AI Will Not Equally Reward Software Platforms

Control, Monetization, and Profit Pool Shift Comparison for DDOG / NOW / CRM / SNOW

Analysis Date: 2026-04-26

Chapter 1: Why the Old Map is No Longer Sufficient

These four companies are most easily placed within the same narrative recently. They all talk about agents, have large customers, complex product lines, and are trying to move AI from the demo layer to the truly monetizable product layer. So the market naturally puts them into a convenient sentence: large-cap software platforms, significant beneficiaries in the age of AI. The problem is, this sentence is too convenient and too crude. It assumes these four companies compete on the same map, will benefit in similar ways, and largely deserve similar valuation language.

However, once the focus shifts from "who has more AI features" to "at which layer AI truly monetizes," the picture changes completely. ServiceNow's Now Assist ACV has reached $600 million, and management has raised the FY2026 target to $900 million to $1 billion, indicating that customers are already paying for a certain capability where "AI truly enters the workflow." Snowflake's Cortex AI currently has a run-rate of approximately $100 million, accounting for only about 2.2% of total revenue. This doesn't mean the product is valueless, but rather that it is far from a "new monetization layer significant enough to rewrite the valuation language." Salesforce's headlines are hotter: Agentforce's headline ARR is $800 million, but 67% of these deals are still free. This indicates that adoption is gaining momentum, but payment quality has not yet stabilized. Datadog's problem is different: Bits AI and LLM observability bring it closer to "moving from seeing problems to participating in problem resolution." However, AI revenue has not yet truly separated from the traditional product portfolio, yet the market has already begun valuing it based on a more distant future.

If we only look at valuations, the cracks become more apparent. NOW can currently maintain approximately 27x P/E and 13x EV/Sales; DDOG superficially trades at 49x Non-GAAP P/E, but if we switch to a more conservative owner economics metric, the picture immediately tightens; CRM appears to have only 14.7x P/E, but "low P/E" does not equate to cheap, as it averages a slowing legacy core with new optionality that is still being priced; SNOW is still around 11x EV/Sales, but whether this platform premium is for data workloads or for a control point that is moving externally is, in fact, an open question today. Superficially they seem like the same sector, but in reality, the market is already pricing them with four different sentiments, it just hasn't fully articulated those sentiments.

If we must distill them into one sentence first, their positions closer to reality today are roughly as follows:

The most important aspect of this table is not the ranking, but how it shatters the old map. NOW's most valuable aspect is not "having AI features," but rather its ability to allow status to truly change within a system that is write-back, traceable, and auditable; DDOG's most valuable aspect is not beautiful charts, but rather its proximity to real system behavior, enabling it to compress fault discovery, diagnosis, and remediation into a shorter timeframe; CRM's problem is not whether it has Agentforce, but rather that its Service, Sales, and Platform business lines have completely different capacities to absorb AI; SNOW's problem is not whether it has Cortex, but rather whether it can continue to command its historical platform premium once data formats, catalog layers, and platform layers become open.

If I had to give a clear ranking upfront here, I would write it as: NOW > DDOG > CRM > SNOW. This is not a comparison of whose launch event is louder. It is a comparison of who is already closer to the monetization layer, the write-back layer, and the thicker part of value retention, versus who is still sitting in a layer that is easier to abstract away or must be split apart.

This is not an abstract classification exercise. It will directly determine what valuation language should be used for these four companies, what evidence should be awaited, and at which layer their thesis should be rewritten. The truly difficult part begins here: it's not about giving these four companies a flashy AI ranking, but about discerning who has begun to pass through the gates of monetization, attribution, responsibility, and value retention; who is still caught up in hype and demonstrations; who must be viewed separately; and who should first have a portion of their platform premium clawed back. The following section is where the true value of these names lies.

Chapter 2: The Same Sector, Four Types of Assets

To truly dissect these four companies, the most effective way is not by industry labels, but by where they stand in the enterprise software value chain.

At the very bottom is the data and workload bearing layer. Its importance has never been low, but its value comes more from storage, querying, distribution, and compute utilization, rather than direct control over final business outcomes. One layer above that is the observation and feedback layer, where the core value is not in "owning the outcome," but in "seeing what is happening in the system faster than others." Further up is the workflow control layer: tickets, incidents, changes, approvals, cases, permissions—these objects are not static data, but workflows where state machines continuously flow. The most difficult and valuable layer is the responsibility and budget absorption layer. Customers pay you not because you do things beautifully, but because you can take on a larger portion of the budget and step in to bear the consequences when things go wrong.

These four companies happen to stand at four different positions. Snowflake primarily remains at the data and workload layer; Datadog has clearly moved to the observation and feedback layer, and is striving to extend into the processing layer; ServiceNow is closest to the workflow control layer, even touching the edge of the responsibility and budget absorption layer; Salesforce, on the other hand, is not a single object; it has parts that resemble traditional core cash flow and parts that resemble agent optionality. Therefore, this group of names does not have a single "one-size-fits-all" analytical starting point. From an object positioning perspective, NOW and DDOG are closer to the upper-level control points, CRM needs to be dissected to be clearly understood, while SNOW remains at a more easily abstractable level.

Chapter 3: NOW: The Company Closest to the Closed-Loop Execution Layer, But Its Upside Is Not Infinite

If we only look at which of these four companies is closest to the true AI second curve, I would still place NOW first. The reason is not that it makes the most noise, but that it already stands where enterprise workflows genuinely undergo state changes. A ticket from open to resolved, an incident from outbreak to closure, a change from pending approval to execution, a CMDB record from incomplete to updated—these are not "recommendations" or "inspirations," but real, auditable, rollbackable, and billable actions. As long as AI can continuously write back into this system, ServiceNow will no longer just be selling advice, but an increasingly automated state machine.

Now Assist ACV has reached $600 million, and the FY2026 target has been pushed to $900 million to $1 billion; Pro Plus can also bring a 25% to 40% price premium. Looking solely at these two points, it has already pulled away from the other three companies. It indicates that customers are not just trying out a "seemingly smart interface," but are separately paying for "allowing AI to enter real workflows." Precisely for this reason, while NOW's valuation is not cheap, it is not without justification. The market gives it 27x P/E, 13x EV/Sales, essentially paying for an asset closer to the workflow control layer, rather than for a regular ITSM leader.

More critically, NOW's control points are not isolated. CMDB, historical tickets, permission configurations, approval rules, cross-SaaS integration, AI Agent Studio, Workflow Data Fabric—these elements are not scattered haphazardly, but form a sufficiently dense context. This makes AI's value not just derived from model inference, but from the model's ability to genuinely execute actions within an environment with permissions, auditing, and state recording. This is also why NOW's AI value more easily translates into budget: customers are not buying "a chatbot that talks better than Copilot," but "an execution framework that combines tickets, changes, upgrades, updates, and auditing." Such a thing is difficult to replace with a single point solution. Even if Microsoft Copilot or Salesforce Agentforce are cheaper and faster in certain single-point scenarios, what they would need to replicate is not just a function, but an entire state machine.

However, this does not mean NOW has no boundaries. One of the most important things to consider is that management actually understands its boundaries better than the market. Pat Casey publicly rejected outcome pricing, which I have always considered to be the most critical double-edged sword in NOW's story. On one side, it's protection: it prevents the company from prematurely taking on asymmetrical risks of low margins and high responsibility in pursuit of a larger AI narrative; on the other side, it's also a ceiling: as long as it refuses true outcome pricing, it has not yet evolved into the type of operator that takes on outcome budgets for customers. Thus, NOW has formed a rare and important shape: it is the most like a closed-loop execution platform among the four companies, but at the same time, it maintains restraint regarding its own boundaries of responsibility. In other words, it is the strongest, but its upside is not infinite.

Therefore, NOW's question is not "whether it can succeed," but how many years the market has already prepaid for it. The variables most worth monitoring are also very specific: extrapolating from public disclosures, the current Pro Plus attach rate still appears to be in the 10%-15% range—can it continue to approach 30%? Can the commercialization of Now Assist / Pro Plus push ACV to $900 million to $1 billion? Will the inference costs brought by AI continue to depress gross margins, especially after GM has already seen a -1.7 percentage point pressure? Can the $7.75 billion large acquisition of Armis be smoothly integrated? And will Microsoft Copilot for IT and Salesforce Agentforce IT truly enter the main battlefield in 2027-2028? If any two or three of these points deviate together, while NOW will still be a good company, the valuation narrative will need to narrow.

Chapter 4: DDOG: Extremely Strong Product Value, But Commercialization Evidence Has Not Kept Pace with Valuation

Datadog is the second strongest business in this group of names, but the stock is more challenging than the business. Its true strength is not merely having a new product called Bits AI, but that it stands at the very foundation, next to real system behavior. Trace, metric, log, RUM, security, LLM observability—these are not outcome variables on financial statements, but the production environment itself. When a company's value is built upon the ability to "see systems going wrong faster than anyone else," it will be very close to true automation. Because the first step of automation is not action, but perception. The more genuine and timely the perception, the higher the likelihood of AI moving from suggestion to processing.

This is where DDOG's product layer is most valuable. Bits AI, Watchdog, LLM observability, and even more peripheral AI capabilities like Casey and Toto, are essentially all doing the same thing: compressing the time between "signal—diagnosis—processing." For SRE teams, this is not just icing on the cake. What is truly expensive is not the cloud bill, but how much time people need to piece together clues when a system goes wrong. Once Datadog can reduce MTTR from one hour to fifteen minutes, or reduce the unit cost of an incident from $2000-$5000 to $500-$1000, it is no longer selling a dashboard, but cutting into enterprises' most intractable time costs. In other words, it is already very close to the "fast feedback execution layer."

However, stocks are not just valued for "effective products." Stocks are valued for "the second curve having penetrated from the product layer to the financial layer." And DDOG's most critical issue now is that this line has not yet fully penetrated. AI revenue is not separately disclosed, Bits AI and LLM observability are still embedded in existing product pricing, and have not formed an independent, identifiable, verifiable billing layer like NOW Pro Plus. In other words, the overall observability business already generates mature monetized revenue, but AI, viewed in isolation, is still in an earlier stage of commercialization; if revenue truly exhibiting agent characteristics were to be separated out, its scale would be much smaller than headlines suggest. You may disagree with the magnitude of the discount, but you cannot deny the fact: the market is already paying today for an AI curve that has not yet been separately monetized.

What's more difficult is that DDOG faces not just one competitor, but a set of structures that simultaneously exert pressure. First, open standards. OTel will not directly "kill" Datadog, but it will reduce the friction for customers to switch data sources and observability stacks. Second, open-source and low-cost alternatives, with Grafana being the most direct representative. Third, large customer in-house solutions, especially the 40%-50% in-house development ratio among the Top 100 internet companies, which effectively sets a natural ceiling for the high-end market. Fourth, are traditional competitors and new entrants. Thus, Datadog has formed a typical and also thorny combination: it is one of the products most easily favored by engineering teams, and also one of the types of companies most easily anticipated in the capital market.

Valuation amplifies this contradiction even more clearly. Superficially, DDOG has a 49x Non-GAAP P/E; but if SBC (Stock-Based Compensation) is viewed as a true shareholder cost, owner FCF for FY2025 is closer to $235 million, corresponding to an owner P/FCF of approximately 188x. This figure is not the company's official statement, but it is sufficient to illustrate the problem: the market might verbally value it with a multiple for a "high-growth software company," while genuinely more conservative holders are closer to paying for an asset where "AI is not yet independently monetized, SBC remains high, and multiple leakages still exist." This is not to say that DDOG is necessarily too expensive to touch, but rather that it requires a very different buying discipline. You cannot assume this stock is safe simply because you like its product. Conversely, DDOG might be the one in this group of names most prone to the situation where "the company is excellent, but the stock might not perform well."

Therefore, DDOG is more suitable for research focusing on two types of evidence: either the market first performs a reset, so that the valuation is no longer built upon highly anticipatory speculation; or the company first provides stronger commercialization evidence, such as independent disclosure of AI ARR, Bits AI or LLM observability forming independent SKUs, SBC/Revenue further converging below 20%, and growth remaining centered around 22%. Before these, DDOG is more suitable for high-frequency tracking, and should not be understood with the same level of certainty as NOW.

Chapter 5: CRM: Not Cheap, But a Split Asset That Must Be Dissected

The issues with Salesforce rarely appear in the headlines. Headlines are often positive: low P/E, a large customer base, high enthusiasm for Agentforce, and Data Cloud still holding second-curve imagination. The real problem is that these elements do not point to a single object. What is most worth emphasizing about CRM today is not whether it has AI, but the extremely uneven distribution of quality within it. Averaging such a company into a "low P/E cloud software leader" would be more dangerous than treating SNOW as a platform stock, because it would lead investors to mistake internal tension for overall cheapness.

Service Cloud is the first area to show problems. Its growth rate dropped from 12% to 6.5%; this is not a minor error, nor is it quarterly noise. This is more like a realistic signal of pressure on the seat-based model, even with a hint of cannibalization. In other words, the area of CRM closest to real workflows and most easily impacted by AI's direct alteration of unit economics has already begun to feel the impact. Meanwhile, Agentforce's headlines are very positive: $800 million in stated ARR, and the entire company is building narratives around it. However, the other half of the truth is equally clear: 67% of deals are still free, and pricing has been adjusted three times within 15 months, indicating high product interest, but a stable balance has not yet been found between billing units, customer expectations, and sustainable product-market fit (PMF). The enthusiasm and quality are not synchronized; this is CRM's biggest problem today.

Even more problematic than the misalignment between enthusiasm and quality is the misalignment in value capture. Even if Agentforce ultimately succeeds, Salesforce may not be able to retain the thickest layer of economic returns for itself. If we look at project implementation quotes and software license revenue together, a significant portion of the economic returns from project implementation and process re-engineering is still captured by service providers like Accenture and Deloitte, while CRM itself primarily receives software license revenue. This leads to a situation that can easily mislead capital markets: customers are satisfied, projects are launched, and management can present headline ARR, but the actual profit pool remaining on the shareholder side is not as substantial as the headlines suggest.

Salesforce's own statements regarding its risk boundaries are also not as optimistic as the market might imagine. It has not truly priced Agentforce as outcome pricing; instead, it's more like it's oscillating between task and seat-based pricing. The more realistic situation is that Marc Benioff verbally promotes an outcome narrative, but the actual pricing remains closer to task pricing, even being pushed down to a per-conversation billing model. The difference between the two is not merely a matter of marketing rhetoric; it's a difference of a 5 to 20 times pricing ceiling. As long as Salesforce doesn't truly bear outcome SLAs, it remains in the stage of "helping you process tasks," rather than entering the stage of "taking on your outcome budget." For an AI stock that has already been widely discussed, this difference is significant, because the premium the market loves to grant is precisely for the latter.

This is also why CRM can no longer be viewed using an overall P/E ratio. A more honest approach is to forcibly disaggregate it. Service Cloud is more like an older business line currently experiencing pressure on its seat model but still generating cash flow; Sales Cloud is in a weaker attribution environment; Platform, Data Cloud, and Agentforce are, on the other hand, a new machine that is faster but whose quality has not yet stabilized. Averaging these business lines into a 14.7x P/E ratio might seem cheap, but it merely masks internal contradictions. The overall P/E might appear low, but when disaggregated, the margin of safety is not wide, far from being as comfortable as "value stock + AI option" sounds.

Therefore, CRM is better tracked using a disaggregated perspective. The variables most worth monitoring are also very specific: how much real paid Agentforce ARR actually exists; whether Service Cloud can maintain growth of around 5%; whether Net Revenue Retention (NRR) will be disclosed for the first time and prove that the customer expansion engine has not stalled; whether the issue of service providers taking the lion's share will subside with product standardization; and whether Microsoft Copilot for Service will truly enter large customer budget comparison tables. Without these, CRM is more like a complex asset requiring continuous disaggregated monitoring, rather than a uniformly cheap stock mispriced by the market.

Chapter 6: SNOW: Not a Lack of Technology, But a Shifting Control Point

Snowflake is possibly the most easily misjudged among these four companies, because it's most likely to lead researchers into the cognitive inertia of "good technology, so the platform story is still intact." The problem has never been that SNOW lacks AI, nor that Cortex, AISQL, semantic layer, or Openflow lack product value. Rather, these AI advancements have not automatically translated into higher-level control.

Snowflake has a strong historical narrative, because it once successfully framed "centralizing data in one place" as a platform premium. However, the trickiest aspect of the AI era is precisely that data centralization no longer automatically equates to control point centralization. Because the source of truth still resides within customers' own ERP, CRM, application systems, and business event streams; data formats are becoming open, and the boundaries between the catalog layer and query engines are being redrawn; model vendors, customer-built solutions, and higher-level platforms are all beginning to directly access the data itself. Therefore, SNOW's greatest danger is not that customers dislike it, but that even if customers continue to like it, it may only be an increasingly important, yet increasingly replaceable, underlying layer within the entire AI stack.

This is also why the action of writing to Unity Catalog–managed Iceberg on April 6, 2026, was so important. Many people might view it as a neutral interoperability feature: customers prefer openness and supporting more standards, which is naturally a good thing. However, from a capital market perspective, it conveys another signal: Snowflake is willing to participate as a writer and query engine underneath the catalog layer controlled by Databricks. The question is not "is support good or bad," but "who is defining the control plane." If the answer is increasingly not SNOW, it means that the premium originally attributed to the platform owner is no longer as stable as before. In my opinion, this is already a strong signal of a control point reversal, because it's not just ordinary compatibility, but a redefinition of platform status.

Bringing commercialization into the picture makes this story even sharper. Cortex AI currently has an approximate $100 million run-rate, serving 9,100 customers, accounting for only about 2.2% of FY26 revenue. This is not zero, nor is it something to be ridiculed as having made no progress; however, it is still very far from "enough to rewrite valuation language." Meanwhile, under a more conservative AI gross margin assumption, model vendors would take a significant portion of Cortex's gross margin; Iceberg and customer-built solutions reduce data lock-in; Databricks, Fabric, ClickHouse, and DuckDB are capturing value at different layers; and cloud infrastructure also continues to take a portion of the profit margin. Thus, SNOW's problem, condensed into one sentence, is actually very simple: technology is advancing, and customers don't dislike it, but where new value ultimately settles is not entirely determined by SNOW. This is what a platform fears most. Because a platform is most valuable not when it does a lot itself, but when others cannot bypass it regardless of what they do. Snowflake is further from this state today than it was three years ago.

Therefore, I will no longer defend SNOW with the phrase "leader in AI data cloud." That statement might have had some explanatory power in 2024, but by 2026, it increasingly resembles remnants of an old map. A more honest view is to study it as a stock where "the data and AI workload bearing layer is important, but platform exclusivity is diminishing." Consequently, the valuation language must also change: it is not a platform that naturally deserves a higher EV/Sales multiple, but rather a name that needs to first prove that a platform premium should still exist. To prove this, at least a few things need to be seen: Cortex's contribution first exceeds 5% from 2.2%; Marketplace revenue begins to be viewed separately; after Databricks' IPO and Iceberg feature parity, the trend of control points continuing to shift externally does not worsen; and NRR does not further drop below 120%. Without these, it's difficult for me to place SNOW on the side of "deserving a higher platform multiple upfront."

Chapter 7: Valuation Bridge: Why These Four Companies Can No Longer Be Valued Using the Same Language

The reason why the market easily misjudges these four names is largely because segment labels are applied first, followed by valuation language. The correct order should be reversed: first, determine what level of control point you are paying for, and then decide what valuation language to apply.

If we simplify the old map, it roughly looks like this: All four companies are large software platforms, all talking about AI, and all have decent gross margins and a solid large customer base; therefore, they can be compared within a "software platform + AI option" framework. This language isn't entirely wrong; it merely omits the truly significant differences. Because once we factor in the billing layer, responsibility layer, write-back layer, and value leakage, the four companies should fall into four different valuation syntaxes.

NOW is better viewed through the lens of "workflow control layer + attach economics." Its core isn't seat growth, but whether AI can continuously attach as a higher ARPU workflow layer. Thus, the key is not a single target price, but a shift in worldview regarding attach rate, Average Contract Value (ACV), gross margin, and competitive suppression. DDOG is better viewed through "owner economics + AI monetization proof." As long as AI revenue has not independently materialized, and Stock-Based Compensation (SBC) remains a structural pressure, you cannot merely use Non-GAAP multiples to tell its story. CRM must be subject to a Sum-of-the-Parts (SOTP) valuation: the slow core must be viewed with slow-core methods, and new optionality must be viewed with new-optionality methods; an overall P/E will only mask the real problems. SNOW is better understood through "data and workload bearing layer + platform discount risk." For it, what's important is not how much a target price can calculate, but whether the platform premium implicit in EV/Sales should continue to be maintained or systematically decline.

The most important aspect of this table is not that it provides four labels, but that it truly isolates the "valuation bridge." That is to say, after understanding the object model, you shouldn't just perceive these four companies as different; you should immediately know why they must be discussed using different pricing languages. Otherwise, even the best business judgment will be negated by incorrect valuation syntax.

Chapter 8: The True Variables to Watch Are Not 'AI Adoption,' But These Five Things

If you only focus on a broad term like 'agent adoption,' 'AI revenue,' or 'customer feedback,' you will fundamentally misinterpret this group of names. A more effective system of variables is as follows.

First, has AI truly formed an independent budget? NOW is currently closest to this line, as Now Assist ACV and Pro Plus premium can already be seen separately. DDOG is not there yet, CRM is more like 'hype first, pricing later,' and SNOW is clearly still far off. Second, has AI genuinely written back to the system? NOW's advantage is most apparent. DDOG is approaching this step between diagnosis and execution. CRM primarily focuses on Service scenarios, while SNOW has almost no true write-back to customer SoR. Third, has the newly added value remained in their own hands? CRM and SNOW are weakest on this point; one sees a significant portion taken by service providers, while the other sees a large portion taken by model providers, customer self-builds, and higher-level platforms. Fourth, can gross profit and owner economics support the new narrative? DDOG is most acute on this point, followed by SNOW. Fifth, does competition target superficial features or the control point itself? The threats of Copilot to NOW and CRM, Iceberg/Databricks to SNOW, and Grafana/OTel/self-build to DDOG all fall into the latter category.

When applied to specific companies, the primary variable can be pressed even more directly: For NOW, watch Pro Plus attach rate and Now Assist/Pro Plus ACV, plus whether gross margin stabilizes around 77%-78%; for DDOG, whether AI ARR is separately disclosed, whether Bits AI forms a distinct SKU, and whether SBC/Revenue can continue to trend below 20%; for CRM, look at true paid Agentforce ARR, whether Service Cloud maintains +5% growth, and whether NRR will be disclosed for the first time; for SNOW, observe whether Cortex revenue share first crosses 5%, whether Marketplace revenue becomes visible, and whether the shift of control points brought by Databricks/Iceberg continues to worsen. As long as these variables remain unchanged, much of the 'AI is hot' discussion is not worth too much time.

Chapter 9: The Strongest Counterargument: Perhaps the Market's Unified Reward for Large Platforms is Actually Correct

The strongest counterargument is not directed at any single company, but rather at the overall ranking in this article. This counterargument would state: You are scrutinizing write-back, accountability, independent pricing tiers, and value capture too minutely. In the real world, large enterprise customers do not re-evaluate software with such precision. They will simply allocate more budget to large platforms that already have default distribution, sales systems, and customer relationships within the enterprise. According to this logic, NOW, CRM, DDOG, and SNOW will all ultimately benefit from AI, and the differences may not be as significant as stated here. Historically, the software industry has repeatedly demonstrated that distribution often reflects in financial statements before product details do.

This counterargument is not without merit. Especially CRM and SNOW, which are both likely to catch the tailwind of 'unified re-evaluation of large platforms' when sentiment improves; DDOG could also well achieve higher multiples due to rising risk appetite even before independent AI revenue disclosure; NOW's ceiling also does not prevent it from continuing to rise, as a ceiling and continued ascent are not the same thing.

However, I still do not side with this counterargument today, and the reason is simple: the strongest evidence for the four companies does not line up. NOW has already built a body of evidence across write-back, monetization, attribution, and relatively limited value leakage; DDOG has only partially completed the first two steps, with the latter two still unproven; CRM's hype and paid quality are still mismatched; and SNOW's technological progress and control-point position are not moving in the same direction. In other words, while a unified reward for large platforms may happen at the level of market sentiment, there is not yet enough research-level evidence to treat it as the base case.

Chapter 10: Decision Interface: How to Prioritize Research, and When to Rewrite the Thesis

If I were to prioritize based solely on research priority and evidence density today, I would look at it this way.

First, NOW. It is not the cheapest, but its structure is the most stable, and it is closest to a truly verifiable AI second curve. It is more suitable for continuous investment of tracking resources. Its premises are also very clear: Pro Plus attach continues to trend upwards, ACV progresses towards $900 million to $1 billion, gross margin does not continue to compress, and Armis integration does not encounter major errors. If these conditions are systematically undermined, it will still be a good company, but the valuation narrative would need to narrow significantly.

Second, DDOG. This is a classic case of 'great business, tougher stock.' A more rational approach is not to fully price in the future at high multiples prematurely, but to wait for one of two catalysts: either the market first provides a more digestible reset, or the company first delivers on an independent AI monetization layer, SBC convergence, and separate AI ARR disclosure. Prior to either of these scenarios, DDOG is more suited for high-intensity tracking and should not be understood with the same level of certainty as NOW.

Third, CRM. It is not suitable for holistic analysis; it should only be studied in parts. If you are looking at it, you must first clarify what exactly you are examining: the defensive cash flow of Service Cloud, or the optionality of Agentforce/Data Cloud. The true catalysts for an upgrade are already very clear: genuine paid ARR, NRR disclosed for the first time and above 115%, and Service Cloud maintaining growth. Without these, a unified P/E should not be used to replace disaggregated judgment.

Fourth, SNOW. This is not to say it lacks rebound opportunities or product highlights, but rather that its valuation should not initially assume it deserves the narrative of 'natural AI platform re-rating.' The most honest approach is to first wait for harder evidence: Cortex's share of revenue first exceeding 5%, Marketplace revenue becoming visible, and the abstraction pressure from Databricks/Iceberg not continuing to worsen. Prior to these, SNOW is more like a stock that needs to first prove its platform premium still exists, rather than a stock that automatically enjoys that premium.

To condense it into the most concise research sequence: First, follow NOW's attach economics, then await DDOG's independent commercialization, then track CRM with a disaggregated perspective, and for SNOW, first wait to see if the platform premium stabilizes.

Chapter 11: The Most Common Misinterpretations for Each of the Four Companies

The purpose of this table is not to reiterate the four companies, but to remind readers that the most dangerous misjudgments for these four names occur at four different levels. NOW's boundaries are easily overestimated, DDOG's commercialization penetration is easily overestimated, CRM is easily obscured by averages, and SNOW is most prone to mistaking 'product progress' for 'platform progress'.

Chapter 12: Not All AI Revenue Is Created Equal, This Must Be Made Clear

Another point where this group of names can easily mislead people is the term 'AI revenue' itself. Many cross-sectional studies compare them in a single table, as if merely being able to report a number means they are on the same commercialization trajectory. This is not the case.

NOW's AI revenue is more akin to a higher ARPU attach layer within a workflow. Its significance lies in customers already paying separately for 'integrating AI into real processes.' DDOG's AI revenue, for now, is more like enhanced value embedded within existing products. It is certainly important, but until it becomes a separately monetized object, it cannot be simply equated with NOW's ACV evidence. CRM's AI revenue is more like a mixture of high-hype adoption and early pricing experimentation; the real issue is not whether there's a headline ARR, but paid quality and monetization stability. SNOW's AI revenue, to date, is more closely a technically sophisticated but still small incremental consumption layer, far from rewriting the entire company's valuation narrative.

Therefore, what these four companies are truly worth comparing is not 'whose AI revenue is growing faster,' but rather: does this revenue correspond to an add-on feature, an independent workflow, task-based billing, or something closer to an outcome-based budget? Once this question is clearly asked, many superficial commonalities will automatically disappear.

Chapter 13: Over the Next 4 to 8 Quarters, What Evidence Is Most Likely to First Reshuffle the Ranking of These Names?

If this group of names is to be re-ranked in the future, the first movers will not be press conferences or slogans, but rather several sets of harder evidence.

For NOW, the key is not another agent, but whether Pro Plus attach economics can continue to hold. If the attach rate continues to trend upwards, Now Assist ACV continues to materialize, and gross margin pressure does not further worsen, then its high valuation will appear more performance-justified; conversely, the market will begin to re-evaluate whether it is a 'high-quality workflow company' or a workflow company whose future has already been priced in for many years.

For DDOG, the key is not another product demo, but whether the AI monetization layer finally becomes independently visible. As long as AI ARR remains embedded in legacy products, SBC remains high, and the leakage channels of Grafana/OTel/self-build remain in place, it will continue to be a name where 'the product is stronger than the stock.' What can truly change its ranking is commercialization evidence catching up to product hype.

For CRM, what will first rewrite the ranking will not be Agentforce's headline figures, but whether paid quality and the old core can both hold steady. As long as genuine paid ARR, Service Cloud growth, and NRR do not all move up together, CRM's 'low P/E + AI optionality' will remain more a convenient phrase than a decision-ready statement.

For SNOW, whether its ranking can be rewritten ultimately depends on platform position, not feature count. If Cortex's share of revenue first surpasses 5%, Marketplace revenue becomes visible, and the shift of control points does not continue to worsen, then the market will have reason to re-assign it a higher platform narrative; otherwise, technological advancements and customer satisfaction alone will not be sufficient to automatically offset platform discount.

Chapter 14: Why These Four Companies Are Not Suited for Management in the Same Basket

Managing NOW, DDOG, CRM, and SNOW within the same 'AI software platform basket' has its biggest problem not in its crudeness, but in that it automatically obscures the sources of risk that genuinely require different treatment. NOW's primary risks lie in attach economics, cost structure, and M&A integration; DDOG's primary risks are the delayed appearance of an independent monetization layer, compounded by SBC and standardization pressures; CRM's primary risks are quality mismatches within its fragmented structure; and SNOW's primary risks involve the downward shift in valuation narrative caused by platform abstraction. The triggers, verification pace, and correction methods for these four types of risks are all distinct.

This means that even if you ultimately track these four names simultaneously in your portfolio, you should not assume they are interchangeable. They are not four 'AI software stocks' of varying degrees, but four types of assets at different levels. Only by acknowledging this first will subsequent valuations, position sizing, and validation rhythms become honest.

To distill it into a more practical statement: NOW is more concerned about valuations getting ahead of themselves, DDOG is more concerned about commercialization proof coming too slowly, CRM is more concerned about internal misalignment not converging long-term, and SNOW is more concerned about platform control points continuing to shift externally. These four types of risks are not simply different in strength, but fundamentally different in their failure mechanisms and entirely different in their validation rhythms.

Chapter 15: Summary

The market loves to frame complex problems as sectors because it's the path of least resistance. However, what truly determines returns is often not the sector label, but who genuinely owns state changes, billing units, and responsibility boundaries within a workflow.

Among this group of names, NOW is closest to pushing AI into the write-back layer of real workflows; DDOG is closest to transforming rapid feedback into rapid processing; CRM is not one company, but at least two machines operating at different speeds; SNOW's issue is not the lack of AI, but rather that its control points are beginning to shift externally. As long as these four assessments hold true, they should no longer share the same valuation language.

AI will not reward software platforms equally.
What it rewards has never been 'who can talk about more features,' but rather: who truly controls state changes, who truly captures the right to charge, and who keeps the added value on their own books.
This is the main theme of my current view on these four names.

Chapter 16: The Relationship Between This Article and My Previous SaaS NRR Report

If you have read my previous SaaS NRR Report, you will most likely ask a very natural question at this point: why was CRM's risk-reward ranking higher in that article, while in this article, I place NOW closer to the AI control point and elevate DDOG's business position? This is not because I forgot the former text, but because both articles were answering two different questions from the outset.

The core of that NRR report was to correct a common misjudgment in the old SaaS era: the market was too accustomed to using Rule of 40, singular NRR, and Non-GAAP profit to price software companies, without further questioning what engines drove these growth and profits. Therefore, that article was more like a valuation map. It primarily answered: if one must make trade-offs among this group of SaaS assets today, whose cash flow quality is more genuine, whose margin of safety is thicker, and whose odds are more suitable for initial portfolio inclusion.

This article, however, takes it a step further. It doesn't first ask who is cheaper, but rather which layer of the AI software stack each of these four companies controls. In other words, it's more like an object map, or a control point map. It primarily answers: in the process of AI reallocating the software profit pool, who is closer to the write-back layer, who is closer to the real workflow, whose right to charge is more stable, and who is more easily abstracted by higher layers.

Thus, while the two articles may appear to have discrepancies, they merely reflect different observational altitudes. The former article is closer to a 'stock perspective,' which is why CRM's relative margin of safety would appear more prominent; the latter article is closer to an 'asset perspective,' which is why the importance of NOW and DDOG in their industry positions is viewed more heavily. The former addresses 'which name offers better odds today,' while the latter addresses 'which type of control point is more worthy of a higher valuation language in the coming years.' This is not a mutual refutation, but rather a difference in sequence.

If we must distill the relationship between the two articles into one practical line, it is this: first use this article to define the object, then go back to the NRR report to determine the odds. First clarify whether what you hold is a workflow write-back asset, a rapid-feedback execution asset, a split asset, or a platform whose control points are shifting outward; then discuss whether it is expensive, safe, or worth a larger position. If you do only the latter, you risk building valuation work on the wrong object definition; if you do only the former, you risk buying a good asset as a bad stock.

Therefore, I do not wish to present these two articles as 'new views overturning old ones.' A more accurate statement is: that article addressed why SaaS pricing language became ineffective, while this article addresses why object definitions have changed in the age of AI. The former remains important because the market ultimately always returns to cash flow and odds; the latter is equally important because if even the object itself is incorrectly defined, subsequent valuations are merely elegantly wrong calculations.

This is also the ultimate relationship I hope readers understand: the same company can entirely possess two faces simultaneously. One is its current stock face, which determines whether you should rush to pay this price now; the other is its asset face for the coming years, which determines whether it qualifies to remain in a higher position long-term. What should truly be done is not to force the two articles to be perfectly consistent in their surface-level rankings, but to first clarify: which question are you currently answering.