AI Impact on SaaS Stocks: Analysis of Salesforce, ServiceNow, and Snowflake

Compiled & Organized by Shenchao TechFlow

Guest: Nico

SaaS Software Stocks Under the AI Nightmare: CRM vs. NOW vs. SNOW—Which One Is the True Undervalued Double-Opportunity? A 10,000-Word Analysis of the Next Wave of Software Stock Opportunities

Podcast source: Nico Frontier Alpha

Broadcast date: May 21, 2026

Over the past six months, Wall Street has characterized a brutal sell-off as the "SaaS apocalypse," with Salesforce, ServiceNow, and Snowflake dropping by more than half from their peaks. J.P. Morgan’s congestion model shows institutional holdings in the semiconductor sector have surged to 99.3%, while software sector holdings stand at just 22.8%, reflecting a historic divergence in market sentiment. At this juncture, investor Nico has offered a perspective contrary to the mainstream narrative: AI is not killing the software industry, but rather eliminating companies that merely sell functional interfaces, while rewarding platforms that provide infrastructure and governance. Although the current software sector lacks the same level of industry momentum as hardware, it offers a more favorable risk-reward profile and better value.

The most valuable part of this episode is breaking down the three companies side by side using the same evaluation framework: Salesforce (13–14x forward P/E, $14.4 billion in free cash flow, $50 billion in share repurchase authorization) represents the “margin of safety” camp; ServiceNow (with its AI Control Tower narrative and NVIDIA’s Jensen Huang endorsing it for three consecutive years) embodies the “clearest AI narrative” camp; and Snowflake (pay-as-you-go pricing, 42% year-over-year growth in RPO, yet still GAAP-negative) stands as the “high elasticity, high risk” camp. Salesforce and Snowflake will both report earnings on May 27, followed closely by Snowflake’s annual conference and Microsoft’s Build conference—these catalysts will form the most immediate short-term observation window.

The "SaaS Apocalypse" and Extreme Market Sentiment

• The software sector has been brutally crushed—not just one company is having problems, but the entire software sector has been sentenced to death by the market.
• JPMorgan's congestion model shows that institutional positioning in the semiconductor sector has surged to 99.3%, while the software sector's congestion stands at just 22.8%, indicating a historically extreme divergence in sentiment.
• The good news for the hardware sector is that everyone has already bought in, and it’s already priced in by the market; the bad news for software is that most people have already sold, leaving room for an upward rebound. Over the next three months, if you’re only looking at industry sentiment, hardware will likely outperform—but if you’re considering upside potential, odds, and value, software may actually offer better prospects.

The Impact of AI on the SaaS Business Model

• Many features that SaaS companies once relied on for charging fees can now be used to create a functional prototype in a very short time using AI, requiring no programming experience whatsoever. The real concern in the market is that the scarcity and moat of SaaS functionality are collapsing.
• If an AI agent can do the work of ten people, then a company that previously needed to purchase 1,000 accounts now only needs 100. This is what Wall Street recently refers to as seat compression.
• An agent doesn't need a UI, no dashboard, no beautiful interface—it only needs data and APIs. This means SaaS software has been disrupted by AI, reduced from the primary entry point of enterprise workflows to merely a backend data storage system.

Salesforce's Transformation and Valuation

• Buying Salesforce isn't essentially a bet on a high-growth story or its successful AI transformation using a valuation dozens of times higher; rather, it's a balanced assessment based on comparing its intrinsic value with its current market price—it is indeed currently in a relatively undervalued position.
• Agentforce has shifted its pricing model from per-user to per-task: past revenue was tied to the number of employees, while future revenue will be tied to overall workload. By successfully implementing the per-task pricing model, Salesforce can smoothly transition from a seat-based economy to a task-based economy.
• Microsoft’s Dynamics 365 plus Copilot is the most significant long-term threat to Salesforce. If salespeople no longer open Salesforce at all, but instead rely on Copilot in Outlook or Teams to automatically update customer records, Salesforce could degrade from a workflow entry point into a backend database.

ServiceNow's AI Control Tower strategy

• ServiceNow isn't trying to recreate ChatGPT; it aims to become the governance, orchestration, and execution layer for enterprise-grade AI agents. Regardless of which AI a company uses, if that AI integrates into business processes, accesses enterprise systems, or performs enterprise tasks, it must go through ServiceNow for governance and orchestration.
• This positioning is similar to Apple’s iOS—Apple doesn’t build every app itself, but all apps run on iOS. ServiceNow aims to follow the same path in the future.
• Huang Renxun's exact words were: "ServiceNow is essentially the enterprise operating system of the AI era."

The Snowflake consumption model paradox

• Snowflake fears not that customers won’t use it, but that they’ll use it too well. When companies find their Snowflake bills too high, they push engineering teams to optimize queries, compress storage, or even replace low-value tasks with open-source tools—this is the double-edged sword of consumption patterns.
• Snowflake's net revenue retention rate has declined from 131% to 126%, and now to the latest 125%. While still healthy, the downward trend indicates that expansion among existing customers is slowing compared to before.
• Snowflake is the fastest-growing of the three, with the most straightforward AI data infrastructure logic and a natural immunity to traditional SaaS business models; however, it is also the most highly valued, faces the fiercest competition, and has the weakest profitability. High reward, high risk.

Historical analogies and final judgment

• The narrative that "AI is killing software" has been oversimplified. What’s actually happening is that AI is replacing software that only sells feature interfaces, while simultaneously rewarding platforms that sell infrastructure and governance. Not all software will be disrupted.
• When the dot-com bubble burst in 2000, the dominant market belief was that “the internet would kill all traditional companies,” but in the end, it wasn’t just internet companies that survived—it was also the traditional companies that embraced the internet early and integrated these tools into their operations. Twenty years later, the same logic applies to this wave of AI.

At the start of 2026, the narrative that “AI is killing the software industry” ignited the entire U.S. stock market. Since then, the entire software sector has been shrouded in the nightmare of being disrupted by AI. Even industry leader Microsoft was not spared, with its stock plunging over 25% during the year—and nearing a 40% drawdown from its all-time high, approaching the magnitude of the 2022 U.S. bear market. Popular software stocks from recent years, such as Salesforce, ServiceNow, and Snowflake, have each lost more than half their market value. This is not an issue affecting just one company, but rather the entire software sector being sentenced to death by the market. Wall Street has dubbed this event “SaaS Doomsday.”

Over the past nearly six months, both retail and institutional investors have been doing the same thing: going long on hardware and short on software, leaving the software sector severely battered. However, recently, several unusual signals have quietly emerged. JPMorgan’s congestion model shows that institutional positioning in the semiconductor sector has surged to 99.3%, while software sector congestion stands at just 22.8%—a historic level of sentiment divergence. At this very moment, U.S. President Trump quietly invested millions of dollars to buy software stocks at a discount; Bill Ackman, Wall Street’s most renowned value investor, simultaneously took a heavy position in Microsoft, the largest software company; and NVIDIA’s CEO, Jensen Huang, the world’s most valuable company, made his third consecutive trip to Las Vegas in person to endorse a software company.

So, is AI going to kill the entire software industry, or is it giving us a once-in-a-decade opportunity to buy the dip? In today’s video, I’ll break down three of the most representative software companies: Salesforce, ServiceNow, and Snowflake.

The story of AI killing the SaaS industry and the crash in software stocks begins back in January of this year. On January 30, Anthropic—the company behind the Claude large model—quietly released 11 plugins on GitHub called Claude Cowork: a simple code repository accompanied by a blog post. Yet within 48 hours of its release, global software stocks suffered massive losses. According to market estimates, the software sector lost a total of $285 billion in market value.

Why is everyone so panicked? A CNBC reporter conducted an experiment that kept every SaaS company executive up at night: using Claude Code, he replicated a website like Monday.com in just one hour, at a cost of only $5–15. Monday.com is a publicly traded U.S. stock company specializing in project management software, with a market valuation in the billions. In just one hour and for a few dollars, a reporter created a project management demo that looks nearly identical to Monday.com.

Of course, this doesn't mean it truly replicated a public company—real Monday.com has enterprise permissions, data security, an integration ecosystem, and sales channels, none of which can be achieved by AI in just an hour; they require time to develop and accumulate. But the most alarming aspect of this experiment is that many of the functional interfaces SaaS companies traditionally charged for can now be prototyped in a matter of minutes using AI, without any programming experience whatsoever. Behind this story lies the market’s deeper concern: the scarcity and moat of SaaS functionality are collapsing. The traditional SaaS model, which charged per user, may no longer be viable under AI’s pressure. This also reveals the ambition of underlying AI model providers—not just optimizing large model performance, but directly entering the application layer to claim a share of this massive market.

SaaS stands for Software as a Service. Its core concept is simple: moving traditional on-premises software, previously installed on corporate servers, to the cloud, where customers pay monthly or annually for access to the software. Over the past 20 years, this model has been the software industry’s most powerful wealth generator.

The core pricing model for virtually all SaaS companies is based on per-user licensing. If a company has 1,000 employees who need to use the software, it must purchase 1,000 accounts and pay ongoing subscription fees—ranging from tens to hundreds of dollars per account annually. The more frequently and longer the software is used, the stronger customer retention becomes, as the company’s entire workflow and data become embedded within the SaaS platform, making migration or switching costs prohibitively high in the short term. This is precisely the fundamental logic behind the highly profitable, asset-light SaaS industry—and the primary reason Wall Street has historically valued SaaS companies at price-to-earnings multiples of dozens or even hundreds of times over the past two decades.

However, with the explosion of the AI wave—especially since entering the Agent era—the foundation of this logic has begun to shift. Market concerns about the SaaS industry primarily stem from two factors.

Layer 1: Seat compression

The most immediate layer of panic is that AI agents are replacing employees, causing a sharp decline in SaaS subscriptions and a significant drop in revenue and profits. SaaS companies charge per user—companies buy one seat for each employee using the service. But with the advent of AI agents, this model has been completely overturned: if one AI agent can do the work of ten people, a company that previously needed 1,000 seats now only needs 100. This is what Wall Street has recently been calling “seat compression.”

The revenue formula for a SaaS company is “number of customers × average seats per customer × price per seat.” Over the past 20 years, all three variables have been rising, but under the impact of AI agents, average seats per customer is now facing its first structural downward risk. The market is concerned that the SaaS business model may be disrupted by AI.

Layer 2: Agent workflows bypass the SaaS interface

A deeper level of panic arises when, beneath agent-driven workflows, SaaS software is bypassed entirely and reduced to a supporting role. This is the core reason the market is truly alarmed. The traditional SaaS business model implicitly assumes that software is used by humans. Salesforce designs UIs, beautiful dashboards, and workflows—all fundamentally aimed at cultivating user habits and increasing user retention. But agents require no UI, no dashboards, no polished interfaces; they only need data and APIs.

Once Claude can directly connect to plugins for your Salesforce, Notion, Google Drive, and Slack, workflows undergo a fundamental transformation. Previously, sales representatives would open Salesforce directly to check customer data, follow up on contracts, and review after-sales information—daily tasks heavily relied on Salesforce’s software interface. Now, sales representatives can simply open Claude to complete these repetitive tasks, while Claude uses APIs to read from and write to Salesforce, eliminating the need for sales reps to interact with Salesforce’s software interface at all.

This means SaaS software has been disrupted by AI, relegated from the primary entry point of enterprise workflows to a mere backend data storage system. The terrifying aspect is that it directly alters the value distribution chain. Previously, users interacted most frequently with SaaS software, but now they spend more time engaging with AI agents. The stage where users spend the most time holds the greatest pricing power. Under these conditions, SaaS software has become a supporting player to AI agents. The strongest moat of SaaS—long-standing user habits and workflow integration—was fundamentally built on the assumption that “users heavily use UI interfaces,” but agents are changing this dynamic. This is enough to trigger widespread market panic.

Meanwhile, the macro interest rate environment remains tight, and major tech companies have directed nearly all of their capital expenditures toward AI infrastructure, squeezing corporate software procurement budgets. As a result, long-duration software growth stocks have experienced the most significant valuation compression. So far this year, the entire software sector has significantly underperformed the S&P and Nasdaq indices during the same period, leading to a polarized market sentiment where investors are blindly going long on hardware and short on software.

JPMorgan's congestion analysis shows that semiconductor industry congestion has reached a historic high of 99.3%, indicating that nearly all investors are positioned in the same direction. More notably, short positions in the software sector are steadily increasing, with the squeeze risk indicator reaching an extreme level of 100%. When panic reaches its peak, market turning points and reversal signals often begin to emerge.

This data does not imply that funds are immediately exiting the hardware sector to shift into the software sector. Rather, it is a risk signal indicating that hardware has become the most crowded sector for retail and institutional trading, making blind long positions in hardware increasingly less cost-effective—naturally prompting capital to seek shifts between sectors. Moving from overvalued hardware to undervalued software is akin to transitioning from a highly congested sector that has been fully priced in the short term, to one still suppressed by fear narratives but where fundamentals may soon improve.

The good news for the hardware sector is that everyone has already bought in, and it’s fully priced in by the market; the bad news for software is that most investors have already sold, leaving room for a rebound. My assessment is clear: over the next three months, if you’re judging solely by industry momentum, hardware will undoubtedly be stronger. But if you’re evaluating upside potential, odds, and value, software may actually offer better prospects. In other words, hardware remains the primary theme in AI, but it’s become overly crowded in the short term; software is the catch-up opportunity, with greater flexibility and higher risk-reward potential over the next three months.

Primarily, the software sector has been severely hit over the past few months. Amid AI-related panic, there was widespread and indiscriminate selling of software stocks, with the market selling first and asking questions later. This has led to many high-quality software companies—those with strong business moats, substantial data accumulation, and active adoption of AI—being unfairly punished.

Moreover, over the next several dozen days, the software sector will face numerous catalysts. For instance, on May 27, Salesforce and Snowflake will both release their latest earnings reports on the same day—these reports will address a core question: Is AI consuming SaaS, or is it redefining its value? Immediately following, from June 1 to 4, Snowflake will host its annual conference in San Francisco, centered on data infrastructure and the real-world deployment of enterprise AI. Then, from June 2 to 3, Microsoft will hold its Build conference, focusing primarily on AI agents, Copilot, developer workflows, and enterprise AI applications. When these catalysts converge, they could reinforce the rebound trend in software stocks. If the market begins to believe that AI agents are not meant to kill software but rather to deploy through software platforms, companies like ServiceNow, Salesforce, and Snowflake could all benefit significantly.

Company Background

Salesforce’s code is CRM, matching the name of its core business—it is the world’s largest customer relationship management software company and one of the most iconic firms of the SaaS era. In simple terms, it helps businesses manage their customers. But “managing customers” here isn’t just about sales reps opening a webpage and entering a few client details; its true value lies in serving as the central system of record for enterprise customer data.

Who the customer is, which employees have followed up, what products they’ve purchased, where the contract stands, whether there have been any after-sales complaints, and how many times they’ve been reached by marketing—these most critical data points across the customer lifecycle are all stored in Salesforce. These are the core customer assets of any enterprise. AI can help generate emails, summarize meetings, and automatically draft sales scripts, but without a reliable customer database, AI wouldn’t know how to do any of these things—that’s precisely why Salesforce holds its central position. AI may disrupt Salesforce’s front-end features, but it’s unlikely to eliminate its core.

Salesforce is, on one hand, one of the most classic traditional SaaS companies, directly facing pressure from agent-driven seat compression; on the other hand, it serves as the foundational data platform for many enterprise customers and is not a trivial tool that can be easily replaced. This is the core lens through which we analyze Salesforce: Is it an outdated software company on the verge of being disrupted by AI, or is it a cash-generating machine that the market has overly pessimistically priced?

Salesforce currently serves over 150,000 enterprise customers, ranging from startups to Fortune 500 companies. The company was founded by Marc Benioff in 1999. Benioff, a former Oracle employee, was once Oracle’s youngest vice president and one of the earliest protégés of Oracle’s founder, Larry Ellison. Later, he left to start his own company, proposing a highly radical idea at the time: enterprise software should not be sold on physical discs for installation on customers’ servers, but rather run in the cloud and subscribed to on a monthly or annual basis.

This idea was extremely radical in 1999. At that time, traditional giants like Microsoft, Oracle, and SAP primarily followed a model of selling software to enterprises for on-premises deployment. At that moment, Benioff alone championed the slogan “No Software,” and eventually, the SaaS business model prevailed, making Salesforce synonymous with the SaaS industry.

Benioff is known for his sharp instincts and ability to bet on the right direction. When he first introduced Agentforce last year, the entire market dismissed it as a marketing gimmick, but over the past few quarters, Agentforce has delivered impressive results. The latest disclosure shows that Agentforce’s ARR has reached $800 million, a 169% year-over-year increase. Whether you believe Salesforce can successfully transition into AI largely depends on whether you believe in Benioff himself.

Product Matrix

Many people think Salesforce is just a CRM tool, but after more than two decades of expansion and acquisitions, it has grown into a vast enterprise software platform.

At its core is Sales Cloud, its founding product, which helps sales teams manage customers, opportunities, and the sales funnel. The sales operations of countless enterprises worldwide are built on this product. Following Sales Cloud, Salesforce expanded with Service Cloud, specifically designed for customer service and post-sales support—everything from phone complaints, email inquiries, live chat interactions, to backend ticket assignment and workflow processing runs on Service Cloud. Further extending outward, Marketing Cloud handles digital marketing, enabling businesses to execute targeted campaigns, email marketing, and track advertising performance; Commerce Cloud manages e-commerce, helping businesses sell products online.

Together, these four components enable Salesforce to cover virtually every aspect of how businesses interact with customers—from acquisition and conversion to customer service and repeat purchases—providing dedicated products for the entire customer journey.

But Salesforce’s ambitions go beyond this. Over the past few years, it has invested heavily in acquisitions. It acquired MuleSoft (which specializes in system integration—helping enterprises connect data across multiple software systems often used simultaneously), Tableau (which provides data visualization and business analytics, turning customer data from CRM into charts and insights), and Slack (an enterprise communication and collaboration platform similar to China’s Feishu or DingTalk). Last year, it also acquired Informatica (which focuses on enterprise data management, helping organizations clean, integrate, and govern data scattered across various sources).

By putting these acquisitions together, Salesforce has effectively built a comprehensive ecosystem centered around customer data, with CRM at its core and layers of integration, analytics, collaboration, and data governance surrounding it. The company’s newest and most critical addition is Agentforce, an AI agent platform launched last year and Salesforce’s most important response to the impact of AI.

Business Model: From Seat Economy to Task Economy

Salesforce’s business model is the quintessential SaaS model, charging per user. Companies purchase licenses based on the number of sales staff who need the CRM, with each license costing around $100 or more per month, billed annually. While individual licenses may seem inexpensive, when aggregated across thousands or even tens of thousands of sales, customer service, and operations staff in a large enterprise, these costs add up to a highly stable recurring revenue stream—this has been the foundation of Salesforce’s profitability over the past two decades.

But since the arrival of AI, this easy-profit logic has begun to unravel. If an AI agent can automatically conduct customer research, write emails, manage sales funnels, and follow up with clients, does a company still need so many salespeople? This is precisely what the market fears most—seat compression. Salesforce is one of the most prominent companies frequently cited in such market discussions.

Benioff himself recognized this issue. Starting last year, Salesforce initiated a bold yet critical business model transformation, retaining its subscription fees while introducing a new usage-based product tailored for the AI era called Agentforce. In simple terms, the traditional model was “pay based on the number of accounts you purchase,” while the new model is “pay based on how many tasks your AI agents perform.” Salesforce refers to this usage metric as Agentic Work Units (units measuring work completed by AI agents).

The logic behind this new model is brilliant. If AI can truly replace a portion of human labor, the number of traditional seats may decrease, but at the same time, the number of tasks performed by AI agents could increase dramatically. Previously, a salesperson might follow up with 20 customers per day; in the future, a single AI agent could simultaneously manage 200 customers. While human seats decline, the volume of tasks executed by AI could double—or even increase tenfold. As long as the usage-based pricing model works, Salesforce can smoothly transition from a seat-based economy to a task-based economy, potentially significantly increasing revenue per customer. Past revenue was tied to the number of employees; future revenue will be tied to overall workload. This is the core significance of Agentforce—it could fundamentally reshape Salesforce’s entire pricing logic and business model.

Of course, this story has not yet fully materialized. Although Agentforce’s ARR has reached $800 million and is growing rapidly, it still accounts for less than 2% of Salesforce’s $41.5 billion annual revenue. Moreover, Salesforce may be facing more severe seat compression than any other SaaS company, because it sells licenses to sales reps, customer service agents, and marketers—a company with 10,000 employees might purchase 3,000 to 5,000 Salesforce accounts. Yet these are precisely the roles most vulnerable to AI agent replacement: drafting emails, following up with customers, generating sales copy, and answering client inquiries—all tasks where AI large models excel. It will be extremely difficult for a new business contributing just 2% to offset the decline in traditional seat sales.

Given all this, why am I still saying Salesforce is worth paying attention to right now? It’s not because I believe the new Agentforce business story will necessarily outperform the old SaaS revenue model, but because Salesforce is currently trading at just a 13–14x forward P/E ratio—a valuation that already prices in pessimistic expectations. It also generates $14.4 billion in free cash flow and has a $50 billion share repurchase authorization.

Therefore, buying Salesforce is not essentially a bet on a high-growth story or its successful AI transformation at a valuation dozens of times higher, but rather a balanced assessment based on the comparison between its intrinsic value and current price—Salesforce is indeed currently positioned at a relatively undervalued level. Of course, this margin of safety is not unconditional; if AI truly causes a significant decline in traditional revenue streams and Agentforce fails to compensate, Salesforce’s valuation may still face further compression. However, as long as its core business remains stable and share repurchases continue, even partial realization of Agentforce’s potential could lead the market to revalue the company, triggering a stock price rebound.

Moat

Salesforce’s strongest moat is the vast amount of data accumulated by customers over the past 20+ years. A company that has used a CRM for 10 years may have stored millions of customer records, hundreds of thousands of sales processes, and tens of thousands of custom fields. Moving all of this data would be equivalent to tearing down and rebuilding the entire digital foundation of the business—the migration cost far exceeds the cost of continuing to pay.

Where are Salesforce’s weaknesses? Microsoft’s Dynamics 365 combined with Copilot represents the greatest long-term threat to Salesforce. As the world’s largest software company, Microsoft’s B2B productivity tools have already penetrated the vast majority of large enterprises globally. Dynamics 365 is Microsoft’s CRM product, directly competing with Salesforce’s core business, and has maintained annual growth rates above 20% over the past several years. Most critically, Dynamics 365 is deeply integrated with Copilot, Teams, and Outlook—tools that employees use daily as their primary software entry points. If sales representatives no longer open Salesforce at all, but instead rely on Copilot within Outlook or Teams to automatically update customer records, Salesforce could be reduced from a primary workflow entry point to a background database. This is precisely what Benioff fears most—and the greatest long-term uncertainty facing Salesforce.

Latest financial report data

For the final quarter of the last fiscal year, the data was as follows: annual revenue reached $41.5 billion, a 10% year-over-year increase; total RPO amounted to $72 billion, up 14% year-over-year; free cash flow was $14.4 billion, up 16% year-over-year; and $14.3 billion was returned to shareholders for the year, including $12.7 billion in stock repurchases and $1.6 billion in dividends. Additionally, Salesforce has just approved a new stock repurchase program of up to $50 billion. The new business, Agentforce, achieved an ARR of $800 million, a 169% year-over-year increase, with 29,000 deals signed.

However, it’s important to clarify that 29,000 transactions do not equate to 29,000 large clients, nor do they all represent high-value contracts. This data only indicates rapid product expansion; what truly determines valuation is whether the company can subsequently increase average revenue per customer and net revenue retention rate. During this earnings call, the company also raised its fiscal year 2030 revenue target to $63 billion.

Overall, Salesforce’s fundamentals are indeed very strong. At the last earnings call, CEO Benioff himself stated that this was the most outstanding year in the company’s history and the best-performing year ever in the software industry. He went on to say that this presents a great marketing and buying opportunity, prompting the company to increase its share repurchase authorization to $50 billion. This tone is unmistakable—the management team is highly satisfied with the earnings results and is directly countering market sentiment, arguing that the market has been overly pessimistic and that Salesforce’s stock has been unfairly undervalued.

When I made the video, Salesforce's stock price was only $180, with a forward P/E ratio of 13–14x. Compared to the software bull market of recent years, where valuations often reached 30x or 40x, this represents a significant compression and is the lowest valuation level in recent years.

Catalysts and Risks

The bullish case is simple: it’s undervalued, has stable cash flow, is currently executing a large buyback program, and its new Agentforce business is accelerating growth. Salesforce’s earnings report on May 27 is highly anticipated and represents the most direct near-term catalyst.

The bearish reasons are that its growth rate is only 10%, which is not fast by software industry standards; concerns about its business model being disrupted by AI have not been resolved; and the uncertainty surrounding the new Agentforce business remains high. The market’s biggest question is whether Agentforce can grow large enough to drive the company’s overall revenue and profitability and enable its full transition to AI. These questions remain to be answered over time.

For the May 27 earnings report, please pay attention to the following: First, whether Agentforce’s ARR continues to maintain year-over-year growth above 100%. If the growth rate slows, it may indicate certain risks in the AI transformation—key will be how management responds to this.

Second, have there been any noticeable declines in SaaS seat fee-related business? If similar trends emerge, everyone should be cautious, as the market may continue to hype the narrative that "AI is eating SaaS."

In addition, it is worth paying attention to whether the company continues to maintain an optimistic outlook for the future, and whether management continues to positively address the impact of AI on the SaaS business model.

Looking solely at last quarter’s earnings report, I find the management team to be very clear and optimistic—they do not believe AI will kill Salesforce; instead, they see AI as a catalyst to elevate Salesforce from a SaaS application company into a platform for enterprise agents. However, from a data perspective, this narrative is still in its early validation phase. Personally, I don’t think it’s necessary to rush to conclude whether Salesforce has been disrupted by AI or has successfully transitioned its business around AI. What matters more to me is that its valuation is currently at one of its lowest levels in recent years, and when combined with the company’s solid fundamentals, the current entry point offers high value and favorable risk-reward potential. That said, the long-term narrative still centers on AI—whether Salesforce can withstand the AI test will ultimately require time to verify.

Company Background

ServiceNow is the software company I mentioned at the beginning, which Jensen Huang personally flew to Las Vegas to support for three consecutive years. If Salesforce manages a company’s external customer relationships, ServiceNow manages internal employee workflows and operations. In simple terms, it is the central nervous system of enterprise internal operations.

Many internal business processes that require approval, routing, execution, and documentation can be run on ServiceNow. When a computer breaks down, employees submit IT tickets; new hires go through onboarding workflows to set up accounts and equipment; system outages trigger incident response; and security alerts are assigned, escalated, and resolved. Therefore, ServiceNow is not just an IT ticketing system—it’s more like a unified platform for managing various enterprise workflows.

ServiceNow was founded in 2004 and is headquartered in Santa Clara, California. The current CEO is Bill McDermott, who previously served as SAP’s global CEO and has spent decades in the enterprise software industry. After officially taking the helm at ServiceNow in 2019, McDermott led the company’s expansion from an IT ticketing software provider toward becoming a full-enterprise workflow platform. His distinctive style—exceling at crafting big narratives, closing large deals, and securing enterprise clients—has become an advantage in the AI era.

Product Matrix

Its core founding business is ITSM, which enterprise IT departments use to manage tickets, incident response, change releases, IT assets, and service requests. In the ITSM market, ServiceNow is the undisputed global leader. Building on this, it has expanded into ITOM (IT Operations Management); while ITSM focuses on handling issues after they occur, ITOM proactively monitors systems, detects problems, and aims to resolve them automatically.

Extending the business further, HR Service Delivery enables employees to handle onboarding, offboarding, leave requests, role transfers, and various other employee requests all within ServiceNow. Additionally, Customer Service Management supports enterprise-level customer service—overlapping somewhat with Salesforce Service Cloud but with ServiceNow better suited for complex B2B scenarios such as large equipment, enterprise clients, and cross-departmental service tickets. Security Operations manages security incident response, while Strategic Portfolio Management assists the CIO in overseeing the project portfolio and determining which IT initiatives to fund or terminate.

Putting it all together, ServiceNow has evolved from a simple IT service management software into an enterprise-wide workflow platform. This is the fundamental reason it achieves a 97% renewal rate—once a company migrates its IT, HR, security, and customer service processes onto ServiceNow, replacing it isn’t just a matter of switching software; it requires rebuilding an entire internal operational system, which comes with a very high cost.

Recent key acquisitions

In addition to its native products, ServiceNow has made several critical acquisitions over the past year.

The first acquisition was Moveworks, which develops an AI-powered employee service assistant. Instead of searching for various portals to find answers, employees can simply ask the AI, which can help them check policies, submit tickets, track progress, and even automatically resolve some issues. After the acquisition, Moveworks’ capabilities were integrated into ServiceNow’s EmployeeWorks.

The second is Veza, which specializes in identity governance and access management. In the age of AI agents, determining "who can access what data" has become critically important—not only for humans but also for agents. Veza addresses this exact issue.

The third acquisition is Armis, which specializes in real-time asset visualization for cybersecurity. Armis can see exactly how many devices are on an enterprise network, which ones have vulnerabilities, and which ones are communicating.

All three acquisitions share a common goal: preparing for the large-scale adoption of AI agents in enterprises. For agents to operate effectively within a company, they need to understand what employees are asking, know who has access to which data, and be aware of all assets on the network. These three acquisitions collectively fill these three critical capabilities. Of course, making multiple acquisitions in quick succession also introduces integration risks—particularly with a $7.75 billion deal like Armis, which we will explore in detail when discussing risks.

Core AI Strategy: AI Control Tower

ServiceNow’s core AI strategy is called the AI Control Tower. This concept stems from a very real issue: in the future, enterprises will not rely on just one AI provider—they may use OpenAI’s GPT for customer service, Anthropic’s Claude for contract review, Microsoft’s Copilot for document collaboration, Google’s Gemini for data analysis, and also develop many internal AI agents of their own.

At this point, questions arise: with so many AI agents running simultaneously within an enterprise, who manages them? Who decides what data they can or cannot access? Who ensures they don’t exceed their permissions? And who is held accountable if an incident occurs? This is precisely what the AI Control Tower is designed to address.

ServiceNow isn’t trying to recreate ChatGPT; instead, it aims to serve as the governance, orchestration, and execution layer for enterprise-grade AI agents, ensuring these AI systems operate within companies securely, compliantly, and auditably. This is what sets it apart from many other SaaS companies. While many firms are asking, “Can we build our own AI agent to compete with ChatGPT, Claude, or Gemini for application-layer access?” ServiceNow has taken a smarter approach: “Rather than competing with you for the underlying models, we’ll manage the execution workflows once those models enter the enterprise.”

ServiceNow aims to ensure that regardless of which AI platform a company uses, any AI that integrates into the company’s workflows, accesses its systems, or executes its tasks must be governed and orchestrated through ServiceNow.

Why ServiceNow?

This stems from ServiceNow’s two decades of accumulated foundational capabilities. It possesses something called a CMDB (Configuration Management Database)—essentially, a complete map of an enterprise’s IT assets and their interrelationships. It records which servers are in use, which applications are running, and the permission relationships between users. It also features a proven workflow engine that has been in operation for over a decade, powering all approval, execution, and collaboration processes within the enterprise. Additionally, it maintains comprehensive audit logs that track every action—who performed it, when, and what changes were made.

After an AI agent enters an enterprise, it most needs three things: awareness of which systems can be accessed, adherence to established workflows, and an audit trail for every step the agent takes. In addition, ServiceNow complements this by integrating Veza for identity and access validation, and Armis for real-time asset visualization.

At this year’s Knowledge conference, this concept took another step forward with ServiceNow’s launch of Action Fabric. This platform enables any third-party AI agent—whether Claude, GPT, Gemini, or Copilot—to invoke ServiceNow’s governance engine to execute enterprise-grade tasks. “I don’t care which AI model you use, but execution and governance must pass through my layer”—this logic mirrors Apple’s iOS approach, where Apple doesn’t build every app itself, but all apps run on iOS. ServiceNow aims to follow the same path.

Jensen Huang endorses

The most compelling endorsement for this positioning comes from Jensen Huang. NVIDIA’s CEO attended ServiceNow’s annual conference for the third consecutive year—not merely as a partner showing support, but as a customer of ServiceNow themselves. NVIDIA’s internal supercomputing quotation system runs on ServiceNow; previously, generating a complete quotation document took five days, but with the AI workflow, it now takes just five minutes.

Jensen Huang’s exact words were: “ServiceNow is essentially the enterprise operating system of the AI era.” This year, the two companies jointly launched Project Arc, with NVIDIA providing a secure AI computing sandbox and ServiceNow delivering enterprise-grade governance—demonstrating a deeply integrated partnership. This illustrates that ServiceNow’s AI Control Tower is not an isolated software concept; it is now being incorporated into the enterprise adoption narratives of AI ecosystem partners such as NVIDIA, OpenAI, Google, and Anthropic.

Latest financial data

In the first quarter of this year, total revenue was $3.77 billion, a 22% year-over-year increase; subscription revenue was $3.671 billion, also up 22% year-over-year and exceeding the upper end of guidance; total RPO was $27.7 billion, up 25% year-over-year; and customer retention rate was 97%. These figures demonstrate that ServiceNow’s fundamentals remain strong—it is still a software platform delivering approximately 20% growth, a 97% retention rate, high profitability, and strong cash flow.

AI performance has been particularly strong. The company raised its annual contract value (ACV) target for AI-related business this year from $1 billion at the beginning of the year to $1.5 billion. Note that this is measured by contract value, not current revenue, and will gradually convert into actual realized revenue over time. However, raising the target by 50% within a single quarter indicates that its AI products are genuinely attracting customer demand and experiencing rapid growth.

Its stock price has declined more than 50% from its all-time high, and its forward P/E ratio is now approximately in the range of 21–24x. For a high-growth, asset-light software company, this is indeed a relatively undervalued range.

Catalysts and Risks

The reasons to be bullish on ServiceNow are clear. First, its AI narrative is highly coherent: the AI Control Tower serves as the operating system for enterprises in the AI era— as AI demand grows, companies increasingly need platforms for governance, auditing, access control, and execution. Second, its new AI business is genuinely scaling, with AI ACV rising from $1 billion to $1.5 billion, validating the story in real terms. Third, its ecosystem partnerships are strong, with OpenAI, Google Gemini, Claude, and NVIDIA all integrating or forming deep collaborations with ServiceNow, reinforcing its strategic position as the “enterprise AI control tower.”

However, the risks facing ServiceNow must also be clearly stated. After the latest quarterly earnings report, despite exceeding market expectations, the stock still dropped by double digits in after-hours trading, reflecting extremely pessimistic market sentiment and indicating that the current market trend has not yet reversed—investors remain skeptical about SaaS companies’ business models and their AI transformation. Additionally, ServiceNow has recently completed three acquisitions in quick succession, particularly the $7.75 billion acquisition of Armis, which will take time to integrate. The market will closely examine how much of the raised revenue guidance stems from acquisitions versus organic growth. External risks include geopolitical factors in the Middle East, which caused delays in several large projects last quarter, negatively impacting subscription revenue growth by approximately 75 basis points.

I remain quite optimistic about ServiceNow. Among the three companies, it has the most coherent, straightforward, and market-appealing AI narrative. Its AI Control Tower positioning is not only immune to AI disruption but stands to benefit significantly from AI’s widespread adoption, making it a strong candidate to become the most critical software platform for enterprise AI implementation. From a valuation perspective, its stock has declined by half from its peak over the past year, resulting in a low forward P/E ratio—similar to Salesforce—placing it at a relatively attractive level. Currently, the value proposition and risk-reward profile for buying are very compelling.

Company Background

At its core, this company is the ultimate data warehouse for enterprise data. While Salesforce manages customers and ServiceNow manages processes, Snowflake manages data—all of it. Sales data, user behavior, financial reports, system logs—you pour it all into Snowflake, then perform analytics, modeling, and run AI workloads on this powerful data warehouse.

Product Matrix

At its core, Snowflake is still built on a data warehouse and data lake, where enterprises ingest both structured and semi-structured data to run SQL queries and perform data analytics—this is the foundation of Snowflake and the primary source of its revenue. Built on top of this foundation is Snowpark, which enables developers to write code directly within Snowflake using Python, Java, and Scala to build data pipelines and machine learning models, without needing to move data outside the platform—completing the entire process from data processing to model training internally.

Moving up further, we have Snowflake’s Cortex AI suite, the key focus of Snowflake over the past year. It includes two core products. Snowflake Intelligence is designed for business users, enabling them to interact with data using natural language. It automatically queries, analyzes, and generates insights from both structured and unstructured data within Snowflake, and can proactively execute multi-step tasks—functioning more like an enterprise-grade AI agent. Cortex Code is tailored for developers and differs from general-purpose coding assistants by being a native Snowflake AI coding agent that understands Snowflake’s data structures, permission settings, and computing environment. It can directly help you build data pipelines, debug queries, and develop AI applications, offering powerful capabilities.

The roles of these two products are clearly defined: Snowflake Intelligence is designed for business users, enabling those without SQL knowledge to directly ask questions of data, use data, and let AI take action based on data; Cortex Code is tailored for technical teams, allowing developers and data engineers to build data applications, data pipelines, and AI applications more quickly.

In addition to AI products, Snowflake has two other distinctive capabilities. The Snowflake Marketplace is a data sharing and trading platform where enterprises can directly buy and sell datasets, as well as access third-party data for analysis. Data Clean Rooms enable cross-organizational data collaboration with privacy protection, allowing two companies to perform joint analysis without exposing their raw data. This is used in advertising for cross-platform attribution, in pharmaceuticals for collaborative clinical trials, and in finance for fraud prevention collaboration. These two capabilities represent hard-to-replicate competitive advantages.

Taken together, Snowflake is transitioning from a data warehousing tool to an AI data platform, with data storage and computation at the base, development tools and AI engines in the middle, and intelligent assistants and data marketplaces for business users on top. Snowflake aims to do more than just help enterprises store and query data—it wants to enable businesses to analyze, share, and build applications on a single governed data platform, while truly integrating AI into their business data. In terms of customer scale, Snowflake currently serves over 13,300 enterprise customers, with the platform processing 6.3 billion data queries daily.

Business model

This is the key difference between Snowflake and the previous two companies. Salesforce and ServiceNow charge based on seats, with a fixed annual subscription fee; Snowflake is completely different—it charges based on actual usage, so you pay according to the platform’s formula for the number of queries run, the amount of computing power used, and the volume of data stored.

This model has its pros and cons. On the positive side, data consumption by enterprises is growing exponentially in the AI era, as every AI task requires computational power and data queries, causing Snowflake’s revenue to naturally rise alongside the surge in AI workloads. On the downside, if companies cut budgets or optimize their workloads, Snowflake’s revenue will also decline.

However, over the past two years, Snowflake has also begun actively promoting multi-year consumption commitment contracts. Its latest financial report shows an RPO of $9.77 billion, a 42% year-over-year increase, indicating that major customers are now locking in their computing budgets with Snowflake for several years into the future, rather than maintaining a purely transactional relationship.

Moat and Competitive Landscape

Its strength lies in data stickiness. Once data is loaded into Snowflake, all downstream and upstream analytical models, query scripts, and data pipelines are built on top of it, making migration extremely costly. This is Snowflake’s core competitive advantage. Additionally, its Data Clean Rooms are highly mature in terms of privacy protection and cross-organizational collaboration, making them difficult to replicate.

The weakness lies in the highly competitive landscape. The biggest competitor is Databricks, whose latest annualized revenue run rate has reached $5.4 billion, growing at 65% year-over-year—more than double Snowflake’s 29% growth rate—and its most recent valuation has surpassed $100 billion. Databricks is stronger in machine learning and AI workloads. If Databricks goes public in the future, it is likely to become one of the most closely watched IPOs in the enterprise software market, at which point Snowflake will have to face direct comparisons in the public markets.

Beyond Databricks, the major cloud providers also pose significant threats. AWS’s Redshift, Google’s BigQuery, and Azure’s Synapse are all continuously evolving and are naturally integrated with their respective cloud ecosystems—they are both partners and potential alternatives to Snowflake. Further down the ladder, open-source and emerging tools like DuckDB and ClickHouse are eroding market share in specific use cases such as local analytics, real-time analytics, and low-cost querying. As a result, Snowflake’s competitive landscape is more complex than that of Salesforce or ServiceNow.

The counterintuitive risks of consumption patterns

Here’s another counterintuitive point: Snowflake fears not that customers won’t use it, but that they’ll use it too well. Since Snowflake operates on a consumption-based model, the more customers query, compute, and store, the higher Snowflake’s revenue becomes. Conversely, when companies notice their Snowflake bills are too high, they push their engineering teams to optimize queries, reduce storage usage, or even replace lower-value tasks with open-source alternatives.

This is the double-edged sword of consumption patterns: when growth is rapid, revenue naturally rises with customer usage; but once customers begin optimizing their usage, revenue growth also slows. This trend is already reflected in the data: Snowflake’s net revenue retention rate has declined from 131% to 126%, and now to the latest figure of 125%. While this number remains healthy, indicating that existing customers are still increasing their spending annually, the downward trend suggests that expansion among existing customers has slowed compared to before. This is due both to the natural deceleration that comes with a larger base and to customers optimizing costs and slowing their spending pace.

Therefore, Snowflake is more like a high-growth, highly scalable AI data platform with intense competition. This is both Snowflake’s greatest appeal and its biggest risk.

Latest financial data

Annual product revenue reached $4.47 billion, a 29% year-over-year increase—the fastest growth rate among the three companies. Product revenue for the latest quarter was $1.23 billion, up 30% year-over-year, slightly higher than the annual growth rate. RPO stood at $9.77 billion, up 42% year-over-year. The company added 740 net new customers in the latest quarter, a 40% year-over-year increase. Additionally, the company signed its largest single contract in history, worth over $400 million. These figures demonstrate that demand for Snowflake has not slowed; on the contrary, enterprise customers continue to sign larger long-term contracts.

However, the issues are also clear. Under GAAP, Snowflake still incurred a loss of approximately $1.33 billion for the full year, making it the only one of the three companies not yet profitable under GAAP. Stock-based compensation amounted to over $400 million per quarter, totaling more than $1.7 billion for the year, placing significant pressure on shareholder dilution.

However, Snowflake remains the most expensive of the three companies, with an EV/Sales multiple of approximately 9x based on future revenue, significantly higher than Salesforce's.

Catalysts and Risks

On the positive side, Snowflake has several key highlights. First, Snowflake does not operate on a traditional SaaS model but rather on a usage-based model, which naturally benefits from the growth of AI workloads. In the short term, the more AI is run, the more Snowflake earns. Second, RPO increased by 42% year-over-year, indicating that large customers are still signing larger long-term contracts, signaling strong future revenue visibility. Third, Snowflake Intelligence and Cortex Code are rapidly expanding, with over 9,100 accounts already using AI features.

In addition, Snowflake has two other significant events coming up: its earnings report on May 27, followed by the Snowflake Annual Conference in San Francisco from June 1 to 4. With these two catalysts occurring back-to-back, I believe the upside potential outweighs the downside. Around that time, the stock price is likely to experience significant volatility.

We must also understand the risks in advance. First, GAAP persistent losses are the biggest weakness. In a market environment that favors profitability and cash flow, Snowflake will face greater valuation pressure compared to Salesforce and ServiceNow. Second, Databricks is currently Snowflake’s strongest competitor; its future IPO could reshape the competitive landscape of the data platform sector. If Databricks achieves faster growth, a stronger AI narrative, and a more attractive valuation after going public, capital may shift from Snowflake to Databricks. Additionally, shareholder lawsuits and insider sell-offs—noise at the corporate governance level—can also impact market sentiment, though they are not the primary focus at this time.

Snowflake can be summed up as the fastest-growing, most directly aligned with AI data infrastructure, and naturally insulated from traditional SaaS business models—yet also the most highly valued, most competitively pressured, and least profitable of the three, offering high upside with high risk.

After analyzing these three companies, let me share my personal subjective opinion.

If you prioritize a margin of safety and favor value investing principles, Salesforce is relatively the most stable option, with a forward P/E ratio in the teens, $14.4 billion in free cash flow, a $50 billion share repurchase authorization, and consistent profitability—offering a substantial margin of safety for building and holding a position. However, its growth rate is only 10%, so its potential for strong price appreciation may be limited.

If you accept the logic behind the AI Control Tower governance layer, ServiceNow may have the clearest AI narrative among the three companies, with growth exceeding 20%, a 97% renewal rate, a forward P/E ratio of 22x, and backing from Jensen Huang’s consistent three-year endorsement. The current valuation offers strong upside potential—but前提是 you are willing to accept the integration risks from aggressive acquisitions and the short-term volatility in its stock price.

If you're seeking maximum upside and can tolerate high volatility, Snowflake is a high-reward bet. The primary risks are that the company fails to achieve profitability, continues to incur losses, and sees a decline in its net revenue retention rate. Additionally, if competitor Databricks proceeds with an IPO in the future, it could reset the valuation benchmarks for the entire data platform sector. The risk and volatility are indeed substantial.

Beyond these three, if you're looking for the most stable anchor in the software sector, Microsoft remains the top choice—it’s one of the most severely undervalued large-cap software stocks in this cycle. However, I still want to emphasize that this is merely my personal framework for evaluation and does not constitute any investment advice. Everyone should make their own investment decisions based on their actual portfolio positions and after rational analysis.

Finally, we return to the initial question: Is AI going to kill the entire software industry, or is it offering us a once-in-a-decade opportunity to buy the dip?

My assessment is that the narrative of AI killing software is overly simplified. What’s actually happening is that AI is displacing software that merely sells feature interfaces, while simultaneously rewarding platforms that sell infrastructure and governance. Not all software will be disrupted.

This is similar to the dot-com bubble burst in 2000, when the dominant market narrative was that “the internet would kill all traditional companies.” In the end, it wasn’t just internet companies that survived—but those traditional companies that embraced the internet early and integrated these digital tools into their core operations, successfully transitioning into the internet era. Twenty years later, the same logic applies to today’s AI wave. The software companies with real moats, substantial data accumulation, and the ability to serve as AI infrastructure platforms will ultimately emerge as the biggest winners. And right now, they may just be at the starting point of a new upward cycle.