Next week will include more coverage from all of the investor conferences this month. I am getting married on Saturday, so I am going to get everything done by Thursday. I will likely be slower to answer your questions Thursday-Sundar and I thank you in advance for your consideration. Happy weekend.

Security and Operations Merging (Even More):

Security is perpetual monitoring of autonomous assets that already reside within a company’s digital estate. They are not adversaries breaking in and wreaking havoc. They are entirely permissible assets that already have broad access. Aside from actual adversaries using agents to automate highly complex and frequent attacks, this is the other single most important piece of sound AI security. When permitted agents go awry for any reason at all, they suddenly turn from profoundly useful efficiency boosters to expeditious destroyers of company operations. Just ask those impacted by Hugging Face how damaging a rogue agent can be and how quickly things can unravel. It’s a new world in which our trusted tools can unpredictably turn on us in a heartbeat. That means security is becoming an operations and agentic monitoring problem. Humans will be the controllers and decision makers in this world, with machines doing most of the work.

Rubrik is positioning itself as the company to provide real time protection against rogue AI agents.  They do this by offering agent visibility (including shadow AI), actionable agent access enforcement, and runtime agent security that understands intent. This is all covered by their Semantic AI Governance Engine (SAGE). And finally, they offer easy rollback of only affected areas to restore operations quickly when the inevitable breach happens. They think other companies can match pieces of these 4 pillars, but not the entire suite of services. To be fair, giants like NOW & CRWD do have the first three things, but they lean on partners for recovery. CrowdStrike already relies on Rubrik for identity rollback, and neither platform has a native equivalent for rewinding agent actions. Natively offering all four parts of agent security in one place is a perceived edge they plan to  exploit as they continue rapidly compounding.

Non-Human Identity Monetization:

It’s still too early to know exactly how this will shape up. The things that matter right now are whether traction for these products is building and whether or not customers are getting what they need.. Both of those things look good, with rapid new user expansion and net revenue retention both positive (covered in the recent earnings review). As long as that continues, monetization should seamlessly follow. And while they might not fetch nearly the same rate, when the explosion in overall volume is as massive as it will be, that won’t matter.

Outcompeting the Pack:

Win rates against incumbents remain “extremely high” as they have for a long time. Bipul was also clearly confident in RBRK’s ability to keep taking more market share against new competition in other categories in the quarters to come. There’s plenty more market share taking to do across all of their product categories and clear signs that their pursuit will be fruitful.

I think Rubrik’s innovation culture is a big reason for this consistent success. For example, Rubrik X is its internal incubation team that has been responsible for building their new growth areas. Bipul spoke about their awareness of the cliché issue of enterprise software firms killing it in one area, struggling to expand beyond the core and seeing their growth curves abruptly mature as a result. They’re determined to avoid this fate and are already showing great signs in identity and AI of being able to pull it off. Rubrik X is their way to get more shots on goal and ensure they have talented people always focused on future product utility.

Why Some Tools Are Taking Off First:

When SaaS first proliferated, companies bought software for non-core work not so closely tied to their proprietary data. That didn’t require an intimate understanding of data and asset relationships to begin driving value, so adoption was faster than more industry or company-specific solutions that took off later.

That same process is playing out in AI. Companies must understand an even more complex web of inter-relationships before they can plug this new technology into their most valuable and sensitive resources with safe scalability. They need to know these hefty costs will come with attractive returns and won’t lead to catastrophic security issues that jeopardize the enterprise. That’s why custom AI solutions built for an organization’s proprietary workflows have been slower to grow, but Rubrik thinks that will come. And? As they prove to be a capable vendor on the generalist agent side, they think that the next wave of enterprise AI adoption will yield immense demand for their suite.

Cloud Migration Inning:

When Datadog went public, the early innings of the cloud migration were poised to carry its structural growth engine for many, many years. When asked “where are we today” the answer is still squarely in the middle. They still have just half the Fortune 500 as customers and see an easy 5x-10x upsell with the big logos they’ve already won.

As AI explodes in popularity, it’s easy to forget that this core business still has so much room to run. And? That’s especially true as AI expands asset and attack surfaces, making its bread-and-butter observability suite all the more imperative (Datadog for AI). With this in mind, Datadog has made recent decisions like decoupling its Bits Security Analyst from its Cloud Security Information and Event Management (SIEM) offering so it can work with third-party SIEMs like Splunk and Microsoft Sentinel.

Not only is their non-AI business sustainably compounding, but it’s doing so at an accelerated clip because of AI. They couldn’t be growing 36% Y/Y clip if their largest business wasn’t also enjoying much stronger momentum alongside new products.

Team Consolidation:

In the Rubrik piece, we wrote about how the walls between the security and operations teams were beginning to come down. Overlapping work has gotten far more routine, and collaboration has become imperative. Pomel echoed that sentiment in this interview. While cloud computing demanded faster work and more efficient communication – which feeds the need for team consolidation – AI does that on steroids.

This deeply favors platform-level companies with high-quality offerings in many, many different areas. Because? Those areas are becoming one, meaning solutions now include many different categories that point products cannot address. Yet another reason why AI favors scaled incumbents with better product breadth.

AI Pricing & Usage:

Pomel said everyone is still in the mode of figuring this out. Usage-based has turned into a popular way to monetize all of this innovation. The vast boosts in productivity continue to create so much more volume that we probably won’t notice lower pricing power for machine and AI-based assets vs. human-based. For their customers, they see 10%-20% of AI budgets going to Datadog at maturity just like in the cloud space. This is supported by strikingly similar product adoption patterns (on a faster usage ramp) for AI natives vs. their other types of customers. And that makes sense. These AI tools are using the same data and tools that people are. They’re just doing it in ways retrofitted for their makeup and with far more automation and possible scalability.

Still, there has been a tiny bit of GPM pressure. It has been extremely modest and more than offset by OpEx leverage, so this isn’t hard for me to accept in exchange for playing a more meaningful role in the AI buildout.

As briefly discussed in the recent earnings review, there’s a lot of concern surrounding quality of usage from AI natives. In 2021, pandemic darlings voraciously consumed DDOG’s services and other enterprise software titans with little regard for efficiency, cost or anything but “grow, grow, grow.” That leads many to worry a period of usage rationalization will materially slow growth like it did in 2022-2023. They’ve learned from their mistakes and gotten a lot more proactive in pushing customers to better consumption practices. They’re not allowing clients to behave in ill-advised ways, which means there won’t be mistakes needing fixing thereafter. 

Usage is shifting from training and frontier model API calls. It's moving to inference, or companies actually using these polished assets to unleash automated agents and apps in which AI finally creates real enterprise value. That next wave of usage should support another leg of AI app layer monetization.  That leg is still in its infancy which is great news for this company. All of those emerging products and workflows will need perpetual observing, optimizing and securing.

Finally, the eventual vision for Datadog’s AI work includes building a fully automated site reliability engineer (SRE) that turns observability of AI into an entirely machine-based endeavor, with humans essentially acting as agent managers.

Bring Your Own Cloud (BYOC):

As a reminder, its Bring Your Own Cloud (BYOC) offering allows users to keep their log data in their own cloud environment while still using the DDOG platform for querying, analytics and investigations. The natural question is "why would you let product value flow to data stored outside your ecosystem?" In reality, open means more overall usage of these products. While revenue per workload is likely higher if data is ingested and stored within DDOG's own platform, closing things off would be a revenue and friction headwind. Customers facing petabyte-scale log volumes and data residency requirements were increasingly choosing between absorbing storage costs, dropping data or turning to BYOC-native competitors. Allowing this form of a la carte usage keeps DDOG as a key vendor for those workloads, even when the underlying data lives elsewhere.

Where the Business Has Improved Most:

Datadog’s execution and performance have gotten noticeably better over recent quarters. While AI adoption clearly is a rising tide that lifts all boats, Datadog’s boat is rising even faster than most of its peers. They also think their platform creates a unique dataset that bolsters their existing offerings to a point of attracting customers like OpenAI while many assumed that AI native could just vibe code an alternative. The combination of great existing tools and an AI embracing culture makes everything DDOG does better. Customers enjoy better outcomes and lower costs with this platform than they do on their own, and so Datadog continues to thrive. It’s as simple as that… They win because they’re better and cheaper. That’s why the rare customer that does churn so routinely comes back to DDOG thereafter. 

“It's pretty clear to me that observability is a major part of any transformation story and the AI story in particular. It's also pretty clear that observability is the last frontier. That's what remains, like keeping tabs on AI, keeping tabs on the machines, keeping tabs on the agents, whatever the job is, is going to be absolutely key in the long run. So, that's a gigantic opportunity.” – CEO Olivier Pomel

One more note here. The race to build the biggest AI infrastructure footprint and the massive associated costs have forced the large hyperscalers to prioritize what they choose to do themselves. AI Observability has turned out to be a common element they choose away from, as again, Datadog is an extremely capable partner with the ability to operate at immense, proven scale. These big customers and big AI natives will probably continue to do some workload observability and product development on their own.  That’s fine, a I think it’s highly likely that they will still rely on DDOG as a big part of their overall observability needs.

The company is also using this scaled context to debut multiple AI models that were considered state-of-the-art for their size upon release. They think they’ll continue shipping innovation that delivers better cost and performance than using frontier models. Datadog will continue shifting away from usage of those expensive models as quickly as they can.

Brand Maturation & Product Mix Shift:

There’s a growing theme in Lemonade leadership remarks to analysts and investors. Their brand in the eyes of consumers is evolving from a renters insurance company to an insurance company. As discussed last week, they envision home and auto eventually becoming much larger portions of the book, as they add more states and scale the newer segments. This is important for the overall Lemonade customer vision. Their idea is to win young customers early with a cheap renters product and best-in-class service that includes 3-5 minute policy onboarding and 50%+ of claims settling in real time. Once these customers age beyond renters, Lemonade’s great service gives them a good chance of winning higher value policies. This is what makes renters such a strategically important gateway. It’s “really difficult” for large insurance incumbents to match, according to Bixby.

But following that graduation, it doesn’t matter how great the renters experience was if the company doesn’t have the other products a customer needs in that state. That’s why Lemonade must expand  all 5 products in all 50 states. It’s almost there for 3 out of 5. Home and auto will take a bit longer. As this happens, things like Lemonade’s 5% customer cross-sell rate more closely matching 30% industry averages and its annual dollar retention climbing into the 90%+ range like large best-in-class competitors should follow. These trends should also provide great sources of operating leverage, as selling more plans to existing customers is much more profitable than winning new ones.

AI Regulator Ally?

We often think of Lemonade as the newer kid on the insurance block. And that’s accurate. On the other hand, they’re a mature incumbent when it comes to using AI throughout every system they offer. That has made them an ally for regulators as they make sense of what this new technology means for the space and how to best regulate it. Nice to see this company perhaps earning a better seat at the table with rule makers.

Why So Confident in 30%+ Compounding?

As a reminder, Lemonade’s growth has been both formulaic and predictable. Their lifetime value to customer acquisition cost (LTV/CAC) has been maintained even with explosive growth, as experience and AI-powered marketing improvements have helped the pace of targeting improvement overcome diminishing returns that so routinely come with scale. It would almost be boring how predictable their revenue engine is if it weren’t so compelling from a fundamental excellence point of view. They’re confident they can grow at 30% because they operate in a massive industry with a best-in-class product with a tiny, tiny book of business compared to giant competitors. Market share gains can easily offset the natural cyclicality tied to the business for a very long time. Lemonade could grow way faster if they didn’t care about margin improvement and could be profitable today if they were ok with slower top-line compounding. To them, 30% strikes the perfect balance. 

“We've had 24 quarters in a row of pretty consistent results relative to our guidance and market expectations. That's a good thing. I would expect that sort of resilience, visibility and predictability to continue.” – CFO Tim Bixby

On the General Idea of Following Competitors:

Bixby was asked about matching competitors as they pull back from high CAT-risk states. He turned the conversation to something more general and interesting. Lemonade doesn’t really care about how its competitors are behaving. It will not make any big business decisions based on what those other guys are doing. That’s because they believe their modern, lightweight, obsessively data-driven foundation (per Bixby) makes them better at learning from experience than bigger incumbents.

“We believe the momentum is toward a world of more data and AI enablement providing a better product and customer experience. We believe that we're among the best at doing that and getting better every day, even though we're not yet at scale. One of the things you'll hear Daniel or Shai frequently say is… I wouldn't trade our data, our system, our capabilities for any other on the planet.” – CFO Tim Bixby

Edge:

We often talk about SoFi enjoying special cost advantages in banking that help it offer customers more value differentiation. Lemonade does arguably a better job of that in insurance. It built its entire tech stack Just like for SoFi in banking, Lemonade offers a product that has close substitutes. It differentiates as well as any other company on user experience, marketing precision and strong underwriting. But? The best way it can easily win customers from everyone else is on cost. And it’s edges like these that allow the undercutting and differentiating to happen while the company marches towards positive EBITDA next quarter.

Why Cloudflare Doesn't Need Hyperscaler CapEx:

While hyperscalers build massive, GPU-based footprints to support gigantic training clusters and inference, Cloudflare gets to enjoy a rapid AI growth acceleration without having to spend all that money. Nice to be Cloudflare. They're fantastic at inference optimization and perfecting the work organization and delegation that CPU-based compute must perform. They don't need to build these massive data centers. They do have a footprint and do need some GPUs, but these requirements are much more modest and they're not finding a need to rapidly build new facilities. Why? Partially because they built these data centers years ago with the ability to easily install GPUs (great CEO). And? Partially because they don't need to do what hyperscalers do. They just need to help customers use their compute in far more efficient and fruitful ways... and facilitate reliable network traffic at scale and best-in-class performance. Those are two very hard things to do, but they're also things that don't require nearly as much CapEx for the winners. There's no company better at executing here than Cloudflare. They are immensely sticky without having to spend crazy piles of money and build humungous compute factories ahead of demand. Cloudflare is perfectly positioned to capitalize on the explosion of agent-based traffic as the orchestrator, scheduler and refiner of that traffic. It's this reality that supports 20%+ growth for as long as any other peer at scale... hence the lofty multiple.

On-Premise and Data Sovereignty Trends:

The trend toward more on-premise architecture for big customers to defend against data leakage are good for Cloudflare's business. This moves away from gigantic, centralized data center models to a model that features more fragmented, remote footprints that are perfect for NET's global presence. The increasingly prevalent data control motivation is also entirely fine for Cloudflare's business. That's how they were built anyway. You all know I love Amazon, Google and Microsoft... but those companies do have a lot of other uses for this lucrative cloud traffic data across the rest of their businesses. Cloudflare is in the business of making that traffic work better than anyone else can. It doesn't have anything to cross-sell that may create conflict between securing customer data and wanting to use that data elsewhere.

AI Product Value:

As a reminder, their "Pay Per Crawl" product is a tool that lets websites charge AI crawlers for accessing their content. Cloudflare uses its massive scale to gain a better sense of changes in web data in real-time. This means it doesn't need to redundantly pull the same identical pages over and over again. It can just take whatever has changed, greatly limiting costs and waste.

In a world where chatbots and agents are voraciously scraping information from around the digital ether to complete tasks, the owners of that information have a heavy motivation to guard against impermissible taking. We've seen how this can get ugly with the ongoing legal battle between Reddit and Anthropic pertaining to that AI giant stealing Reddit information for model training without payment. Cloudflare helps customers avoid this by using its 20% share of internet traffic to create standardized guardrails. It determines what AI tools can use and facilitates payment for this highly valuable asset (rather than effectively stealing it). The actual payments happen through its monetization gateway that settles payments in stablecoins via the x402 protocol, whose backers include Visa, Mastercard, Amex, Google, Shopify and Stripe. Their "Wallets" payment architecture blends perfectly with these scaled authorization systems by enabling customers to set up agent-based payments with strict, reliably enforced rules.

All of this is very exciting, but Cloudflare CFO Thomas Seifert cautioned that the priority right now is scale, not revenue. They think broader adoption will lay the seeds for more powerful long-term revenue compounding if they stay patient for now. Given they said similar things about Workers Platform (and many things before that) and it's now materially contributing to their financials in a big way, I think trust is warranted.

Cloudflare thinks its moat and product leads will grow in the age of AI as change accelerates and it adapts and innovates faster than others can.

Modeling Challenges:

Cloudflare has found modeling more difficult in the AI world. Like many other software platforms, it's shifting more of its business from seat-based to consumption-based. It's a lot harder to know how much more traffic and consumption a customer will deliver in this new world than it was to know how many people they'd hire a couple years ago. This is leading them to be more prudent in their already impressive forward-looking guidance and (I think) could yield bigger upside vs. consensus than we're used to for the next few quarters. Great teams like this tend to lean conservative amid decreasing visibility.

Margins:

GPU-involved workloads are lower GPM than traditional CPU-only counterparts. That's because NET has been optimizing CPU efficiency and performance for a lot longer, so it enjoys higher utilization and value per CPU-based compute than GPU-based. They're closing the gap and also enjoy relatively lower OpEx intensity in some parts of the GPU side as well. It's very easy to trade a modest short-term gross margin headwind for a lot more business when the EBIT margin headwind is so small and the source of profit pressure is ephemeral.

Recently, an Anthropic employee went viral for saying he thinks there’s a 10% chance of the company creating a model that drove human extinction. This felt like a planted story for the media, but I have no evidence of that being the case. Just a guess. Regardless, this led to both Anthropic and OpenAI calling for slowing down model development in the name of safety and privacy. I don’t buy that. I don’t think either company cares much about those things. If they really did, they wouldn’t need everyone else to slow down in order for them (the leader) to make the same decision. I do, however, think both Anthropic and OpenAI care a lot about competition and beating it. They see closed-source models vastly undercutting them on token costs. They also see cheap open source models suddenly getting very good too. Convincing regulators that there’s an issue as pressing as human extinction to deal with could lead them to tighten the grip on those less controllable open source alternatives. I think that’s the true motive for all of this noise.

Relatedly, Meta’s Mark Zuckerberg this week took to social media to effectively say something along the lines of: “We already paced the rollout of new products. We don't need to slow down. We already have customer interests in mind. META doesn’t need the world to slow down in order for us to make a responsible decision like that. We’ll just do it on our own.”

A couple things on this. First, Meta is deeply profitable and has a lot more of its own cash to spend on compute than the Sam and Dario calling for slowing down. Second, Meta also is a lot more ok with open source proliferation than those two.

In terms of whether or not we should be dampening pace of progress, here’s how I think about it. If these mass extinction threats from Anthropic/OpenAI are even remotely real, then yes it seems like we should be slowing down. Civilization matters more than making models a little better more quickly. That’s pretty easy to agree with. It’s just that our main adversaries won’t slow down, so the responsible decision might be the one that causes the USA to lose this important innovation race. That’s what makes this hard. And to complicate things more? We don’t even know if this is a responsible decision, or? One that’s in response to ridiculous, baseless claims inspired by a quirky Silicon Valley founder who does plenty of BSing.

In the end, I think some level regulation is pretty inevitable just as it is for any other new industry over time. I do think there will be new rules. I don’t know exactly what they’ll be or when they’ll come, but I’d imagine they’re not nearly as tight as the most strict forecasts and not as loose as they are today. Reality tends to always be somewhere firmly in the middle.

The Fed hiked rates a quarter point. Warsh (very briefly) spoke about resilient and improving economic growth and employment data. He doesn’t say much and he doesn’t offer any help on forward-looking expectations. I get that some find his lack of willingness to answer any questions amusing. I find it pretty annoying and think he could do a lot better job with messaging. That’s an important part of his job and I’d give him a failing grade in that area early on. Need to do way better, in my opinion. He also spoke about inflation not returning to the 2% target quickly enough, which led to the decision to raise rates. While rate hikes will do nothing to help normalize oil supply, Warsh and the Fed believe this will help lower the risk of inflation spreading beyond energy. At the end of the day, making money more expensive to borrower won’t do anything to help create more oil availability in the slightest. For this reason, I don’t think this week’s hike will do anything. But? It doesn’t really matter what I think. What matters is that they hiked and expect to hike one more time this year. Last week, I wrote a piece on my plans amid potential rate hikes and those plans haven’t changed following this week’s news.