SE reiterated ~25% Y/Y Shopee e-commerce marketplace growth.

EPS is expected to grow by 86% next year and by 27% the following year.

Its e-commerce platform, called Shopee, is performing well. Gross orders rose 24% Y/Y vs. 40% Y/Y growth last quarter while GMV rose by 25% Y/Y vs. 29% Y/Y growth last quarter. Core marketplace revenue rose by 43% Y/Y. This bucket consists of transaction fees and advertising (+49% Y/Y), as well as value added services like logistics (+29% Y/Y).

From a profitability point of view, Shopee re-inflected to positive EBITDA as expected. This positive EBITDA milestone occurred in both Asia and Brazil and is expected to continue going forward. Last quarter, the company generated a positive contribution profit for the very first time in Brazil of $0.09 per order. Encouraging to see that breakthrough followed up by another subsequent profit inflection this quarter. Also encouraging that it took SE just 5 years to turn profitable in Brazil. Buyers in that country rose 40% Y/Y while customer cohort quality improved.

While competition from Temu and Shein in Indonesia or MercadoLibre in Brazil continues to be a concern, market share trends remain positive and SE is quite pleased with its competition positioning. No changes in Asia or Brazil. There has been a lot more cross-border competition in its markets, but most of its sellers are local, so that impact is minimal.

Notably, item and advertising commission take rates also rose for SE as its markets saw further competitive “rationalization.” Great to hear. For ads specifically, the take rate rose by 30 bps Q/Q. Platform investments, targeting algorithm enhancements, easier seller onboarding and a new dashboard all helped SE’s ad business continue to command a larger piece of the pie. Sellers rose by 10% Y/Y while revenue from these sellers rose by 25% Y/Y.

“Many of our markets still have very low e-commerce penetration rate. This puts us in a great position to continue to grow as e-commerce penetration improves.”

Founder/CEO Forrest Li

Like for other global marketplaces, a big piece of SE’s competitive positioning is its fulfillment network. This is where Shopee Express (SPX) Delivery, its rapid delivery service, comes into play. It has worked hard to more deeply integrate with logistics partners over the last few quarters, and it thinks this work is paying off. 50% of SPX orders were fulfilled in 2 or fewer days (vs. 70% in 3 or fewer days last quarter) while costs continued to fall. Other services like its “change of mind” 15 day return policy are also helping drive higher basket sizes, which means more order batching to cut costs (as well as more revenue). Generally speaking, large investments to enhance efficiency and service levels have yielded significant input cost intensity relief. It has created an end-to-end, vertically integrated supply chain for SE and its merchants to drive interoperability and better outcomes. It’s passing some of the coinciding savings onto buyers and sellers to deepen its value proposition. And speaking of price advantages, it offers the best prices in its markets according to a Portrait survey.

Finally, live streaming on the marketplace continues to boost engagement and be a popular tool for merchants to use to stand out. Investments over the last year here have “paid off.” Streamers rose by 50% Q/Q while streamers who bought something from a merchant rose by 15% Q/Q. A lot of this success is coming in Indonesia, where it’s the largest live streaming e-commerce platform and continues to enjoy steady operating leverage.

Basket size is higher for these shoppers, which means better margins and retention. It also just debuted a new YouTube partnership where content creators can put shoppable Shopee links right on their pages.

SeaMoney’s consumer and small business loan book rose by 73% Y/Y to $4.6 billion. It reeled in 4 million first-time borrowers and generated 60% Y/Y growth in loan customers. The company is aggressively leaning into originations and is doing so while delivering resilient credit metrics. Its 90+ day non-performing loan (NPL) rate actually improved a bit Q/Q to 1.2%. Furthermore 48% Y/Y growth in credit loss provisions materially lagged overall asset growth. SE is effectively using its rich customer data profiles within Shopee (tons of payment history) to give its underwriting models a key edge over others. This approach seems to be working, just like it does for other similar business models around the globe.

Credit across SE’s markets is very underserved and demand is robust. Still, it’s going slowly here to ensure strong performance remain strong. For example, it starts first-time borrowers without ample data on very small credit limits. It slowly, carefully raises those limits over time as a customer demonstrates strong tendencies. That’s pretty common practice, but it’s easier said than done to stick to this discipline over chasing faster growth. That’s always tempting.

It’s also getting more aggressive with originating credit for non Shopee users. For example, it’s running a successful new program for phone purchases in physical stores. Much more to come here, as now 50% of its loan book is from outside of the Shopee ecosystem. This does eliminate the lucrative advantage of underwriting customers you know better than anyone else. Still, if they can continue delivering these credit metrics despite shifting to this source of growth, it should be full speed ahead.

Sea Limited thinks Garena bookings growth will reach 30% Y/Y for 2024. For this quarter, bookings rose by 24% Y/Y to greatly outpace -16% Y/Y revenue growth for the segment. This is related to negative revenue growth during recent quarters still being reflected in the Y/Y comp. Bookings calculates forward-looking demand, and offers strong evidence of revenue growth quickly rebounding in the quarters ahead. This segment was aggressively helped by the pandemic pull-forward, but seems to be getting back to a point of more structural, reliable demand. Free Fire continues to be its standout game several years after launching. SE’s continued obsession with in-game enhancements (it thinks) is the largest reason why, with 25% Y/Y daily active user growth and 25% Y/Y download growth offering evidence.

Momentum in its core markets is solid while North Africa is quickly turning into another promising expansion opportunity. It hosted the largest ever e-sport event in Morocco and generated significant social media buzz to build brand awareness.

The big debate surrounding SE’s business over the last few years was growth stickiness. Skeptics argued that its revenue growth was entirely and permanently reliant on spending more marketing dollars. They thought of recent margin improvement as temporary and unsustainable if SE ever wanted to grow revenue again. They thought ramping competition would yield rising customer acquisition cost (CAC) and permanent margin pressure. Optimists believed in rising unaided brand awareness, easier comps and the firm’s track record of growth before the pandemic. They thought of these factors as obvious evidence for SE being able to generate top line expansion without sales & marketing (S&M) scaling in tandem.

This quarter bodes very well for the bulls. S&M fell 4.3% Y/Y overall and by 11% for its e-commerce segment, which is where marketing reliance was thought to be sharpest. And as you can see in the demand section, it delivered 30% Y/Y revenue growth despite these cuts. G&A rose by 12% Y/Y while R&D rose by 8% Y/Y to both greatly lag the pace of revenue growth as well. The EBITDA explosion is the byproduct of all of this and should have the optimists quite good about their point of view.

This was an excellent quarter. It tells all of us that SE doesn’t solely rely on more OpEx to fund revenue growth. It’s showcasing fixed cost leverage and compelling economies of scale. Its consumers are a lot more loyal than some thought than, with this financial data offering clear signs. This company took a lot of heckling throughout the post-pandemic hangover for global e-commerce. It responded by fixing its cost base, shifting its focus, delivering more unique value and executing. The team deserves a ton of credit.

A link to my most recent earnings review can be found here.

Lemonade talked a lot about leading with price when it went public a few years ago. Since then, the theme was put more on the back-burner, while the idea of wanting to offer cheaper plans was maintained. During this event, price competition again became a core topic. And why do they care so much about this? Because for every 10% discount they can offer to customers, they enjoy a 50%-150% boost to plan conversion rates. That allows them to harvest more overall profit dollars, thanks to the overall growth tailwind offsetting the lower fees. At the same time, intentionally lower premiums per plan also (all else equal) means a higher loss ratio.

Lemonade knows that conversion gains make accepting this concession more than worth it. This is a choice it will likely make as it remains in rapid growth mode. As that growth eventually slows with scale and age, price is a lever that it could easily pull to drive more long term operating leverage. It is focused on maximizing profit dollars rather than perfecting short term gross loss ratio (GLR). Considering this, I find the excellent GLR performance in 2024 to be even more notable.

Lemonade’s business model provides two durable cost advantages to make undercutting the competition an even more obvious decision. First, it has no agents. It does pay a 16% premium share to its growth spend financier (essentially cost of debt), but those commissions are finite. They last 3 years until General Catalyst (the partner) is repaid. These are not perpetual expenses. Furthermore, this also means there’s no disintermediary (as leadership constantly reminds us) between Lemonade’s platform, brand and the end consumer. So? It has more open access to all of a customer’s relevant data. This means it knows when to surface a cross-sell offer at the perfect time; it means the firm can augment claim handling speed/accuracy and can further personalize experiences. Generally speaking, all of that leads to higher lifetime value (LTV) without added customer acquisition cost (CAC).

Lemonade’s architectural design makes up the other inherent cost advantage. This isn’t “60 stitched together systems” that competitors like Geico try to glue together. This platform doesn’t need expensive system integrators to slowly, painfully conduct any small tweak to the offerings. It doesn't have massive rosters of insurance agents who understandably fight incumbents as they try to modernize at every turn. These agents want the status quo and that makes it harder for their parent companies to change it.

Compare that to the Lemonade’s singular operating system (called Blender). From this scalable, overarching, light-weight core, it feeds every single model the needed data, context and instruction to work. From Jim (customer service model), to lifetime value models or Cooper (internal work automation model), it helps everywhere. Lemonade ships 50 software updates per week, with no disruption and with rapid, data-driven learning to constantly guide product road-map. It built its platform the right way and it can now sprint faster than anyone else.

This idea is somewhat subjective and abstract, so let’s attach some numbers to it. Since 2021, Lemonade’s headcount has compounded at a 2% clip while its top-line grew at an average pace of 25%. It is quickly approaching $1 million in premiums per employee, and sees that 4Xing over the next 6 years to depict its continued plans to rapidly compound top line with little added cost. And for Lemonade’s internal work, since 2022, the company has pocketed a full $120 million in OpEx.

This is already an instrumental part of cost leverage, with expectations of OpEx savings ramping to $1 billion over the next four years. Notably, as leadership put it, this industry isn’t like airlines, for example. Airlines cut costs by giving you less leg room or “taking away your peanuts.” Lemonade cuts costs by creating more rapid, more convenient, more automated and more delightful interactions.

It’s in a prime position to offer the best of both worlds: better service and lower expenses. You get the peanuts and the leg room for less money. For Lemonade’s business, this means a unique combination of value creation including:

Lemonade aggressively front-loaded fixed cost at its birth to set itself up for decades of efficient scaling. While that led to bloated losses that were easy for investors to pick on, it now positions them ideally for low-cost expansion. Just like for SoFi in banking or Uber in ride sharing, it’s these sticky cost advantages that yield differentiation within a fiercely competitive field. Lemonade surely augments these edges with real-time claims handling and rapid onboarding that incumbents can’t match, but lower price is the biggest factor. Many will endure an antiquated, slow, frustrating interface for some savings. Nobody will put themselves through that just to pay more money at the end.

One more quick note on AI, automation and cost controls. While Lemonade is determined to keep headcount growth slow, it is not interested in laying off its remaining talent. It did have to rightsize headcount from over-hiring through the pandemic (like everyone else), but new GenAI innovation will not mean more job losses… just slower headcount growth. Lemonade’s claims handlers, for example, are simply shifting their roles from manually handling customer inquiries, to perfecting GenAI prompts, testing those prompts and deploying them to millions of interactions. The team took us through a demo of its Blender model and how employees interact with it on a daily basis to ensure responses are always becoming more complete & valuable. This is why the proportion of inquiries handled in a fully automated fashion continues to rise (chart below). This paves the way for lower relative headcount needs vs. other competitors.

The Pet Insurance story is remarkably positive. It has gone from $12 million to $254 million premiums in 5 years while gross loss ratio fell from 110% to 69%. It’s getting better & better with converting compelling customers and avoiding un-compelling cohorts.

Success is related to its AI-fueled automation paired with human touch in action. Jim (customer service model) can automatically approve items for coverage from unstructured screenshots & PDF documents. It can also rapidly ping a coverage specialist to handle issues with all needed context to handle issues in minutes rather than days. All of this means moving insurance applications from weeks-long, paperwork-packed ordeals, to rapid, friendly service. 

For Pet insurance alone, AI help has cut $60 million in LAE expenses and moved variable costs per claim from $65 to $19 since 2021. It did all of this while maintaining best-in-class customer service scores because, again, cost-cutting automation means better service too. These benefits will merely grow with scale and model improvements.

This execution also offers a small positive hint for the car insurance story that we’ll discuss next. Why is pet relevant to car? Don’t you still travel via horseback? Me neither. Interestingly, claims frequency for both products is much higher than other plans. This allows Lemonade to more expediently collect data for model seasoning. And if the last few years of pet insurance results is any indication at all, car should be able to find a good path too.

Lemonade is now the 4th most searched pet insurance vendor a few years after launching. As COO Adina Eckstein was quick to note, they got here with 5% of the marketing budget of Progressive.

Telematics Definition: Lemonade Telematics uses an app to track every single driver action – from a tap on the break to a dangerous text on a highway.

Car insurance is a large part of Lemonade’s growth engine vision over the coming decade. It has gone quite slowly with integrating Metromile, upgrading underwriting algorithms, investing in needed telematics processes and waiting for needed premium hike approvals from regulators. Those approvals have come. And? Lemonade thinks its product feature set and underwriting models are ready for primetime growth in 2025.

Lemonade thinks this product is going to come with far better pricing for its target customers than anyone else. Why? Because telematics is unique to Lemonade. Not because every other incumbent doesn’t claim to also use this technology, but because Lemonade does it better. Two reasons for this. First, its app-based telematics software is perpetual. It doesn’t stop tracking drivers after a few weeks or months. Other leaders here like Progressive do. Leadership was quick to note that while 9% of car policies in the USA do use telematics, about 1% use perpetual telematics vs. 92% for Lemonade. That matters a lot, as median policy prices change by about 23% following the typical tracking window. The always-on tracker also perfectly positions Lemonade to immediately offer service when needed. They know when that person behind you rammed into your bumper and can help you initiate a claim with little paperwork in seconds.

Secondly, Lemonade rightfully calls its telematics algorithms proprietary. Aside from one other competitor (I think Root), all other telematics users pull from the exact same algorithm templates to power their offerings. That’s what Lemonade also used for version one of its product and has since moved to custom algorithms that have delivered large observed performance gains.

A better, perpetual telematics approach will, per the team, help Lemonade routinely undercut competition by 15% for 66% of customers and by 25% for 25 of them… in fewer than 15 minutes. It was quick to point out that Geico built an insurance empire on this marketing ploy.

How can it drive such powerful savings vs. the most competitive alternatives? Aside from all of the inherent cost edges we’ve already covered, its superior telematics software for this specific product is the secret weapon. Here’s how Lemonade puts it. The algorithms uncover the ⅔ of car policy holders that are paying too much and “subsidizing the other ⅓.” Legacy underwriting treats customers as far more monotonous than they truly are. Granular underwriting reveals their true risk and will tell Lemonade exactly who to pursue at industry-leading rates. Furthermore, retention rates for cross-sold car policy holders is 70% better with loss ratios 10 points lower. This merely feeds its ability to undercut other prices.

Over the next several years, if the company executes here, pulling away those over-payers will drag up the mean risk scores for incumbent insurance books. That will make average premiums for all others naturally rise and will create a larger pool of the market that Lemonade can undercut profitably. They didn’t mention this during the event, but to me this is intuitive.

All of this is exciting, but there’s still a lot to prove here. Lemonade needs to show the world that they can effectively underwrite car policies at scale in 2025. They’re very confident, and if they’re right about this, like they’ve been right about everything else, they’ll be right. A 700,000-person car insurance waitlist (wow) offers screaming evidence that they are. The existing customer cohort expressing rising interest in this offering will be the first to be pursued next year.

As a reminder, Lemonade uses General Catalyst (GC) to finance 80% of its external growth spend in exchange for a 16% share of premiums. This is in place until GC is repaid with interest within 3 years. This means Lemonade’s growth spend is not a cash drain, which is why cash flow margins lead income statement margins by such a large amount. For context, it’s already FCF positive as of this quarter, but over a year away from an expected EBITDA inflection. All of this was to lock-in sustainable balance sheet health while it remained deeply unprofitable and still needed to fund its growth. As leadership reminds us, other insurers use actual agents to bear the risk of pursuing growth and then they share premiums with those agents. Lemonade is doing something very similar, but with GC financing the growth.

They took us through an example of how this arrangement has helped them in the near term:

As you can see, Lemonade is forgoing a bit of the premium upside to give it a lot more ability to lean more aggressively into growth. It’s widely thought that Lemonade will use this arrangement on a temporary basis and shed the contract as it becomes more profitable and more able to use its own balance sheet to spend as aggressively as it needs to. For now, this gives them the ability to either greatly boost growth spend or greatly diminish balance sheet usage. In reality, it practices a healthy balance between these two ideas.

The company offered an illustration of how meaningful an inflection point they currently are sitting at. The next doubling of its top line will yield a roughly 19x in incremental cash flow contribution. This is amid a stable expected benefit from synthetic agents, and so is a positive byproduct of cost leverage expected from every other bucket.

Lemonade shared an excel spreadsheet really getting into the weeds of the impact. We’ve already covered all of the important points. If you’d like to get further into the weeds, you can find the model here.

While growth spend is financed (so not a cash flow statement expense) it’s still an income statement expense and will become a source of cash use over time if Lemonade cuts the synthetic agent agreement. Encouragingly, it’s showing a real ability to lean back into more spend here (as expected) with a stable and strong 3x LTV/CAC ratio. Specifically, it will spend $121 million in 2024 vs. $55 million in 2023 with zero deterioration in return metrics. This shows us how much opportunity it has to find compelling, profitable growth. It plans to keep accelerating spend here in 2025 and 2026 as well. And to me, I support this wholeheartedly.

Its tiny marketing budget vs. industry titans has already made it the 2nd most-searched renters product on Google (and again 4th for pet). Furthermore, 70% of its customers are younger than 35, and so come with very long spend runways. Finally, its smaller budget has still yielded the large market share gains seen below. If it can do all of this with the spending handcuffs voluntarily on, imagine what it can do as it unleashes this growth engine.

One more note here. I do think it’s somewhat fair for bears to complain about us bulls using cash flow metrics here. Growth spend is a large part of the OpEx here and this ignores that reality. At the same time, I don’t think this matters. The cash flow build they expect in the coming years is only possible if EBITDA and net income ramp up equally meaningfully (just on a several quarter delay).

All of this leads us to Lemonade’s updated financial targets. Before we dig in, a bit of context. As I frequently praise, Lemonade’s ability to control its growth and margin… for a company this small and young… is not normal. It can rapidly decide when it wants to grow more quickly or deliver more operating leverage. It’s still on the exact same path to EBITDA that is set in 2021; it just shattered all goals set at its 2022 investor day; it has beaten guidance in all 17 quarters since going public. I say all of this to lend credence to innately uncertain multi-year forecasts. While these forecasts are always risky, this team has a uniformly positive track record of delivering.

With that said, it raised its 20% annualized premium growth target to 30% annualized growth. It thinks it will carry this 30% CAGR all the way through 2033 to reach $10 billion in overall premiums. When doing some back-of-napkin math, CFO Tim Bixby alluded to an exit EBITDA margin of 12% of premiums (higher % of revenue). It reiterated its path to positive EBITDA by the end of 2026 (could be there today if it cut all elective growth spend) and positive GAAP net income by the end of 2027. That’s despite its intention to spend a lot more on growth. So let’s just assume it gets to an 8% EBITDA margin in 2029 and meets its 30% growth goal.

That would leave us with about $270 million in 2029 EBITDA. Assuming a 20x EV/EBITDA multiple, that would give us a 2029 enterprise value of $5.4 billion and a return CAGR of 23%.

I’ll refrain from getting more aggressive in my assumptions here, but I do think that aggression would be somewhat warranted. Why? First of all, the 30% target is an “at least” guide. Furthermore, this 30% CAGR through 2033 assumes that it takes 1.5% of the car insurance market. It took pet and renters market share a lot more quickly… without an existing base of clients to cross-sell to.

Final notes here – it thinks adjusted gross profit and cash flow are the metrics to focus on. Adjusted gross profit reflects underwriting expenses and cost to service claims (LAE). Cash flow reflects all other variable expenses. I’d also really encourage all investors to add EBITDA and eventually GAAP net income to the fold there like I will continue to do in all of my coverage. Again, growth spend will always be part of this business model. Cash flow is being uniquely held up. It’s not at all shady, just worth noting.

The company offered early 2025 targets calling for $1.2 billion in premiums (27% Y/Y growth to mark more acceleration). It sees growth ramping to 30% through the end of the year and staying there in the years ahead. It sees 5% OpEx growth ex-growth spend, which I believe equates to about 13%-15% total Y/Y OpEx spend assuming growth spend rises modestly as a proportion of overall OpEx.

I don’t think much needs to be said here. I’ve long called this company the puppet master of controlling their growth and margin. I’ve long praised this unique talent that no sub-10-year-old insurer should already possess. I’ve defended this company as the stock spent years doing absolutely nothing. Why? Because they continued to show such clear signs of strong fundamental progress. This week is the coming out party and the moment many more are realizing “oh, I guess this company is real after all.” Gratifying, to say the least.

While I trimmed a small piece of the stake last week following the rapid doubling in the stock, I have no plans to do so again. That could change if the stock keeps moving parabolically higher. This is becoming less speculative by the week and I’m comfortable with it growing to a larger portion of my portfolio than I previously was.

Nvidia designs semiconductors for data center, gaming and other use cases. It’s unanimously considered the technology leader in chips meant for accelerated compute and Generative AI (GenAI) use cases. While it specializes in chips, it does a lot more than that too. Its toolkit includes chips, servers, switches, networking and cutting edge software. It designs the entire next-gen data center layout with slick software integrations so customers can enjoy the best of accelerated compute. Nvidia calls these data centers “AI factories.”

The following items are important acronyms and definitions to know for this company:

DGX: Nvidia’s full-stack platform combining its chipsets and software services.

Chips:

Connectivity:

Software, Models & More:

EPS is expected to grow by 47% next year and by 22% the year after. I think estimates will be stable following this report.

H200 powered rapid Y/Y & Q/Q growth at massive scale. Whether it’s demand for large language model (LLM) training and inference, chatbots, content delivery personalization, robotics or any other GenAI use case, business continues to boom. Supply bottlenecks have still not vanished, but they’ve improved materially. And as improvements play out, customer demand is eager to buy up the added availability. H200 ramped to “double digit billions” in quarterly revenue, representing the “fastest product ramp in company history.” This product doubles inference performance and efficiency vs. the already strong H100 offering. As of right now, H200 is available through AWS and Azure (and CoreWeave), with Google Cloud and Oracle Cloud Infrastructure (OCI) “coming soon.”

Consumer internet companies like Meta contributed to 100%+ Y/Y data center growth, while Sovereign AI demand continued to soar. Still, cloud service providers powered half of all this demand. That strength should be sustainable, as long as the return on investment for using and renting out Nvidia GPU capacity remains as strong as it is. As of last quarter, every $1 spent by these players (Amazon, Microsoft, Alphabet, Oracle etc.) netted $1.25 in revenue for four years. That means a sub-1-year payback period, which is immensely easy to invest in. This is why CSPs are such amazing partners for Nvidia. These companies represent aggregated demand from entire industries, as they can easily rent out excess capacity to customers at sky-high returns and organize more Nvidia demand.

“CSPs deployed NVIDIA H200 infrastructure and high speed networking with installations scaling to tens of thousands of GPUs to grow their business and serve rapidly rising demand for AI training and inference workloads.” 

CFO Colette Kress

Within the GenAI infrastructure opportunity, compute revenue rose by 132% Y/Y to $27.6 billion, while networking revenue rose by 20% Y/Y to $3.1 billion. Compute continued to rapidly grow on a Q/Q basis, but networking was down Q/Q for the first time in three quarters. Within networking, InfiniBand and SpectrumX did continue to sequentially grow, while leadership called demand for this slice of the market “strong and growing.” Q/Q growth is expected to resume this upcoming quarter.

“Every customer is racing to be first to market.”

CFO Colette Cress

All 7 custom Blackwell chips are in “full production” and NVDA shipped 13,000 GPUs to customers during Q3. Demand was called “staggering.” Sales will viciously ramp from here, but Blackwell is still expected to be supply-constrained at least until next summer. Supply constraints should mean that it keeps selling everything it can make at a sky-high margin.

Where is overwhelming demand coming from? The same thing as always: fundamentally superior products. Recently, MLPerf concluded that Blackwell yields a 2.2x performance leap over its H200 chips (which already led markets). Better performance means more scalability at lower cost. For example, OpenAI will use Blackwell to cut costs to run its GPT3 models by a factor of 4.

“Just 64 Blackwell GPUs are required to run the GPT3 benchmark compared to 256 H100 or a 4x reduction in cost.”

CFO Colette Cress

Separately, the recent production issues that were leading to lower-than-expected yields have been implemented, and the volume recovery has played out as expected. As a reminder, this drives 30x inference speed gains vs. predecessors. As competition races to catch up to Hopper, Nvidia is already leaping well beyond Hopper performance with Blackwell. And? The Rubin Platform will come after Blackwell to yield more explosive improvements. Competition is sprinting to keep up with Nvidia. AMD has even moved to a one-year platform update cadence to match Nvidia. And while that’s great, the issue is that Nvidia continues to sprint as well. 

Leadership was asked about recent reports this past weekend pointing to overheating issues from Blackwell. Per the team, there are no production issues to note. Raising guidance for 2024 Blackwell revenue says they’re right.

Jensen was asked about the debate on length of runway for large language model (LLM) scaling and performance gains, which leads us to an important point. The CPU to GPU transformation was made necessary by pace of CPU performance gains greatly slowing. Next-gen applications and models needed to process a lot more data and perform more complex tasks, and CPUs could not keep up with those rising needs. They still work for other functions, just not GenAI models or apps. GPU performance gains are still nascent, and that’s vital considering new large language models will need 10x the compute capacity of existing products on the market.

Considering model advancement will drive Nvidia’s own demand and pricing power, this topic mattes a lot. According to Jensen, there are three ways of scaling models. Pre-training (adding data to models), Pos-training (retraining deployed models with more data and “reinforcement learning”) and Inference Time Scaling (making models think longer to produce better answers). All three of these opportunities are fostering excellent demand growth for Nvidia’s accelerated compute hardware. And per the team, all three methods (including pre-training) have very long runways to drive more improvements.

There are two main pieces of the Nvidia software opportunity. The first one is Nvidia helping to build applications like full-earth digital twins or drug discovery accelerators to help companies make the most of its world-class infrastructure. 

The second part of the opportunity is equally important and often overlooked. Nvidia’s leading hardware and large install base naturally invite more developers to want to build within this ecosystem. Traffic is what creates opportunities for these developers to get paid. All of this work includes software algorithm upgrades that extract more performance from its chips. Specifically, advancements here have 5Xed inference capabilities for Hopper Y/Y. It has a new NIM release planned to boost Hopper Inference performance by another 2.4x. Software-based vertical integration doesn’t just create more cross-selling opportunities to drive customer stickiness and loyalty. It also creates larger hardware performance leads to inspire even more loyalty.

Moving on, I think agentic AI is an especially interesting opportunity here. This form of AI is more goal-oriented and, in a way, unleashes algorithms to complete a complex task in a more autonomous, less instruction-oriented manner. This will push AI agents from recycled chatbots that create more frustration than value, to something that can be truly delightful and convenient. It has an operating system called Nvidia AI Enterprise (includes NeMo, NIMs etc.) to help companies pursue agentic AI goals. Revenue for enterprise AI is set to 2x Y/Y, as Salesforce, Servicenow, SAP, Nutanix and system integrators like Accenture launch new agentic programs with it. Accenture is also using this platform to internally cut steps to marketing campaign completion by about 30%.

Nvidia’s strong support throughout the software and app creation layer matters a lot. It raises retention rates and means it will find monetization opportunities even after the GenAI hardware monetization boom shifts to software and apps. No other GenAI hardware darling does remotely as good a job of combining proprietary software with cutting-edge hardware.

For software within the Industrial AI opportunity, its Omniverse product is already creating tangible value. This allows companies to build digital twins and run massive simulations of factory workflows, healthcare outcomes etc. It enables powerful testing in a zero-stakes environment to turn trial and error into well-intentioned operations. For Foxconn this quarter, it cut energy needs by 30% in Mexico.

This was an excellent quarter. Nvidia is a world-class organization with one of the best founders this planet has ever known. This performance simply marks more masterful execution. Rapid growth at massive scale and elite margins continues to be exceedingly impressive. Still, over the last two years, Nvidia has trained Wall Street to expect jaw-droppingly amazing beats and raises. It has delivered a string of quarters that financial history books will rave about.

Tonight’s results carry a normal level of greatness that any other company would love to deliver to their shareholders. But Nvidia isn’t every other company.

As we exit this quarter, I think it’s important to revisit two core questions. First, is it leading the technology race? The answer is an obvious yes to me. But secondly, how long will massive sector demand growth continue amid this GenAI explosion. The small guidance beats will push sell-siders to question if this cycle is starting to show signs of slowing down just a tad. And it will also make those analysts less emboldened to issue aggressive positive upward profit revisions for 2025 and beyond. Nvidia will continue to dominate this technological wave… but how long this wave can continue to accelerate just got a bit less certain.

Snowflake’s overarching platform is called the Data Cloud – a “single data foundation to eliminate data silos.” This infrastructure unlocks the ability to affordably store, organize, query and learn from data sources at gigantic scale. It offers these services with elastic compute capabilities to allow for flexible scaling up and down of usage. The architecture naturally separates the functions of data storage and consumption, unlike legacy data warehouse solutions. That means data consumption capacity is untethered from public computing resources. This removes the computing capacity bottleneck and enhances the scalability of data storage.

Under this framework, I can store as much data as I want without the requirement for immediate processing. That processing utilizes computing capacity. In Snowflake’s case, the storage is done in a centralized data repository in the Snowflake Data Cloud. It’s processed only as needed. Data is utilized virtually, which removes the need for dedicated hardware. This scalable (or “elastic”) reality limits waste and cost. Snowflake does all of this for clients in a managed fashion to minimize client talent and infrastructure needs. There are a few key products to know & track:

The Snowflake Data Warehouse is where structured data is stored and (on command) processed. Structured data is formatted data. It’s utilized for record keeping and report creation. Data can be easily fetched via structured query language (SQL).

Snowflake Data Lake does what the warehouse does for unstructured data. Unstructured data is unformatted and used to uncover new insights and patterns.

“Snowpark” is its application-building platform. It frees developers to work with data in any source code language. With it, developers can process and visualize data (through Snowpark functions) and build apps (through Snowpark Native Apps). Snowpark is their data-equipped playground to build new things. GenAI models are voracious data consumers. Snowpark Container Services allow GenAI models to run closer to the data that they require. This enhances performance and expedites model training. Movement of apps, workloads and developer attention from Apache Spark to Snowflake is a key source of growth here. 

Cortex AI is another important new product. It’s what Snowflake calls its “AI layer” and offers a slew of GenAI-powered tools to (as Snowflake always says) bring AI, application-building and analytics right “to a customer’s data.” That conjoining routinely lowers data transfer and storage costs. Cortex AI offers unstructured text summary, sentiment analysis, helps beginners write SQL etc.

Snowflake data sharing is its secure product for, as the name indicates, sharing data among the rest of Snowflake’s participating users. As more opt in, a compelling network effect of relevant data builds and Snowflake’s value proposition deepens.

Snowflake’s revenue model is consumption-based in nature. This means visibility compared to SaaS business models is not as strong. It also means customers can more easily scale down (or up) usage when times are bad (or good).

They didn’t talk about guidance for next year. Still, leadership did talk about “feeling really good” about it after having “built good muscle in our sales organization” and identifying new workload opportunities through product innovation. This innovation will “move into production” next year, with a “good backlog” creating more optimism.

Note that 18x gross profit is about double the market average. It trades for 192x forward EPS. EPS is expected to compound at a 55% clip over the next two years. I’d expect positive profit revisions to take place following this report.

Snowflake has been struggling a bit as of late. Growth had materially slowed while margins tanked as it aggressively invested to plug product gaps and improve go-to-market. This quarter represents the culmination of that work and a lot of improvement. Importantly, it wasn’t macro getting easier that powered the success, but Snow's execution. It marks the removal of SNOW-specific bottlenecks to allow its still compelling value proposition to be more fully unleashed. Snowflake’s low cost, wildly easy-to-use, interoperable, malleable platform is beloved by its customers. The company routinely allows its clients to shed engineering resources and slash total cost of ownership. It does this while enabling access to virtually all data sources from all corners of the world to be ingested, combined and organized in the secure Snowflake environment.

While it did fall behind Databricks in some categories like Notebooks, this quarter makes it look like go-to-market and execution were the true causes of its temporary struggles.

Its revamped selling system, which more tightly ties product, marketing and sales roles together to create one finely tuned go-to-market machine, was credited as a large part of this quarter’s outperformance. Beyond better selling practices, it’s also working more closely with hyperscalers like Amazon to drive sales. Over the last year, it has collected $3.9 billion in AWS-sourced bookings, which represents 68% Y/Y growth. It’s also starting to improve its somewhat challenging relationship with Google Cloud, but more work to do there.

And while it’s making great progress on go-to-market, tightened focus and sharpened execution were the reasons for its most recent set of changes. Snowflake decided to restructure its teams and cut “redundant management layers” to expedite decision-making. The company also cut lower-performing salespeople and started conducting more frequent performance reviews. It’s also now prioritizing the use of its AI work (discussed later) internally to automate tedious processes and cut more costs. Expense “rigor” will be a larger focus going forward.

While better execution and go-to-market are helping, it has also quickly addressed its pace of innovation and product release cadence. It shipped more product updates this quarter than during all of fiscal year 2024. And to ensure this pace remains rapid, it debuted the Internal Marketplace for Snowflake employees. Like its public marketplace, this allows for open context and data sharing and teamwork within Snowflake’s departments. Better communication always has a way of driving faster progress.

This past quarter, it released Unistore into general availability. Unistore is Snow’s hybrid table product, which can ingest, store and organize both transactional and analytical workloads. It has long been an analytical workload specialist within the data warehouse part of the business, and this unlocks transactional workload demand. Its product support for open source Iceberg tables (called Open Catalog) is enjoying strong adoption. As a reminder, the open source wave is a threat to 11% of Snow’s revenue coming from data storage. The support for this does, however, bring lower friction associated with using all other Snowflake products, as it creates an easy means of using Iceberg and Snowflake products together. This should be a tailwind for the other 89% of its revenue. Notably, it now has 500 clients that have adopted Iceberg tables. Considering storage remains 11% of its revenue, it’s confident in assuming the headwind here will be very small. And where there is any impact, added adoption of the other data engineering products (like Cortex and Snowpark) will offset this. Between unlocking Iceberg storage traffic and transactional workload demand, it has created a $200 million revenue business out of “data that previously would not have been addressed by Snowflake.”

The Cortex AI feature family is enjoying “strong adoption.” It’s now expanding adoption beyond text support to also work with images, audio and video. All in all, 3,200 customers are using 1 or more AI/ML features. 

“Cortex is starting to take off. It's still very much in the early innings, and we're very optimistic of what that's going to do in the future based upon what we're seeing.”

CFO Micahel Scarpelli

The key risk surrounding Snowflake over the last year was based on people questioning its positioning in GenAI. It is cheaper to use the same vendors and systems for data and AI than separate, siloed vendors, but was Snowflake doing enough on the product side to win over more customers? All of these product launches and faster-than-expected adoption say yes it is. It’s effectively layering on AI services to modernize and future-proof its data architecture core. This, in turn, means more of the data will stay in and flock to Snowflake, as it gives customers more use cases and value within its ecosystem.

Positioning here was the exact opposite of Nvidia. Sentiment was deeply negative, estimates had come way down and this company’s long-term competitive positioning was being questioned. This quarter was much better than feared and benefitted from easy positioning, but I still think these results were quite positive overall. Everything that needs to be gaining traction is doing so.

Product innovation is beginning to show signs of working and its heightened cost controls added some dearly needed margin relief. I think optimists can argue that this company is turning a corner and establishing itself as a core data platform in the GenAI era. I’d cautiously agree with that idea. The stock remains very expensive but the fundamental picture is brightening and the ship appears to be stabilizing.