Content already sent this week:

Read 30+ more detailed earnings reviews from this season on Meta, The Trade Desk, Amazon, AMD, SoFi, Shopify, Airbnb & so many more companies here.

Broadcom creates & manufactures a slew of semiconductor-related equipment within data center, networking and industry-specific use cases. Chips and high-performance compute (HPC) can’t all be packed into the same corner of a data center. GPUs must be able to connect to one another to drive better bandwidth and performance, with faster, more efficient model training and inference to cut costs. This is where Broadcom thrives. 

It also offers a range of software tools, which significantly broadened out with its VMWare acquisition. VMware offers virtual, localized layers of software that sit on top of hardware. This allows the centralized hardware to run several different operating systems from the same place. The company, which is now a Broadcom unit, calls these “virtual machines” or virtual private clouds. By reducing hardware requirements, VMWare saves its clients money. 

This company does not compete with Nvidia in terms of designing GPUs. It does, however, create application-specific integrated circuits (ASICs) for more specialized workloads. It also makes variable processing units (XPUs) which are the high-performance accelerator subsection of ASICs. All XPUs are ASICs but not all ASICs are XPUs. These are often used to optimize data center, networking and GPU performance. In some cases, this can replace various needs for more generalized chips like GPUs. Furthermore, its core niche focuses on networking and connectivity, which competes with Nvidia’s switches and its SpectrumX networking product.

Broadcom’s EPS is expected to grow by 36% this year and by 18% next year.

Semiconductor Solutions – AI:

The AI portion of this bucket grew by 77% Y/Y to reach $4.1B and materially surpassed expectations. It sees this momentum carrying into Q2, with expectations of 7% Q/Q and 44% Y/Y growth to reach $4.4B. Networking performance from its Jericho and Tomahawk switches was excellent and so was its XPU result on the compute side of things. And despite all of the DeepSeek noise, Broadcom is still seeing “hyperscale partners invest aggressively in their frontier models.” Interestingly, most of that demand is still for pre-training, as the shift to inference brought on by training deflation and reasoning models hasn’t dramatically shifted its own demand to inference. Importantly, it thinks its products are well-positioned to cater to both use cases.

To stay ahead of the competition and nurture demand growth, it’s accelerating R&D spend in two areas. It’s pushing full speed ahead to tape out (finish a design concept) the very first 2 nanometer XPU so it continues to pack more compute into each chip. Fewer nanometers enable higher chip density and productivity. It’s also working to double Tomahawk 5 capacity to enable more chip connections, larger, more flexible compute clusters and better productivity. Its Tomahawk 6 switch is also now taped out to “enable AI clusters to scale to 1M XPUs.”

These investments are well-placed, considering fantastic hyperscaler demand. As a reminder, it has 3 larger hyperscaler customers (such as Alphabet) that it expects to deliver a serviceable addressable market (SAM) of $75B by 2027. That’s more than its current annual revenue base and it feels poised to capture the bulk of it, based on its 70% market share of the related hardware. Last quarter, it talked about two new hyperscaler customers and it expects to tape out XPUs for these clients this year. But wait, there’s more. During the quarter, the company began “deeply engaging” with two additional hyperscalers. Importantly, the 4 newer prospective clients are incremental to its $75B SAM estimate.

Importantly, these 4 companies are not yet customers. These are just design wins for now. In the world of semiconductors, design wins are often a dime a dozen, with actual deployment much more rare. But? Broadcom is unique here. It is highly picky with the type of client it will build XPUs with. It only works with the highest-volume vendors and only does these for vendors with a strong likelihood of large-scaled deployment.

Where is all of this hyperscaler demand coming from? Hardware superiority. Broadcom is convinced that its best-in-class hardware paired with its customers’ “excellent software” is a winning combination for creating custom XPUs to optimize performance of their infrastructure. Whether that’s optimizing for capacity, GPU utilization, memory, latency, resource allocation across training and inference or anything else, Broadcom is world-class in this regard. Hock considers these 4 prospective companies to be “partners” at this point in time. They’ll become customers when AVGO begins shipping to them at volume. Time-to-market for projects like this is somewhat lengthy. Moving from concept to design to production often takes a couple years, but the company does think it can turn these partners into meaningful revenue before FY 2027.

“In the process of working with the hyperscalers, it has become very clear that while they are excellent in software, Broadcom is the best in hardware. Working together is what optimizes via large language models.”

CEO Hock Tan

Semiconductor Solutions – Non-AI:

Non-AI revenue within this bucket fell 9% Q/Q. The recovery for this cohort was called “slow.” In terms of subsections, broadband rose by more than 10% Q/Q due to strong telco spend levels; server storage fell a bit Q/Q, but should grow sequentially in Q2. Enterprise networking was and should remain flat Q/Q. Wireless drove most of the weakness here as expected. Overall for non-AI semi revenue, it sees 0% Q/Q growth next quarter. Encouragingly, bookings growth will remain strong on a Y/Y basis, which points to potentially accelerating growth later in the year.

Infrastructure Software:

Growth was temporarily aided by two things this quarter. First, it is the final quarter of inorganic contribution from VMWare in the Y/Y comps. Second, several large deals slipped from Q4 to this quarter, which were signed as expected.

Broadcom continues to do very well with VMWare integration and cross-selling. Part of this work has meant converting from a license-based business to a subscription-based business, which is 60% complete. The other major part has been up-selling customers to the VMWare Cloud Services (VCS) package. This “enabled the entire data center to be virtualized and customers to create their own private cloud environment on-premise.” It’s no longer predominately virtualizing CPUs in isolation. It is virtualizing CPUs, GPUs and every other piece of the data center. 70% of its 10,000 largest customers are now using VCS vs. 45% Q/Q.

The product that virtualizes GPUs is called VMWare Private AI Foundation. This was announced last May with Nvidia, and now has 39 enterprise customers. Broadcom sees continued healthy demand for CPU bookings (VMWare’s main financial driver), and sees a great opportunity to do more with high-performance chips too. It also continues to prioritize on-premise service with VMWare Private AI Foundation alongside public cloud deployments as well. It thinks the GenAI explosion will keep data privacy and security more important. It thinks this will slow public cloud workload migrations as some enterprises “recognize they want to run AI workloads on-premise. That’s not something we often hear from others.

“Customer demand has been driven by our open ecosystem, superior load balancing and automation capabilities that allow them to intelligently pull and run workloads across both GPU and CPU infrastructure and leading to very reduced costs.”

CEO Hock Tan

Another strong quarter for a leader in GenAI connectivity, custom chip design and hardware optimization. The firm admirably integrated VMWare with little headache and continues to find awesome traction with large hyperscaler clients. As long as this GenAI hardware boom lasts, and mega-cap CapEx budgets say it will last at least through the year, Broadcom should continue to be a very central part of it.

On Cost Controls:

On Alphabet’s last call, many investors thought leadership walked back previous commitments to cost controls. Per Ashkenazi, that is not the case. There are no changes to the firm’s commitment to controlling costs and cutting bloat. That’s where last week’s layoff announcement came from. From 3rd-party vendor contract negotiations to cloud infrastructure optimization to back-office automation and everything else, the firm remains in cost efficiency mode.

Along similar lines, she’s quite confident in the firm’s $75B CapEx budget being put to very good use. This goes back to why its full-stack AI approach – across hardware, models, distributed apps and developer tools – means this CapEx has several productive uses. If one piece of the tech stack proves to need lower budget, this CapEx is highly “fungible” to reallocate elsewhere. These dollars will go to waste. And for more evidence of that, recall that the majority of this CapEx is from short-lived assets like GPUs and servers rather than long-lived assets like data center construction. Also recall that Google Cloud remains capacity constrained and unable to fulfill all of its current demand. Translation? This CapEx is in response to near-term demand signals and revenue opportunities. This isn’t investing for potential value creation in 5 years… more like 5 months.

On Falling Behind in Search:

She pushed back hard against this notion, citing circle-to-search, AI Overviews, and the launch of Gemini 2 as key examples of rapid innovation. She also reminded us that this AI volume is monetizing at similar rates compared to legacy search, which had been a key concern for investors. Now, it’s clear that this innovation will not cannibalistically come at the expense of its bottom line. AI Overviews are directly raising repeat query demand and its dedicated Gemini app is raising overall engagement too. I’m personally a daily active user (DAU). Furthermore, AI is creating an explosion in query use cases and raising the overall size of the opportunity. So? Just because Perplexity or OpenAI can find a little market share on their own does not at all prevent the search business from also finding sustainable, profitable growth. Near-100% market share is not a prerequisite for that to happen. A smaller piece of a larger pie works just fine.

With 5 trillion annual queries, leading video market share with YouTube and 6 products with over 2 billion users, it has more data to season models more effectively and generate deeper, more precise and more engaging search experiences. With Gemini, DeepMind, global cloud infrastructure and its budding semiconductor business, it has all of the assets in place to leverage this data as well as anyone else can. And, with its thriving advertising business and scaled impression menu, it has a direct means to monetize all of this work. I would also call this capability unique, as shoppers are 98% more likely to trust a recommendation if it’s offered by a YouTube creator they follow and as Alphabet obsessively works to use AI to augment return on ad spend (ROAS).

While many think Google Search is decaying, recent results and company positioning tell a very different story. Isn’t it interesting that this existential risk seems to fade as the stock price rises and builds back as it falls?

Other News:

Alphabet launched AI Mode to expand on its highly successful AI Overviews launch. It offers a full page of helpful links (some shoppable and I’m sure monetizable), “advanced reasoning” and multi-modal, agentic capabilities made possible by the launch of Gemini 2. It’s now available for some Google One subscribers in early access.

MongoDB is a key player in data storage and analytics with a document-oriented setup. This differs from legacy relational-style databases and next-gen versions like Snowflake’s. How so? Legacy relational databases store data in static rows and columns linked by implemented formulas. These databases look like giant Excel spreadsheets and use structured query language (SQL) to work. Legacy relational databases cannot seamlessly handle unstructured data like MongoDB’s data lake can. This is a large limitation, considering how important unstructured data is for GenAI use cases. Legacy relational databases struggle to scale and unlock the most advanced querying. The datasets are fixed, with formatting and filtering more limited. The inability to provide “not only SQL” (NOSQL) can slow performance and diminish value. MongoDB’s NOSQL database and document-style storage fix these issues, which is why the pace of migration continues to ramp up.

The firm’s most exciting product is called MongoDB Atlas. This is a cloud-native database service that uses a group of servers (or a cluster) to actually store data for app creation within its platform. The nature of MongoDB’s product allows clusters to be easily added to or subtracted from for easier tweaking as needs fluctuate. It also offers MongoDB Realm as a mobile environment for app creation and MongoDB Search for data querying. Finally, it offers MongoDB Data Lake specifically to house unstructured data.

“MongoDB removes the constraints of legacy databases, enabling businesses to innovate at AI speed with our flexible document model and seamless scalability.”

CEO Dev Ittycheria

Some more products & terms to know:

Note that most of MongoDB’s growth is based on acquiring new customers and migrating their workloads onto the platform, as well as consumption-based revenue from customers using its data products and growing workloads.

Reminder:

MongoDB pocketed $40 million in unused Atlas revenue and another $40 million in multi-year non-Atlas licensing business last year. That $80 million in pure margin revenue led to an $80 million revenue headwind and Y/Y profit declines assumed in its 2024 guidance. Keep this in mind as we go through its financials and why they look underwhelming.

Pretty poor annual guidance here. Notably, MongoDB is among the most aggressive guidance sandbaggers in the coverage network. They routinely set the bar ridiculously low every single year to set themselves up for easy outperformance. Some rightfully say its consumption-based model means it must be prudent. But this is consistently in another league of conservatism. They attribute it to difficulty forecasting multi-year non-Atlas deals, which is somewhat fair. But they’re still an outlier in this regard.

The reasoning for the massive profit misses was similar to when it offered initial annual guidance for FY 2025 a year ago. Following “two very strong years” for non-Atlas licensing revenue, it thinks it has a “more limited set of account opportunities.” This will lead to a $50M revenue headwind and a high-single-digit Y/Y decline for this non-Atlas bucket. Excluding this impact, revenue growth guidance would have been about 16% Y/Y instead of 13% Y/Y. Furthermore, this is extremely high-margin revenue, so the bottom-line impact is more severe. During a follow up interview at the Morgan Stanley conference, leadership said this was the source of half of the Y/Y margin contraction in guidance. Aside from this item, R&D related to Voyage, aggressive investments to build out their AI product offering and more marketing budget were other margin headwinds cited. Perhaps this time is different, but if history is any indication, their actual expectations for the year are probably higher than what they guided to. Leadership spoke about temporarily severe margin headwinds before the firm went public when it was building out Atlas. They see a “similar opportunity in GenAI,” and want to invest to take advantage. They’re willing to accept short-term margin pain for a higher revenue and profit ceiling down the road. EPS is expected to fall by 25% Y/Y this year and rise by 29% Y/Y next year.

Go-To-Market:

Over the last few quarters, like Zscaler, SentinelOne and many other enterprise software names, MongoDB has been making various changes to optimize its go-to-market. It shifted focus to larger enterprise contract opportunities. This entailed adding support for a cohort of high-upside strategic accounts coming with especially compelling room for workload growth. It also tweaked sales incentives to prioritize the size of new workload wins and pace of workload expansion. This was in response to disappointing workload growth in FY 2024 and into 2025. Addressing this in FY 2025 put its base of workloads in a much better place. That helped inform its stable Atlas consumption growth guide for FY 2026, as the business continues to scale. Still, per leadership, the lingering effects of poor FY 2024 performance (fewer quality workloads to upsell) led to the 16% Y/Y normalized growth guidance. Keep in mind this is supposed to be a high-growth company — and is priced as one.

Finally, the company sought to be more vocal about all of the flexibility, scale and other perks associated with building NOSQL apps. It thinks these changes helped drive better-than-expected Atlas consumption and what it viewed as a strong pace of new workload acquisitions during the quarter. I would have loved to see that coincide with upside in $100K ARR customer growth, but outperforming Atlas customer growth is positive. Per leadership, sales productivity rose throughout the year and it sees that continuing.

Demand Trends:

Q4 delivered strong retention and better-than-expected (shocking) consumption growth for Atlas. Specifically, that growth was stable compared to last year. It was expected to slow a tad. As briefly mentioned, it expects this consumption stability to continue through the upcoming year. That confidence is informed by go-to-market fixes beginning to bear fruit, as realigned incentives drive more workload growth focus and a shift to bigger deals yields higher returns. This is creating a larger base of workloads to drive more expected consumption growth this year. So far in fiscal year (FY) 2026, stability has played out, with leadership expecting Atlas to approach $2 billion in revenue this year. That represents a little over 40% Y/Y growth.

For non-Atlas demand, the company once again enjoyed a much better-than-expected contribution from multi-year on-premise licensing deals stemming from an Alibaba contract and general segment strength. Outperformance added $10M to the Q4 revenue number. This is the second straight year this has happened, and leadership does not expect this to recur in FY 2026. They said the exact same thing last year, so perhaps they’re being overly prudent like they so often are. Still, it did offer tangible reasons for this pessimism, as it has fewer clients eligible to sign large, multi-year deals this year. It also continues to enjoy a desired shift from non-Atlas to Atlas revenue, which will continue. MongoDB is required to recognize all of the revenue from non-Atlas service upfront. This leads to considerable lumpiness. This quarter, it added a new non-Atlas ARR disclosure, which does a better job capturing run rate growth for this segment. Non-Atlas ARR rose by around 5% Y/Y, with its main focus on supporting the cloud-based operations.

Dev Ittycheria participated in the Morgan Stanley conference after the earnings call. He was asked if he believe the company can get back to and sustain 20% revenue growth. He said “the short answer is yes,” and continued to talk up stable Atlas consumption growth for evidence. Lapping the non-Atlas revenue headwind truly just has a massive impact across the income statement. He was adamant that the core Atlas business remains healthy and stable.

App Modernization & AI Accelerating the Urgency:

MongoDB’s customers remain in GenAI app experimentation mode for the most part. There are some small pockets of customers moving from this phase to more developed concepts and even deployment, but not many just yet. The pace of this ramping is what will support financial upside for MDB amid this technological boom. It expects that to “gradually” play out in the new year with “modest” incremental revenue help.

“I would say that we do think this year is a year of transition. We are really excited about the opportunity in AI, but we also do recognize that customers, especially large enterprises, are moving quite slowly in the deployment of custom AI apps.

Interim CFO Serge Tanjga

Most of the AI use cases are fairly simplistic, chatbots, document management use cases, etcetera. But we are seeing people get very interested. And we think, architecturally, we are well designed for the world of AI.

Over the longer term, the company feels its suite of data & app-layer-based products will thrive as hardware monetization shifts to software. With the pace of technological change driven by GenAI, not embracing this technology means companies will fall behind faster than ever. It’s determined to help make the change less disruptive and intimidating in a few ways. First, it’s great at automating tedious prep work for data and app migrations. This is a prerequisite for making the jump. More recently, it has played a more active role in directly helping companies recode and modernize their legacy applications so they can seamlessly run on MDB’s platform. Its Relational Migrator product is its full-service tool to take the headache out of this often cumbersome process. MongoDB is also investing to support better professional services operations to ensure companies with finite resources have the help they need to purchase its products. This offering is off to a great start, and it has its eyes on Java apps running on Oracle to modernize and bring onto its platform this year.

“In FY 2025, our custom app modernization pilots demonstrated that AI tooling combined with services can reduce the cycle time of modernization. This year, we'll expand our customer engagements so that app monetization can meaningfully contribute to our new business growth in fiscal '27 and beyond.”

CEO Dev Ittycheria

“MongoDB was designed from the outset to remove the constraints of legacy databases, enabling businesses to scale, adapt and innovate at AI speed.”

CEO Dev Ittycheria

Expanding its AI Opportunity:

MongoDB recently bought Voyage AI to augment model trustworthiness. A routine issue with GenAI model outputs is factually incorrect responses or “hallucinations.” This issue creates immense friction in enterprises building apps with these models. Voyage AI is a specialist in surgically precise data retrieval thanks to two main products. Its Vector embedding models organize data as a number sequence to more closely and scalably group data points for easier machine learning and pattern recognition. Next, it offers reranking models that ingest and reorganize materials based on data relevance for a given model input. It sifts through all queried information to eliminate waste and inaccuracies. Voyage’s products are quite popular and loved on Hugging Face, while Anthropic is a customer.

“Integrating Voyage AI's technology with MongoDB will enable organizations to easily build trustworthy, AI-powered applications by offering highly accurate and relevant information retrieval deeply integrated with their operational data.”

CEO Dev Ittycheria

Like most other enterprise software firms in the coverage network, Voyage is about giving MongoDB a broader suite of GenAI services to make its overall offering a better, more cohesive, less “cobbled-together” platform for customers. Now, it offers its document-oriented database (perfect for unstructured data), help for app and data modernization and re-coding, and world-class embedding and reranking models all under one roof. 

MDB will pay $200M in stock and $20M in cash for this company. It announced a $200M buyback to offset the dilution.

Customer Wins & Outcomes:

Good quarter and another annual guide that I view as extremely, perhaps unnecessarily prudent from this team. Their core Atlas product continues to perform very nicely and the lumpy nature of the rest of its business continues to create consistently strange comps and lots of noise. The company’s product suite remains well positioned to capture reliable growth for a very long time, and I think its Voyage M&A was a great decision. This isn’t my favorite or least favorite name in software. While the stock price reaction has been ugly, this quarter and expectations for more non-Atlas-related volatility didn’t sharply affect my views of the company. I don’t like it enough to own it here, but I fully understand those who do.

Supporting Branded Checkout – Create a Competitive Experience

Chriss feels like PayPal has “leap-frogged” the competition in terms of mobile and web checkout flows. I would instead argue that they’ve made rapid progress in bringing the product up to par with the rest of the field, but that’s really all the company needed. When you have the most trusted and ubiquitously available brand in commerce, you don’t need to be better than the other guys. Simply “as good” will do. Specifically, updates to checkout flows have yielded anywhere from 1-4 points of checkout conversion. 30% of the U.S. and a little over 10% of the globe are now on these latest flows vs. about 5% six months ago. It has plans to get to 80% penetration globally by 2027. As this percentage rises, cohorts with access are delivering very solid results and giving the team more confidence in ramping towards 9% branded checkout growth by 2027 while maintaining growth at or above the rate of e-commerce by that time as well.

The exciting thing about this business is that despite having an antiquated experience (especially on mobile) for years, the checkout business is stable, churn rates haven’t spiked higher and monthly active users (MAUs) have continued to engage more and more. This resilience despite a poor offering shows us how much better things can be as they catch up to other alternatives.

“We feel really good about our ability to take a very stable branded checkout experience and results that we posted the last few years and really start to inflect it.”

CEO Alex Chriss

Chriss sees checkout product parity and also world-class payment flexibility (including crypto) as step one to righting the ship on this core branded checkout business. These were the initial major milestones in its aim to keep accelerating core branded checkout and pursue 9% annualized growth starting in 2027.

Supporting Branded Checkout – Provide More Value:

Step two of supporting branded checkout growth is all about using AI, integrations and its unmatched customer data profiles to drive checkout personalization and superior value. On the merchant side, this utility is accessed through the new PayPal Commerce API. Thanks to PayPal’s increasing willingness to openly integrate with partners, this API gives merchants vast access to rewards programs, marketing tools, and anything else they can think of with one simple API onboarding. Perhaps more importantly, they also get access to 80 million rich customer data profiles. PayPal helps them use these profiles to identify and target customers often “before they’ve even entered a website” and allows them to malleably tweak site layouts based on granular customer preferences. Shoppers don’t even need to be existing merchant customers for PayPal to execute here. This is currently being rolled out to select merchants, with a broad rollout taking place over the coming years. As that happens, Chriss hinted at directly monetizing the commerce API, as well as using it to nurture checkout conversion rates, its advertising platform and broad-ranging merchant service. 

What about for consumers? PayPal’s ubiquitous integration roster and dense customer data profiles mean it can connect to other rewards programs to help customers “stack” benefits when shopping. It also means the company actually knows how to personalize customer shopping experiences in an effective, conversion-raising way. If a consumer has cash-back rewards to use from another financial service… PayPal can plug into them to make sure they’re saving the customer more money. If a shopper is an Amazon Prime member or loves to use BNPL, PayPal will automatically tell a shopper they’re eligible for free shipping and surface a BNPL button for them to use. If PayPal knows you love the color yellow and need a new rug, they’ll make sure you see it earlier in the shopper journey from one of their merchant partners. If PayPal knows the shopper loves using their button, they can actively support their own placement in real-time, to get higher market share and higher merchant conversion.

All of this makes PayPal more valuable to consumers by allowing us to pocket more rewards and enjoy curated experience. And? Merchants are happy to accept the higher usage of reward redemption in exchange for more volume.

Brand Checkout Competition:

“It feels like every few months, there's another rumor about some competitor that's coming in and potentially taking branded checkout market share. The rumors so far, as far as I can tell and [based on] all the data that I look at, are nonsense. I can tell you, we look at the data very, very closely. We look at it on desktop and mobile. We look at competitors. And we have not seen any degradation in our share of checkout from competitors launching products.”

CEO Alex Chriss

Why is Checkout Modernization Taking so Long?

The pace of checkout modernization is somewhat slow here and won’t be wrapped up for another two years. This has been one of the more frustrating parts of the investment for me, but it’s still a byproduct of how poorly run the company was under old leadership. PayPal was running a web of fragmented, discombobulated integrations with disparate 3rd parties and a clutter of heavy APIs. They were operating fifteen years of integrations with no sunsetting and very little maintenance. At massive, massive scale, one can quickly see how that can become a bit disorganized. They also gave merchants “very little reason to want to upgrade,” which is quickly changing with the modern checkout flow, Pay with Venmo, Fastlane and Commerce API all serving as strong reasons to modernize. As PayPal moves from 30% of its U.S. merchants on the newest flows to 50% and beyond, it will get more aggressive in marketing and pushing for upgrades to expedite the pace of legacy API discontinuing. As an important aside, that should also lead to far better operating efficiency and cost cuts.

Importantly, once this process is finally complete, subsequent updates will become far quicker, more frequent and less disruptive to PayPal’s business. With the new commerce API, they’ll actually be able to move quickly for a change. I realize Chriss is now a full year into his tenure, but this was a complete mess of a company to fix and I continue to think he’s making great progress. I don’t see a market-wide sell-off amid macro and geopolitical concerns as reason to change my mind.

Final Notes:

“And I think as a company, PayPal had been in a bit of the doldrums after being known for many, many years as an innovator in payments and in commerce. We hit a dry spell, and I think we picked that up.”

CEO Alex Chriss

Progress Towards 2025 and Long-Term Targets:

DraftKings is “tracking really well” vs. its 2025 targets. They’re “actually doing a bit better” in terms of sport outcomes (which provides material potential upside) and thinks the year is off to a “fantastic start.” They feel “really good” about guidance and also see “other levers” that could provide more upside vs. estimates. Keep in mind that this guidance does not rely on Missouri launching on time (which it won’t) or states like Georgia legalizing gambling (which isn’t happening). So those are not risks to guidance, but do limit potential upside surprise for this year. Furthermore, the industry-wide acceleration in handle growth that Robbins discussed on the last call has continued into March. DKNG is above where it needs to be on handle growth to meet guidance. It does not need more acceleration and “could even probably handle a bit of a decline and still be ok.” The margin of safety for 2025 guidance seems quite large. Acceleration has a lot to do with getting through the election season and mental bandwidth not being dominated by that event. For more evidence of this being the case, NBA ratings rebounded right after that ended too. 

Now for long-term targets. In Q4 2023, DraftKings set 2028 EBITDA and revenue targets at $2.1B and $7.1B, respectively. As of right now, it thinks it’s “ahead of schedule.” For revenue specifically, it’s a full year ahead of schedule. Again, this guidance did not include potential legalization in states for sports betting or iGaming. Importantly, for DraftKings, the only state that has passed and launched sports gambling or iCasino since the event is North Carolina (it isn’t in Delaware). North Carolina represents just a 6% boost to its sports gambling footprint, which is not nearly enough to explain this outperformance. The two sources of its over-indexing success are far more encouraging than that. It continues to enjoy thriving customer acquisition at lower cost in even its most mature states. The runway for growth everywhere is longer than expected. Secondly, through intentional efforts to improve the parlay menu to better compete with FanDuel, it now sees more upside to its hold rate.

Live Betting:

Jason Robbins hinted at live betting strength beyond what’s baked into guidance, providing potential upside for 2025 numbers. Through M&A (mainly Simple Bet), it thinks it has extended its lead over the field in terms of assortment, uptime and capabilities. While they’ve had to play catch-up for parlays, they’re “playing from ahead on live betting” and see potential strength here beyond what has been modeled. Live betting is now up to 50% of total volume, which is still well below roughly 75% across the pond. Notably, popular North American sports are inherently better for live betting than soccer (more frequent stoppages), which gives Robbins confidence in getting that to 75% over time. And vitally, live betting is proving to be “almost entirely incremental, if not entirely incremental” to handle growth. It has observed low single-digit cannibalization from cross-selling sports bettors to its slot customers, and thinks the impact will be even lower for live bets.

Churn and Stickiness:

There are a lot of sports gambling products. DraftKings continues to quickly slash marketing and promotional intensity. And yet? Customer acquisition is exceedingly strong and churn remains low. Impressively, net handle and revenue retention (which rewards existing customer revenue growth and penalizes churn) are over 100% for all cohorts. That is not something one would expect for a business model like this, and just goes to show how much more loyal consumers are than one may think. 

iGaming:

For the “first time in a few years,” Robbins sees iGaming regulatory momentum. There was a stretch of time when he “didn’t even know which state would be next or how long it would take.” But now, he thinks they’ll “get some iGaming bills done” this year or next year. He also said that he expects a “few sports gambling and iGaming states” to pass this year. It really sounded like he was excited but was trying his absolute best not to get ahead of himself as politicians are anything but predictable. Again… pure upside to 2025 and future estimates and a somewhat inevitable tailwind. Every state has budget issues to address. This is an obvious way to do that on a voluntary consumer activity. The segment continues to grow at a steady 20%+ pace without any new states coming online. And with the top-rated products in the market, it is best-positioned to capture future growth opportunities that are an all but when not if.

Separately, 50% of total DKNG iGaming content consumption is first party. Robbins thinks that leads the industry. This not only makes the company a more unique and differentiated player, but makes it less reliant on clunky 3rd parties. These 3rd parties have not spent the money to build an optimal back-end like DraftKings has. So? They lag and glitch more frequently. Lower reliance on these… well… less reliable partners is another edge for this product.

Finally, Golden Nugget and DraftKings continue to be wonderfully complementary brands. Golden Nugget skews older and more female, while being a more broadly-known brand for the passionate iCasino-only players. They’re deciding whether or not to keep JackPocket’s iGaming business in New Jersey or focus on DKNG and Golden Nugget brands for that.

Taxes:

Everyone loves to talk about the risk of taxes. Nobody likes to talk about how they create more inefficiency while weakening subscale players and driving industry consolidation. That consolidation will accrue to DraftKings and FanDuel. The only way this doesn’t happen is if states like Illinois pass policies that penalize larger scale with higher taxes. That decision is looking like an anomaly at this point, and it has already been able to recoup some of the EBITDA losses from that decision.

“I've kind of accepted that any particular year, we're going to have some good things going on in some states and some potential risks in others and we’ve got to manage both and that's just the reality of being operational in such a wide swath of states.” – Founder/CEO Jason Robbins

More Notes:

DKNG is using AI to automate a little over 10% of its code writing and to handle about 20% of customer service inquiries at this point. It has plans to use this technology in every facet of its operations.

Robbins sees prediction markets as more of an opportunity than a risk. It would open up several different bet types, while DraftKings and Fanduel have fended off several formidable competitors to maintain dominant share. If Barstool and ESPN can’t win the hearts and minds of sports bettors… I don’t think Robinhood will either. I’m sure they’ll find some volume, but I also think incremental business from prediction markets will outweigh that headwind. And? This assumes betting on sports will become a legal part of prediction markets, which is far from certain.

The debt raise (as it told us to expect on the Q4 call) is not from a point of need. They just wanted to get debt markets comfortable and familiar with the company. The deal was 5x oversubscribed, so the reception was warm. DKNG was even able to upsize the deal due to favorable terms stemming from this demand. It plans to use this excess cash to potentially accelerate buybacks.

WhatsApp, Business Messaging & Meta AI:

Meta AI now has more than 700M monthly users, with WhatsApp continuing to be the most popular place to use the product. In places like India and Mexico, where populations live on this app, it is already enjoying strong business messaging traction within its WhatsApp for Businesses app (in beta). Companies are routinely shedding manual customer support needs and improving interaction quality simultaneously. Going forward, Meta plans to use the increasingly multi-modal Llama 4 model to greatly enhance use cases here with the aim of expanding use cases and adoption across the globe. Specifically, it wants to get better at letting customers submit information via image, with an ability to automatically ingest and transcribe that data for rapid model outputs. It also plans to add more audio features for these customer service engagements, so companies can field queries in any language with ease.

Early Meta AI Use Cases:

Meta AI is gaining popularity in requests for more context on a given item or topic that users are exploring. This use case integrates perfectly with its Ray-Ban glasses, to ask Meta AI to translate a sign or explain a landmark for travelers. More recently, it began grouping and displaying search highlights at the top of page results. This is somewhat similar to AI Overviews from Alphabet. While this use case can be easily monetized, Meta is not focused there just yet. Like always, it wants to build ubiquitous scale and habituation from its base of users. Only then will it monetize with ad placements. For now, most of the direct AI monetization continues to happen within ad campaigns and targeting. It now has a broad base of tools to help automate content creation in addition to running optimal campaigns, which should create new upselling opportunities. Indirectly speaking, Llama is indirectly bolstering Meta’s own discovery, targeting and personalization algorithms to create more engagement and more ad impressions to sell.

Meta AI is also being used in ChatGPT-like ways. Chris Cox (Chief Product Officer) offered an example of his kid asking Meta AI about Dungeons and Dragons and getting responses far more informed than most people can offer. When I hear this, I think next-gen, AI-based therapists, life coaches, consultants and more to keep rounding out the ways consumers can use these apps and to keep boosting usage.

CapEx:

Cox reiterated how flexible Meta’s CapEx spend is. Most of its teams at this point need GPUs and the data center capacity Meta is currently building. If advancements from companies like DeepSeek make pre-training less important, it can easily shift some of these assets to reasoning and post-training needs. Furthermore, given its open-sourced approach, cheaper model innovation is good for Meta. It can take all of this innovation for itself, and use it in its highly popular foundational models to make them more valuable and more accessible. It is not reliant on monetizing training workloads like OpenAI, for example. It wants this field to evolve as quickly as possible, is confident in its dominant positioning and knows industry-wide innovation will make all of its products better.

Main Things to Expect from the Llama 4 Launch:

“The models will interact with voice and audio natively, rather than translating voice into text, sending text to the LMM, getting text out, turning that back into speech and having speech be native. This is a big deal.”

CPO Chris Cox

Final Notes:

On Max:

Duolingo Max is now 5% of total paying subscribers. Notably, Max has been around for a while, but the ramp from 0% to 5% has come over the last few months. It has been a direct response to the video chat tool that debuted late last year. Duolingo fully expects to make the conversations more aesthetically pleasing, engaging and personality-containing, as Lily (the character one talks to) grows up. It’s extremely early here and exciting to think how developments can augment Max adoption and raise subscriber lifetime value (LTV). Joining this tailwind with price optimization in countries like India is expected to support strong bookings growth in the new year.

Between this and its pursuit of rapidly introducing new advanced English content, it is quickly moving towards its goal to teach as well as a human tutor.

Data Moat:

Depending on your source, Duolingo has a 90% market share of online language learners. It is better positioned to capture more consumers as they move their education to digital settings than any company on the planet. This also means it has the most data to more effectively and frequently run impactful split testing for expeditious product improvement. It’s the same playbook Duolingo has run flawlessly since inception. More data means more insight to feed more product iterating, better engagement, intense word-of-mouth growth, generation of more revenue to feed the flywheel more.

In the future, it hopes to become the brand associated with language learning. Like “to Uber” means to take an on-demand ride, it wants people to describe their language proficiency as a “Duolingo 50 in French.” With dominant category positioning, this is somewhat realistic and can make Duolingo the ubiquitous, sticky leader of this quickly growing field.

The Office of the Comptroller of the Currency (OCC) removed language that prevented chartered banks from operating cryptocurrency platforms. There are a few more regulatory hurdles that need to happen before SoFi can relaunch this. While this is a revenue opportunity, it will likely remain a small one for quite some time. It was a very small business before getting shuttered, but the asset class has also grown a lot since then and so has SoFi’s member base. Furthermore, not having this is an instant non-starter for the especially passionate crypto investors who require access through wherever they’re housing their accounts. I think this diminishes top-of-funnel friction more than anything for SoFi Invest, with some kind of small but purely incremental revenue benefit to enjoy as well. Estimates I’ve seen on the incremental impact range from about $20 million a year to $90 million or 0.6% to 2.8% of 2025 revenue.

SoFi also placed a nearly $700 million personal loan securitization deal this past week. This was the first new asset-backed securitization deal since 2021. As the 10-year yield continues to fall, rate cut expectations rise, quantitative tightening ends and unemployment remains at a strong 4.1%, capital market activity should remain robust. That is fantastic news for SoFi, considering more capital market demand means more ability to originate more credit without stressing its balance sheet too much. While the stock has been swept up in broad market turbulence, the macro backdrop remains improving for this company. As long as unemployment doesn’t explode higher from here it should be entirely fine. Its new annual guidance assumes unemployment rises to 5% this year, leaving plenty of margin of safety.

Starbucks named Cathy Smith as its new CFO. She’s currently the CFO of Nordstrom and was previously the CFO of Target, Express Scripts, GameStop and Walmart International. She will make $925K a year in cash with bonuses of up to $1.16M, a $5M signing bonus, a $6.4M equity grant and $4.5M in annual equity compensation. Starbucks also promoted its Japan CMO to CEO of that country’s business.

Per the Wall Street Journal, CEO Brian Niccol spoke with corporate employees following a large round of layoffs to demand more accountability and a “more effective decision-making process.”

Microsoft is testing non-OpenAI models.

Apple is delaying the next-generation of Siri. It’s “going to take them longer than they thought to deliver these features.”

CrowdStrike signed a Falcon channel distribution deal for Arrow Electronics (ARW) to sell the platform to their base of customers.

Zscaler got upgrades from Bank of America and Rosenblatt following its strong earnings.

Mercado Libre plans to grow Mexico investments by 38% Y/Y to $3.4B in 2025.

Shopify debuted AI store setup. This allows companies to enter conversational prompts to describe the look and feel they want for their site. Shopify then generates it for you, with the help of partners like Meta and Llama.

Uber and Waymo launched in Austin. And through a partnership with Hyundai, AVRide will launch in Dallas this year… with Uber. AVRide has already launched in Abu Dhabi with Uber.

Amazon’s AWS will invest $8B in Indian cloud infrastructure.

Output Data:

Inflation Data:

Consumer & Employment Data:

My macro post from earlier in the week lays out my thoughts on the rapid flow of headlines coming in and my views on macro. Nothing has changed since then. I continue to expect more short-term volatility, modestly higher unemployment and lots of tariff headlines in the coming weeks/months. Job growth is quickly shifting back to the private sector, but that will create some employment dislocation in the near-term and headwinds to consumer sentiment that need to be worked through. I continue to view sharp pullbacks associated with this noise as opportunity. I continue to think accumulating in small, frequent bites is the right decision for me.