For review:

Leadership:

What interesting timing for an Investor Day. As most of you know, the company had a surprise CEO change just 4 days before the event. CJ Desai, who had been with the company for less than a year, bolted to Meta for what I can only assume was a mountain of cash and equity. While this is not a great look, it’s hard for me to blame someone for getting a raise and what he views as a promotion elsewhere. I understand why it leaves a bad taste in the mouths of many, but I tend to side with the “go get paid” crowd. I do not think this is a red flag for MongoDB. I think Meta just rolled out the red carpet like they did for their entire superintelligence team. 

Former CEO and Chairman Dev Ittycheria came back to lead the company in the interim and there is a formal search underway for a new CEO. Leadership was asked how MongoDB plans to fill the C-suite connection gap that Desai’s departure creates. Some believe that his long and impressive resume fostered relationships with high-powered executives that allowed MongoDB to enjoy a bit more up-market success. And while that may be somewhat true, Ittycheria believes the success was  a “team effort” and a byproduct of ongoing work that will continue with or without Desai.

MongoDB 9.0:

MongoDB 9.0 is the latest iteration of its data platform. Compared to the previous version it boosts transaction throughput by 20%, reading speed by 35%, update pace by 30% and large instance throughput by 100%. Faster reads and writes mean customers' applications respond quicker and handle more users on the same hardware, which is both a cost story and a performance story. Faster core, better unit economics, happier customers. That's the idea just like it has been for several years and will be for several more.

I think the most exciting piece of this launch is Enhanced Queryable Encryption, which they see as one of one in the market. This frees customers to query their data while maintaining safe encryption. For example, a company can fetch a customer by the first letters of their name, or a card by its last four digits, with both the stored data and the search itself encrypted end-to-end. Just like MongoDB’s on-premise feature set caters best to highly regulated industries, so does this. Banks and hospital systems can enjoy modern, scalable and low-latency data handling without sacrificing security.

Atlas Infinite:

Atlas Infinite is now in limited preview for select customers. Until now, renting capacity from MongoDB entailed renting compute and storage in a static bundle. Atlas Infinite separates these two things, allowing each to scale independently. Why now? Changes in agent traffic are a lot less predictable than the traditional, seasonally-oriented human patterns. Agents can be dormant and then process more work in the blink of an eye than a person would in a day’s time. Sometimes, that requires a lot of compute and/or storage… sometimes none at all. This reality makes separate and flexible scaling up/down of these two things all the more imperative. With Atlas Infinite, customers pay only for what they use for each resource, rather than being forced to pay for a set amount for both. Customers use the same platform, security features and interface that they do under Atlas Core, which means no learning curve and very little friction to use this new option. The product is now in preview for AWS customers (and soon other clouds) with the full launch coming in Fiscal Year (FY) 2028 (currently in their FY Q3 2027). The meaningful revenue contribution is expected to come in FY 2029.

It must be noted that Snowflake and Databricks on the online analytical processing (OLAP) side have offered this to customers for years. They’ve also beaten this company to market within online transactional processing (OLTP), which is MDB’s bread and butter. MongoDB would argue that its OLTP offering is far more advanced than Snowflake and Databricks, and they’re right. Still, I would have liked to see them first to market here. This was missing from their product suite and it’s good to see them plug the gap. Late is far better than never.

Voyage AI:

Voyage AI provides mathematical representations of what text means, so a computer can compare ideas instead of just matching words. These are called “embedding models.” It also helps identify and select the very best results from a search output through “re-ranking models.” These models place near the top of Hugging Face rankings and enable model output quality that routinely surpasses world-class frontier models on their own. It’s these models that power its highly popular Vector Search (search by meaning/intent instead of keywords), which has delivered 2x customer growth over the last 6 months. This is perfectly complementary to Atlas’s agent work. Again, memory (not Micron chips but agents remembering previous interactions) is so important for AI value. That problem is all about feeding the right context to the right workload at the right time. That’s exactly what Voyage AI products augment and improve, making this such a natural fit. And the traction backs up that view.

Atlas Agent Engine:

This is MongoDB’s managed environment for creating, orchestrating, operating and perpetually optimizing AI agents. To me, this is one of the best ways in which MongoDB can morph from an AI data vendor to a full platform. They have a scaled document-oriented ecosystem that’s well suited for hefty unstructured data processing. They have high-quality embedding and re-ranking models to ensure information is optimally fed into agents and models to minimize hallucination rates. 

This is how they combine all that value with guardrails to create data agents that properly utilize this. It’s completing the value creation circle. The agents built through this engine will feature enterprise-grade memory that allows them to avoid functioning as “goldfish.” That eliminates constant redundant training once a new session is started. Customers get short-term and long-term memory and they fully own and control all of the data that flows through these systems. Aside from this, they view sound security and a long list of integrations as differentiators. MongoDB had something similar to this in the past called the MongoDB AI Applications Program (MAAP). The product struggled to find a market fit and they think that’s because the company tried to do too much themselves. They didn’t openly invite integrations with popular 3rd-party tools and that philosophy has now sharply changed. I think that positions the Atlas Agent Engine to be a lot more effective. 

Updated Financial Targets:

Over the next 3 years, MongoDB expects at least 20% revenue compounding with roughly 25% Atlas growth and 10%+ on-premise growth. Sell-side consensus calls for just under 20% compounding over that time and MongoDB has a consistent history of greatly under-promising on multi-year forecasts. I think there’s considerable upside potential for this target. During this period, MongoDB expects 1 to 2 points of annual EBIT margin leverage and an 80%-100% FCF conversion rate too. This is roughly in line with expectations, but again MongoDB loves to sandbag. And? When they beat, the vast majority of that revenue outperformance flows down to net income. Finally, they expect to maintain GAAP net income profitability during the planning period.

More Notes:

Opening Remarks:

Much of this event was spent working through the existing product suite, how it ties together and why it positions Zscaler to be an AI winner. There weren’t a ton of new product announcements, but instead the company telling a story to investors and trying to fight the narrative that they can’t dominate outside of their core product categories. They spent most of the day promoting the narrative that they’re a true platform and that AI will be the largest tailwind they’ve ever seen. The three main reasons are the same as they’ve been over the last several earnings reports:

AI Tailwinds:

AI is materially helping overall interest in Zscaler’s platform in two ways that we talk about frequently. The first is AI-enabled attacks that continue to become more complex, numerous and intimidating. Broken Record Alert: Models are powerful and dangerous at the same time. AI chatbots, MCP servers, and agents all enlarge the attack surface — and AI models can now discover vulnerabilities at scale, shrinking discovery-to-exploitation from months to minutes or seconds. More DDoS protection, more firewalls, more VPNs can't fix that. It's a different problem. Zscaler solves this problem by hiding applications from attackers rather than naively hoping they’re free of material vulnerabilities (which most are).

Second is the safe deployment and management of agents. These machine-based assets going rogue can do far more damage in a tiny fraction of the time it takes a savvy hacker. That heightened risk as well as the exponential boom in overall traffic tied to agent proliferation is expected to structurally augment overall demand levels for years to come. They call this the “giga wave” compared to the original and ongoing cloud computing “mega wave.”

Zscaler Internet Access (ZIA) and Zscaler Private Access (ZPA):

Helpful Definitions:

Zscaler debuted Autonomous AppShield as another useful tool under the ZPA umbrella. This conducts automated, frontier model-powered vulnerability scanning and patching. This work is done without customer application downtime and, through a new partnership with Red Hat (IBM), Zscaler expects to expedite remediation and find issues before they’re even public knowledge.

Zero Trust Branch:

Branch offices traditionally connect to the corporate network through specialized boxes (SD-WAN appliances) that create private tunnels. The problem with this? Once you're inside the tunnel, you're trusted and can access anything. In the last twelve months, Zscaler unified its Branch Connector and its microsegmentation technology into purpose-built appliances that replace that whole setup. They’ve positioned this as true Zero Trust alternative to SD-WAN for branch-level security. What does that mean in practice? Every device gets isolated. Nothing communicates with anything unless a policy explicitly allows it. A compromised thermostat can't become a doorway into the production line, because the thermostat was never allowed to talk to the production line in the first place.

AI Gateway and Securing AI:

The AI Gateway sits between every enterprise AI interaction…  every prompt, every agent action, every model call. Zscaler sees this product as differentiated in ways that are subjective and must be proven out over time via effective demand and revenue growth for the product. They see their MCP gateway as best-in-class, with an unmatched ability to understand agent intent and how to keep these unpredictable assets on track (have heard this from so many companies). This confidence stems from their large ZPA/ZIA foundation and the ocean of valuable outcome data that comes with it. They’ve owned a large portion of modern network security market share as an unfathomable amount of traffic was processed. This gives them a leg-up in terms of training AI systems to solve network-based agent problems. Zscaler is poised to handle the ongoing boom in autonomous traffic, with cost optimization and an overarching AI access graph that helps control and track AI assets while mapping everything they can touch.

Then there's AI-based endpoint security (acquired via the SquareX M&A). This controls AI usage by blocking impermissible data transfers between endpoints and protects against phishing attacks as well as other model-based risks. New attacks routinely target AI-based endpoints while other attacks can come from agents that are allowed to be where they are and have simply gone rogue. This makes now the right time to dip its toe into the endpoint world while it leaves core endpoint/extended detection and response (EDR/XDR) to CrowdStrike. In fairness, CrowdStrike is also pushing hard for the same AI endpoint security pie, pinning these two as more direct competitors than they’ve been in the past. But still, they remain friendly partners throughout the majority of their product suites.

Agentic Security Operations (SecOps):

This launched last month as the combination of Zscaler’s security operations center (SOC) and Red Canary’s Managed Detection and Response (MDR) and threat hunting products. As many companies from various parts of security push to be a customer’s centralized SOC, Zscaler thinks it has as good a chance as any. The pitch is data. Zscaler thinks it has better, more meaningful data than anyone thanks to its scaled inline exchange. I would argue that Palo Alto and maybe Cloudflare have better access to scaled data due to their size and platform breadth but I do think Zscaler is the 3rd-best positioned in this regard. And, they parlay that with product quality that is as good as any on the market.

Now, why does the SOC position matter so much and why is every security company trying to become it? The SOC is where a company's security team (agents and people) actually works. It's the screens, the alerts, the incident response and the operational hub of cybersecurity. Whoever owns the SOC owns the daily workflow of the security team, and that makes it the stickiest position in the industry. Once analysts live in your console, ripping you out means retraining the team and rebuilding every playbook. Nobody takes that lightly.

It also greatly accelerates and deepens quality of data flowing into the platform, fueling an ongoing flywheel of product improvement driving more business, which, in turn, keeps the whole process humming.

That's why they're all coming for it. CrowdStrike, Palo Alto, SentinelOne, Zscaler… every one of them wants to be the console the analyst opens in the morning. The endpoint players like SentinelOne and CrowdStrike will tell you they’re best positioned to win the opportunity. The network players like Zscaler say the same thing. And the end-to-end platforms like Palo Alto say their unmatched product breadth provides better data telemetry and SOC positioning too. All of them talk a big game and that big game will be (or won’t be) supported by accelerations in company growth.

Financials:

Zscaler guided to $8B to $10B in ARR by 2031, representing a 17%-22% CAGR during that time. M&A is expected to add 1-2 points to annual growth. Offering this amount of revenue guidance was seen as a big surprise, while the outlook delivered slight upward revisions to consensus revenue estimates over the next few years. While the change wasn’t at all large, it was still important as it reversed a slightly negative trend that lasted for most of 2026.

For profits, Zscaler expects to maintain a roughly 80% GPM while G&A expense leverage (thanks mainly to AI) should allow gradual operating leverage throughout the planning period. All in all, that makes low-to-mid 20% operating income and free cash flow compounding a solid bet for the company. That puts it at a PEG ratio of 2X or modestly below that. The FCF growth multiple is closer to 1.75X.

Quick Take:

Really not much new here. I feel like the 4-hour event could have just been them giving us the Autonomous AppShield product launch and the 2031 ARR guide. At the same time, that ARR guide is immensely important and highly encouraging to me. They always sandbag, which to me makes 20%+ top-line compounding for the next 5 years (18%+ organic) highly achievable.

I continue to watch this company very closely and am more interested in an entry today than I was with the stock lower a few months ago. At the same time, I do want to see at least one more very solid quarter of execution before entertaining the thought of jumping back in. We are still just 6 months removed from what I viewed as a highly disappointing quarter and the valuation doesn’t create much sense of urgency for me right now.

I want to preface this by saying the stock is already way off of its 52-week highs and there is a lot of bad news already being priced in. For important context, it’s the cheapest high-growth software name using either PEG ratios or FCF growth multiples. With its expected growth profile, that would normally make me eager to jump in with large purchases and incremental accumulation if it got cheaper. But I’m not doing that in this case at this point in time. I am open to doing so in the future, but not yet. There are three reasons for this:

First, I fear that there’s more to the model update delay cited last quarter than a timing issue. If it was only timing, then the revenue guide should have been stronger like it usually is. It’s fair to say I have a bit of PTSD from admittedly holding onto The Trade Desk for too long. That company had a charismatic CEO making big promises and explaining subtle deterioration in fundamental performance via highly understandable things that turned out not to be true. Perhaps I’m being unfair to APP and overly patient by wanting to wait for a couple clean quarters before I entertain starting a position, but I’d much rather forgo some profits than risk significant losses. 

Second, I think there are real competitive concerns from Unity’s emerging rival offering. With APP’s best-in-class 80%+ EBITDA margin, I think there’s a chance Unity could drive some pricing pressure to steal market share. APP’s request for a temporary restraining order (TRO) against Unity was denied by a San Francisco Superior Court judge. APP accused Unity of essentially harvesting APP’s ad data and feeding it into its own ad platform to gain an unfair competitive edge, which Unity denies. Regardless of which company is right, Unity is not barred (for now) from operating the software development kit (SDK) that’s supposedly doing this while the dispute moves to arbitration. Unity also has highly valuable runtime gaming data to season its own algorithm with incremental signals, making it a formidable emerging competitive threat.

Third, I had a conversation with a buy-side friend of mine who I have a great amount of respect for and who has access to a humungous data budget. For APP’s e-commerce growth engine, the rapid storefront growth we've seen some publications cite is noisy. According to their data, they think that growth is greatly over-indexing towards fake sites out of Hong Kong that aren’t reliable leading indicators for revenue. They think there’s a good chance the e-commerce business eventually does inflect in a meaningful way, but they’re not going to turn bullish until that happens (and it hasn’t yet). They believe (and I fully agree) that upward estimate revisions stemming from the core gaming business have likely run their course and the next leg of that must come from e-commerce. Taken together, this is the current buy-side narrative and why price action has been so awful lately. None of this stuff entails new revelations that aren’t already understood by the funds that control most investment dollars. This is me explaining why the sentiment has gotten so aggressively bad for a name with fundamentals that still look so wonderfully good.

Finally, I have no interest in owning the name today. They need to ease my anxiety levels in all three of the areas I just mentioned. I am not confident enough in this happening to buy shares ahead of the potential successes. If they deliver 2 consecutive clean quarters, that would tell me my fears are overblown, the forward estimates are reasonably reliable and this stock comes with great value. I would seriously consider owning it (even at higher prices) at that point. Not until then.

As commerce patterns rapidly evolve and agents begin automating product discovery, it’s natural to ask how this impacts the leader in U.S. e-commerce. Is this good for their market share or detrimental to it? I think it’s a net positive. Shopping Agents optimize for the best possible outcome. The three things that determine best outcome are product relevance, price and delivery speed. We’ve seen highly (highly) capable disruptors like Shopify attempt to match it on the delivery side and fail pretty emphatically, showing how deep this competitive moat truly is.

It doesn’t matter if agents are doing the shopping or people are, they’re still graded on human satisfaction and that satisfaction is still tied to the same three things it always has been. Considering this, I think Amazon thrives in this era. I think consumer traffic patterns keep rapidly evolving and some small businesses with especially great products and abnormally low prices will suddenly gain more traffic. But? Amazon will still over-index in agent outputs because they’re still usually the leader in the variables that create consumer value for each transaction.

I hold this opinion despite Amazon currently blocking agents on its platform. While everyone yells about it being some grand and permanent move, it’s not. The CEO already told us that integrations will likely come eventually.

For now... Amazon is building its own agents before openly accepting others. It’s savvily figuring out how to set up 3rd-party integrations with as much value as possible.  They have dominant e-commerce market share and a large ad business to protect. They must nail agent-based monetization & structure before opening the floodgates to make sure the edges they’ve worked so tirelessly to establish are maintained… and so they will. Once they get this right & the floodgates do open? I think this will prove to be a large enough commerce GMV accelerator to more than offset any headwind to their sponsored listings business. And lets not forget how quickly Amazon Prime Video ads are growing, which aren’t impacted by this at all. Amazon will be just fine.

Rubrik introduced Code Guardian as an extension of Project Hourglass (AI partner alliance) and a complement to Rubrik Agent Cloud (RAC). Code Guardian is an AI red-teaming tool powered by Anthropic's Claude that takes a cloned, air-gapped copy of a customer's code repositories and attacks it with sophisticated multi-step threat scenarios. This way, partners can find the weaknesses in client code without ever touching production. Rubrik keeps pushing beyond backup-and-recovery into proactive AI security, and it's doing it through the partner channel, which is how this stuff scales without hiring an army.