-flying, highly expensive pure-play AI darlings... but that would be deviating from my process and my wheelhouse. So I'm not going to do it.

I think software's structural tailwinds will begin to matter much more as the currently insatiable GPU buildouts eventually slow. I think that, as well as more favorable AI costs, will improve growth, margins and sentiment. And I think I want to stay highly exposed to this general sector, I just  don't make timing cycles an important part of my process. I like structural, boring growth.

Robbins gave an interview in Vegas this past week at the G2E conference. It wasn’t public, but I was able to get a summary of it.

He confidently said he doesn't see Prediction Markets taking meaningful market share. He added that most of the volume these guys are getting is from California and other states where gambling isn't legal. When looking at 15%+ Y/Y growth in New York State betting volume this past week, the data is still saying he is right. This level of growth in that state has been consistent in recent months and weeks, despite it being 6 years since legalization occurred. He added that the product is far ahead of anything prediction markets offer and reiterated how DKNG's market maker status gives it more control over risk, limits and betting menu. Meshes well with UK data pointing to prediction markets not really gaining much market share vs. legal books. He would not touch a question of whether or not he thinks prediction markets should be allowed.

If he's right about DKNG's competitive positioning, this selloff will likely be short-lived. We'll go back to focusing on this being a co-market share leader that is rapidly growing and expanding margins at a dirt cheap 16x forward FCF multiple (growth multiple well under 0.5X). Attention will again turn to the increasingly profitable financial engine and other items like how early we are in iGaming legalization. If he's wrong, this competitive risk will stay front and center and will probably keep weighing on the stock. It was great to hear these comments. They make me more confident in holding shares amid the recently ugly price action, although I’m going to wait until the Q3 report comes to entertain adding to the stake. 

The take rate in New York this past week was bad. It was 3%. This will impact Q3 results, as varying degrees of luck always do. But? Volume is the metric that shows whether or not they’re losing market share and it again looked very good for DraftKings. That’s so much more important to me right now. 

Finally, there was an interesting passage from a Bank of America buy reiteration note on DraftKings that I wanted to share. It meshes well with what leadership tells investors and what we’ve been discussing in recent weeks:

“Beyond state regulation, we believe there is a strong moat for existing OSB operators that have invested billions of dollars in 1) product, 2) technology, and 3) customer acquisition and retention. These investments lessen cannibalization risk in their existing 29 states. In these markets, PMs offer an inferior product for core sports bettors, with a fraction of the markets, worse speed/latency and no bonusing/promotional reinvestment to keep bettors active. Better pricing for PMs is debatable, but we have shown the fees are more comparable at ~3.0-3.5% for PMs of estimated handle vs. 4.0-4.5% for OSB. Investors in OSB are rightly concerned about competition from PMs, but the math shows there is substantial room for PMs to grow without having any impact on the existing public-company OSB landscape. We estimate the 1) 21 states (incl. CA, TX, FL) where OSB is not accessible plus 2) the 18-20 year old population is a $150B+ handle market double the $150B for the current 29 states. Translating this to “volume” for PMs would reach upwards of $1.3 trillion dollars vs. Kalshi’s current run-rate of ~$40B. This means Kalshi could grow ~25x and not take a single dollar from the existing OSB operators.” – Bank of America Analyst Shaun C. Kelley

The $20B currency swap agreement has been finalized and the USA purchased Argentine Pesos today. Bessent just sent out a long tweet that was quite upbeat and positive about the future Argentina/USA relationship. Great to hear and should be very good for Mercado Libre sentiment, as well as nominal revenue growth in that region. 

"The success of Argentina's reform agenda is of systemic importance and a strong, stable Argentina that helps anchor a prosperous Western Hemisphere is in the strategic interest of the United States... their success should be a bipartisan priority." – Treasury Secretary Scott Bessent

I know Bessent is a political figure but this is highly important for MELI and is something I need to share with you all to give you the full picture of what's happening here. Not here to comment on political strategies, but need to mention them when these strategies impact our companies.

Mercado Libre added a new business-to-business (B2B) unit across all of its core markets. B2B relies more heavily on different things like granular employee permissions, custom discounts, bulk ordering and other items that aren’t as important when selling to consumers. Creating this dedicated team signals MELI’s intention to turn B2B into a large growth level for the firm. This has been monumentally important for Shopify accelerating its own growth over the last two years, and the opportunity in Latin America should be less tapped and even more compelling than in the states. There’s no other company better positioned to capture a large piece of this pie than MELI, and it’s exciting to see them more intentionally going after it. They’ve been testing this product for more than a year, with 4 million users before a formal launch.

Mercado Libre also launched a new car-selling tool in Argentina. It offers prices, sets up 3rd party inspections & offers upfront payments or handles auctions on behalf of sellers. They’re doing this through a partnership with a company called Flash. And in other partnering to boost assortment news, MELI is aggressively pursuing authorization in Brazil to add pharmaceuticals to its marketplace. This will be done through 3rd party merchants, similarly to how it approaches perishable grocery offerings. Both announcements should deliver subtle boosts to overall user frequency & MELI GMV. And to make sure this new assortment enjoys as much momentum as possible, MELI is providing free same-day delivery in Brazil for all orders above $3.56.

We've seen a ton of progress in SoFi Invest over the last few months:

You may see this list and wonder: “what’s so special? Everyone has these products.” And you’re so right. The thing is… SoFi didn’t. It had large, gaping product holes in its brokerage suite that were forcing it to compete with a hand tied behind its back. These launches effectively fill that void and allow it to compete more effectively. Considering its vast cross-selling potential and all the value creation that stems from it, I like their odds.

Leadership told us a few quarters ago that it was finally ready to make SoFi Invest a higher priority in the product suite. Innovation sharply sped up accordingly, and I think that will continue to accelerate as it finally finishes integrating Technisys into Galileo. That will unlock a unified, fully-owned tech back-end that lets SOFI experiment, customize and build without reliance on any 3P vendors. The cost edge associated with owning this tech stack is something we cover constantly, but the innovation edge is another important factor to consider, and it's about to be unleashed.

This is all exciting, and it doesn't even get into Hong Kong plans. It also doesn't cover its aim to greatly grow the investment margin lending business by cutting borrowing rates to match competitors. The owned tech stack, lack of branches and its LTV edge from cross-selling more products will enable all of this.

It's clear that the monetization gap for SoFi Invest will keep closing. For context, leadership saw this product as a full 50% under-monetized, with any future catch-up providing clear opportunities to accelerate growth. Leadership also thinks this offering can be AS LARGE AS SoFi Money down the road. And right now? Brokerage fees are just 5% of its non-interest income within financial services specifically. So much growth ahead if they're right like they always are. These updates lay the foundation and I expect revenue growth acceleration for SoFi Invest to follow.

Citizens JMP had quite the notable EPS revision for SOFI today, raising their Q3 EPS target from $0.07 to $0.12. Consensus currently calls for $0.08.

Optimism is mainly based on the thriving loan platform business, but also due to rising SoFi Invest confidence (see the note above for all of the recent product momentum there). Stablecoins and crypto were both part of that optimism. They also noted resilient credit health, which we always love to hear.

The opinions on the company were uniformly positive. The only negatives they had to offer were on how much the shares have already rallied, potential macro weakness and any future market corrections leading to SoFi shares pulling back. That's fair and, candidly, pretty obvious in my mind. To me, by far the most important thing is the large raise to their EPS target and the complete lack of anything negative to pick at for SoFi the company.

Bank of America also raised their EPS target for the quarter from $0.08 to $0.10 in a quick update as well, while maintaining  an underperform rating. This is related to their belief that shares are “priced for perfection.”

The SEC is investigating the firm’s data collection practices. This investigation is a follow-up to the short reports published on APP over the past couple years. Allegations from those reports included:

APP responded by basically saying they’re cooperating. This could be nothing. Investigations can be all bark and no bite, especially ones that are in response to short seller requests like this could potentially be. We’ll have to wait and see what they find.

If the SEC does discover anything damaging, it could potentially push APP to change its business model. Not my base case expectation, but it’s possible. And if it has been unlawfully using data, that means it would have to implement much tighter data leveraging controls. Tightening would lead to some level of signal loss that could dampen targeting efficacy, which is the holy grail in advertising performance. In my mind, avoiding that outcome is far more important for APP than escaping a potential fine.

Relevant to Meta – EssilorLuxottica CEO on the Ray-Ban Glasses in a Bloomberg article from this week. In it, he said “glasses are poised to become the central device in people’s lives, possibly replacing smartphones.” Bold and exciting to hear. For now, new models are flying off the shelves and this product is on a great path. In other Meta news, data from Wells Fargo is pointing to sharply accelerating time spent growth for both Instagram and Facebook.

Needham’s fall survey yielded upbeat takeaways for Duolingo. They’re more “bullish on ARPU expansion & less concerned about cannibalization from GenAI enhanced translation tools.” The only negative is that they haven’t seen overwhelming traction for Advanced English yet. But the overall conclusion was positive. They reiterated an overweight rating with a $460 target. 

Amazon raised pricing for Nvidia-powered compute rentals. This could embolden hyperscalers to make the same move and is a positive hint for future chip demand.

OpenAI added Uber, Spotify & Booking.com integrations (among others) to allow those products and services to be offered through its own interface. OpenAI sent out a blog post teasing some AI-native tools it’s building. These include products that could compete with DocuSign, Salesforce and HubSpot. Finally, the massive private company introduced “Custom GPTs” that allow customers to customize ChatGPT, with broad access to popular integrations that some fear could disrupt the traditional software as a service model (SaaS). While that could be true for disparate point solutions providing little overall value, I don’t see that happening for true platform plays that are capable of effectively innovating. These companies have years (if not decades) of highly relevant data to train their own algorithms and models and build great products for a wide range of use cases. That data is what separates good products from great products and sort of useful tools from tools that can handle any fringe use case with ease.

RBC initiated Cava coverage with an outperform rating based on strong market positioning, store growth runway, and expected comparable store sales durability.

We’re not getting much macro data with the current government shutdown. We did get some important China trade news on Friday via a social media post from Trump. In it, he imposed 100% higher tariffs on China in response to their decisions to restrict rare earth metal access. He also talked about canceling an upcoming meeting with China’s president, but later walked back those comments. This feels like the perfect excuse for Mr. Market to take a much, much needed breather. 

We've seen this same risk play out a dozen times. And each time, the end result is better than feared for Mr. Market (outside of very specific pockets such as apparel). I expect that to happen again. And, in the near term, I am selfishly rooting for the noise to grow louder and for markets to take that noise very seriously. Incentives point to the eventual outcome being as economically favorable for both sides as possible. If Mr. Market wants to throw a fit while we noisily get to that outcome, I'd say thank you.

I'd re-enter accumulation mode if we get a meaningful pullback and let the cash pile start to shrink once more. We are not there yet. This is a tiny move amid several months of uniformly fun price action. Just expressing my plans if the selling does accelerate in the future.

Two things are true in my mind: