My SoFi investment case is progressing nicely. I will have it for you before the start of next earnings season.

Access to capital market loan buyers matters a lot for SoFi. It’s how the company can continue to originate quality loans beyond its current regulatory capacity, without taking too much balance sheet risk in the process. It allows the company to accept new customers, delight them with acceptable loan terms, generate gain on sale premiums and collect that data for future cross-selling. It can do all of this without requiring them to get overly aggressive or irresponsible in terms of leverage. Simply put, capital market access is vital for SoFi’s flexibility and financial results.

Rate cuts directly help capital market demand. They make future expected cash flows from underlying loan pools worth more and also lower borrowing costs to make hurdle rates more attainable. This happens while SoFi (per the team) will get more aggressive on unsecured originations (thanks to cuts) while enjoying more demand for its secured origination and refinancing businesses. The effect is a positive feedback loop of tailwinds that allows companies like SoFi to get bolder in catering to more borrowers. It’s a perfect storm (in a good way). And again, that means more member growth and more opportunity for cross-selling.

We’ve now gotten a double rate cut and have seen borrowing rates begin to fall. And? This week, a new SoFi personal loan securitization was filed on the SEC’s website. Ratings for this pool have not been released and we don’t yet know what the gain on sale margin metrics are for the assets. Still, considering SoFi’s large capital ratio cushion and its strong gain on sale margins through the worst of this cycle, it’s hard to believe it would accept unfavorable terms. If demand wasn’t there, it could easily and profitably just keep the loans on the balance sheet. I’m speculating a bit, but to me this is common sense.

What we do know is that the securitization is quite large, and so quite material for adding incremental origination capacity. Specifically, the securitization represents more than 14,600 loans, which is roughly 50% larger than its previous two deals. The floodgates appear to be reopening.

Subscribe

As part of Duocon 2024, Duolingo launched a few new AI tools. It added new “immersive adventures” or quests to its app with a boatload more content variations and a new GenAI video call tool as well. With the video product, learners can have real-time, detailed conversations with AI chat bots. These bots are programmed directly into the front-end Duolingo character Lily. This helps any learner feel confident in talking to a fluent counterparty, without feeling nervous about saying something silly. 

Lily has memory and can fetch previous, relevant answers to enrich the conversation while tailoring speaking difficulty levels to learners. It doesn’t tell you that you’re wrong when you offer a confusing or incorrect answer. Instead, it asks you to rephrase or clarify something, like you’re actually talking to a person. In true Duolingo fashion, this quirky tool is backed by data science and AI algorithms. It strives to minimize the self-correction part of the brain that prevents us from taking chances and learning via trial and error.

Every time I’ve visited a Spanish-speaking country, I’ve wanted to speak more Spanish but I didn’t for fear of offending or embarrassing myself. For me personally, this sounds useful. Video calls and the new quests will be laced into the more expensive Duolingo Max tier, juicing long term pricing power, retention and engagement. These are the kind of launches that wake investors up to how positive GenAI should be for the firm and the launches that will allow its fabulous financial success to be properly rewarded. We saw that playing out this week as sell-siders lined up to praise the aspects of Duolingo’s business.

For the last several quarters, and especially since the OpenAI translation product launch, bears have been loud and steadfast in their belief that AI will kill this company. I believe they are wrong and I’ve said so many times. Duolingo uses GenAI to vastly accelerate and fuel its content delivery network. It partners closely with OpenAI to gain access to their latest models. And? It is for learning and entertainment, rather than cheating (sorry Chegg). A few months ago when this negative sentiment was peaking, I asked readers to contemplate the following:

I think the answer to all three questions is a resounding “no.” Those are the common Duolingo use cases, and GenAI does nothing to replace them. In reality, GenAI supports Duolingo’s rapid and constant product improvements. Without it, this new video product and a newer Duolingo podcasting product would have never been created. It would have taken too much time, too many developers and too much energy to make pursuit rational. That’s no longer the case. Humans now act as the foundational building blocks of new content, with AI models taking that foundation and driving countless variations. This company, with its wildly impressive team and unmatched dataset, is poised to utilize this new technology to extend its lead over everyone else in the pack.

For evidence of its success, look no further than DUOL’s financial engine (tables below). It’s the unmatched ability to create great products through obsessive, data-driven split-testing to motivate viral, global, word-of-mouth growth that drives this. It doesn’t prioritize sales & marketing like its competition, but rather R&D. Great products are the best marketing engine a company can have. No company on the planet is currently delivering its combination of growth, leverage and GAAP profitability. Furthermore, no software company coming remotely close to its growth and margin profiles trades for as reasonable of a multiple DUOL. Here’s how the company stacks up against two of the highest quality software names on the planet:

Also consider this: just one year ago, analysts expected $2.30 in 2024 EPS. The stock traded for $150 per share and a PE of roughly 65x. Fast forward to today, and explosive upward profit revisions mean the $275 stock stick trades for 62x $4.33 2024 earnings estimates. This stock’s success is not being driven at all by multiple expansion… just outperforming profit growth. That’s ideal. Duolingo has quickly turned into one of my largest winners and largest holdings, and the (relatively) modest PEG ratio makes it clear why I’m happy to keep it near the top of the portfolio. Plans on the holding were spelled out in a Stock Market Nerd Max article sent earlier this week.

As part of the Duocon event, the company announced a new partnership with “Loog.” Together, the two will offer a portable piano for Duolingo Music. All of the firm’s financial success to date has been driven by language learning and the Duolingo English Test (English proficiency for universities and employers). Music, Math and all future subjects are future monetization opportunities. Early traction is great, but Duolingo always takes its time on monetizing any new product. It wants to make sure the offering is optimized and reasonably scaled before doing so. Just like Meta.

For the Math product, Duolingo is in content creation mode. It’s adding real life objects such as rulers, dollars and measuring glasses to connect problems to tangible scenarios. It also added 4 new games for brain training and learning and launched the app for Android in 6 languages.

For the Music product, it added a lot of modern songs with its new Sony Music partnership. It also added a more complex curriculum and ear training for listening skills too. Like for Math, this recently launched for Android.

Micron trades for 12x forward earnings. EPS is expected to grow by 600% this year, 43% next year and -14% the year after. Very cyclical business.

Meta packed more announcements and innovation into this event than any developer conference I’ve seen in a very long time. We’ll dig through all of the important highlights here, with some commentary on the other presentations given.

Meta’s open source approach to GenAI Llama foundational model creation is working. It’s inviting world-class developers to build with these tools, which inherently means model optimization, improvement and new inspiration for Meta to capitalize on. In a world of models racing to commoditization, this is how you become the low cost provider and truly stand out. And while Llama has garnered significant praise and success to date, the Llama 3.2 model should build on this momentum as another open source, frontier model release. 

Llama 3.2 is Meta’s first multi-modal foundational model, meaning it can field and understand text and images… and also voice. Zuck thinks voice-based interaction with GenAI models is the most natural (I fully agree) and sees this as another large unlock for adoption. Speaking of adoption, Meta AI (powered by Llama) is well on its way to becoming the largest GenAI assistant on the planet, with 500 million MAUs. This is an example of where offering free access to Llama models (besides proprietary pieces Meta keeps for itself) leads to financial value. Meta AI laced throughout its apps creates more engaging experiences, more queries and, so, more ad impressions. It’s also a key piece of Meta’s customer service tools for enterprises via automated fielding of questions and the creation of creator-level AI avatars. Monetization will be both indirect (due to more time spent on its apps) and direct via more opportunities for selling higher value business subscriptions. And, as we’ll expand on below, these models are vital for maximizing the value of its various lines of hardware to bolster adoption there. 

The model comes in 11 billion and 90 billion parameter versions and is available everywhere except the EU. Europe continues to be very strict with AI regulation on a relative basis. I’d argue that those regulators are shooting their union in the foot in terms of competitive positioning and economic growth, but that’s a topic for another day. Interestingly, there are also 1 billion and 3 billion parameter small language models (SLMs) that “outperform” all other models at those sizes. This should be important for more niche use case traction, as these are less expensive to customize and train on top of. These are also on device models that will soon run on the Orion hardware described later.

Zuckerberg sees Llama becoming the “Linux of AI” in terms of being an industry standard. This is why closed source competition is slashing their own pricing, and why that process will likely continue. Open source is the only approach that can work as the world races to cut costs for training and inference. Monetization needs to happen elsewhere… not for actual model selling. That’s Meta’s opinion and my own opinion too. 

For years, Meta has been talking about blazing the mixed reality (MR) innovation trail and laying the foundation for future adoption. This doesn’t just mean miniaturizing the technology, optimizing display, and resolution… It also means driving down the cost to improve access. To really make this the next computing form factor, affordability is a key ingredient.

Along these lines, Meta announced a new version of the Quest 3 headset called Quest 3S. It will start at $299 and will begin shipping next month. That’s $200 cheaper than Quest, thanks to a constant obsession with optimizing input costs, manufacturing processes and other efficiencies at every turn. While Quest 3 is hailed by the firm as the “first high-quality mainstream MR experience,” this should lead to another “large influx” of users in its ecosystem, according to CTO Andrew Bosworth. 

Still, if Meta is basically selling this hardware at cost, why should we care? This gets back to how Meta plans to monetize this opportunity over the long haul. It hopes to build out an ecosystem with the most traffic, enjoy the coinciding 3rd party developer traffic that follows, and collect margin on the app and software side of things. It also plans to lean on its unmatched dataset to train models more effectively. The potential software profit driver only realizes its true potential if you focus on driving down cost at the initial point of sale.

“We don’t just innovate to advance the state of the art. We innovate to drive access.”

Zuckerberg

The approach stands in stark contrast to Apple. That tech giant is, like it always does, charging a large premium for its own headsets and collecting immediate margin at that point of sale. I personally think the app and software opportunity will eventually be far larger than for hardware, but Apple also thinks its fortress ecosystem will be enough to command that premium and still enjoy great traffic and software revenue. If anyone can pull this off, it’s Apple. I prefer Meta’s approach, but both will likely work to varying degrees.

Quest 3S builds on what Zuckerberg views as the “best MR devices you can buy today. This comes with the “same defining features as Quest 3” including: the same processors, elite pass through technology, hand tracking, touch controls and its full library of apps. Other updates include:

Everything above sounds exciting, but these Quest (and Apple Vision Pro) headsets are still years away from potential mainstream adoption and ubiquity. Why? Because they need to be comfortable enough and stylish enough to wear around all day long. No headset on the market comes close to that today.

To solve this, Meta and others are in a technology miniaturization race. They need to pack in all (or most) of the use cases from the clunky headsets while also making these as easy to wear as the Ray Bans on our faces. For 10 years Meta has been working on this, and Meta Connect 2024 provided a compelling progress report on this lengthy journey. During the event, Zuck unveiled the rumored Orion Project glasses. He took us through demos from several notable individuals who genuinely became emotionally delighted by the advances of this technology. While I am certainly a big fan of Zuck and Meta, I think anyone watching this event would have been impressed. So what’s the big deal?

Orion is the very first fully holographic, augmented reality (AR) glasses in existence. You can wear the glasses to chat with your friends from across the globe, or play a game of ping pong with them. It comes with Meta AI (now multi-modal via Llama 3.2), live streaming, photos and so much more. It’s 100 grams, has no wires and offers a fully wide field display. The lenses don’t use pass-through technology and aren’t even made of glass. Instead, they are actually made of silicon carbide (with a very light magnesium frame). Tiny projectors built into the glasses “shoot light” into the frames to enable 3D display of holograms. Pretty neat. The frames are also cooled with the same tech that NASA uses to cool space satellites. Orion comes as the Meta Ray Ban glasses continue to fly off of the shelves. And while Meta now knows it doesn't need full AR glasses to drive adoption, it sees Orion as the true prize for fostering ubiquity in the coming years. It thinks this will deliver on its vision of creating a sense of social presence in remote settings.

The interaction interface here is also novel. It does offer hand tracking, voice-enabled Meta AI and eye tracking… but there’s more. Orion offers the first, wrist-enabled neural interface for this type of hardware. 

Orion has already delivered on all project objectives, but Meta has a bit more work to do on sharpening display, shrinking the hardware to make it more fashionable and finding more manufacturing efficiencies to drive down cost. For now, the glasses are still a prototype and what Zuck describes as “a glimpse into the future.”

Meta doesn’t have the greatest reputation for being developer-friendly. Accordingly, the company has made this issue a core focus for the last year. It has created more open environments for easier, lower-code customization (like the AI studio). It also pushed through more integrations with the tools developers know and love to make work more comfortable for them. This has helped the Meta Horizon app store 10x its library of apps since April.

As I’ve been saying for about 2 years now, Meta’s approach is the winning approach in GenAI. It is accepting commoditization when that’s inevitable, masterfully focusing on being the low-cost provider where needed, creating tools to motivate more user and developer traffic and setting itself up for ample app monetization down the road. It is removing friction when needed to expedite adoption of this new technology and leading the innovation race to make sure it’s best positioned to capitalize on the faster pace. 3 years ago, the argument was “how can Meta ever compete with Apple in hardware if they’ve never built anything meaningful?” 3 years from now, I think the new debate will be “how massive can this opportunity actually become for Meta?” The pace of innovation here is world class and the hardware adoption is coming more quickly than expected.

Zuck and company faced cross-app data sharing restrictions from Apple head-on and overcame all of those challenges despite it blowing up the core business. It took that competitive punch from King Apple and let it light a fire under the entire company. It used this to drive a sense of urgency in both addressing ad targeting gaps, but also pushing to control the next computing consumer hardware beyond smartphones. It dedicated itself to controlling its own destiny. These are the reasons I admire Zuckerberg’s leadership so much and remain so confident in this firm despite it already being my biggest winner ever.

The Justice Department will soon file an antitrust case. The complaints are based on unfair deals signed with PayPal, Square and others that prevented competition in the debit card space. Visa dominates this category, with a 70%+ market share.

Visa is an incredible company with a fortress moat, secular growth tailwinds, significant value added services to cross-sell and an ability to thrive on global economic growth. This is one of the best businesses and business models on the planet. This is why it trades for a PEG ratio north of 2x, with a 27x earnings multiple and has multi-year earnings growth expectations of 12%.

Selfishly, I hope the noise from this lawsuit becomes exceedingly loud and drags this company down as a result. However, as is usually the case, I anticipate considerable bark with little bite. I see it making a few concessions to hollowly appease regulators before continuing its top-and-bottom-line compounding march against any and all competitors. I would absolutely love to get my hands on this one if we can see the forward multiple move close to 20x.

Raymond James initiated Uber with a “strong buy” rating and a $90 price target. The interesting part of this news was its commentary on autonomous vehicles (AV). It sees Waymo as reliant on Uber for about 4 hours of utilization per day and modestly incremental to Uber demand. It also doesn’t see Tesla achieving needed performance gains in self-driving cars as quickly as previously thought. Raymond James thinks these two items create more positive sentiment around Uber’s business. 

I tend to agree for now. Uber is the only way to optimize utilization rates and allow for a rational transition from driver to driver-less. The issue is where Uber stands once fleets are mature and Waymo racks up its own market share. Will Google, Amazon and Tesla compete? Probably, and Uber’s monopoly-like presence could be at risk if that happens. I expand on these ideas in Section 4 of this article. None of my views on Uber have changed since this post.

If you listen to Nu’s leadership team speak, they talk about Brazil, Mexico and Colombia being the first of many markets it expects to enter. The issues of underserved consumers, poor access and expensive, antiquated financial services are not specific to Latin America. Nu sees a significant opportunity to drive more value creation across the Middle East and Northern Africa too.

While this could be seen as a distraction from the core, it doesn't bother me. Nu has roughly 50% of its team working on new products, new markets and new ideas. That has been true for years while it has surgically executed its pursuit of rapid, consistent, margin accretive growth. They have shown an ability to walk and chew gum at the same time and, in my mind, have earned the right to go pursue these more speculative markets.

Most importantly, I hope readers in the southeast impacted by Hurricane Helene are okay. My thoughts and prayers go out to you. There have been mounting concerns about the potential for Hurricane Helene to create large losses for Lemonade. It does not offer home or auto insurance in Florida, only pet and renters. It does offer home insurance in Georgia and Tennessee, with premiums between the two states somewhere between 3% and 4%.

Lemonade has been shifting its book away from catastrophic (CAT) event risk and consistently talking up considerable upgrades to its underwriting algorithms. The company also fared quite well last quarter in a tough CAT event period. In my non-hurricane expert opinion, I think all of these items give the firm a good chance to navigate this awful hurricane as well.

PayPal added business account adoption for transacting with cryptocurrency.

Celsius energy drink volume rose 32% Y/Y through the first week of September, with pricing falling 20% Y/Y. This data does not currently correlate with Celsius results, as its distribution partner Pepsi continues to shrink/optimize their Celsius inventory held.

Flutter (owns Fanduel) gave a very robust North American gambling market outlook of $70 billion by 2027. DraftKings also responded very positively to this news. This new estimate is 50% larger than Flutter previously estimated. It also offered strong revenue targets, representing a 14% CAGR through 2027 (roughly as expected), $5 billion in EBITDA by 2027 (4% beat) and added a $5 billion buyback.

Alphabet will launch an AI assistant for Volkswagen cars and is investing $3.3 billion in two new South Carolina data centers.

Uber added WeRide as another autonomous vehicle (AV) partner. The two will start offering WeRide on Uber’s app starting in the UAE. WeRide has the only AV license in the UAE.

Disney added the ability for Disney+ family plans to add an account from outside of the physical residence. It costs an added $6.99/month for basic and $9.99/month for premium.

China is aggressively providing monetary support to juice economic growth. This includes material rate cuts, market liquidity, mortgage deregulation, stock buyback incentives and more. Just as a quick aside, I greatly prefer gaining exposure to the Chinese economy through U.S. based firms with businesses in that nation. I don’t personally invest in the ADRs.