Salesforce is cheap potentially. The share price is down 38.3% year to date, and down 55% since its all time high in December 2024. Essentially, the share price of Salesforce today ($163) is flat since almost 8 years (September 2018).In fiscal year 2018, Salesforce generated $10.5 billion in revenue. In the current financial year, Salesforce is expected to generate $46.1 billion in revenue. Revenue is up 4.4x, but the share price is flat. Perhaps Salesforce was expensive in 2018, but it certainly is not today. The chart below shows that Salesforce trades currently at only 8.62x forward CFPS, lower than at the bottom of the global financial crisis in 2009, and way below its long term average of ~30x.This raises the question if Salesforce is an AI loser?At first sight, Salesforce does indeed look like the archetypal AI loser. Its core products are expensive, often cumbersome, sold mainly per human seat, and surrounded by an implementation industry that generative AI can compress. An agent can update a record without opening Sales Cloud, answer a question without viewing a Tableau dashboard and resolve a case without a human Service Cloud user. So, the AI fear is legitimate.However, Salesforce is not behaving like a business in terminal decline. Revenue is still growing, margins have expanded materially, and the company generates $14.4 billion last year, guided to roughly $15 billion. Management has also launched the largest accelerated share repurchase in corporate history.The market is therefore pricing in a meaningful deterioration before it has clearly appeared in the financial results. The crucial question is whether AI destroys Salesforce’s economics—or merely changes where within the company those economics accrue.In this article, I assess Salesforce product by product to determine which businesses are likely AI winners, which are losers, and which face a battle between consumption growth and seat cannibalization. I then reconstruct owner earnings, separating maintenance capex from growth investment and treating stock-based compensation as a real cost. Finally, I project where the company’s cash will go (acquisitions, dividends, debt repayment or further buybacks) and provide bear, base and bull case return scenarios.The result is a more complicated thesis than “cheap software stock.” Salesforce may be one of the clearest examples of a company whose interface is threatened while its underlying data, permissions and workflow infrastructure becomes more valuable.So, is Salesforce a melting ice cube—or is the market offering a durable AI infrastructure business at roughly 13 times owner earnings?1. Salesforce can lose the interface — and still winThe fear that Salesforce is an AI loser is legitimate, but it is incomplete.AI agents may dramatically reduce the need to open Salesforce applications. They can update opportunities without visiting Sales Cloud, resolve customer cases without a human Service Cloud user and answer questions without opening a Tableau dashboard.However, the disappearance of the interface does not necessarily mean the disappearance of Salesforce. An enterprise agent still needs permissioned customer data, identity, governance, workflow logic, API access and an auditable system in which to record its actions. In many large organizations, Salesforce already owns those layers.Salesforce’s own usage data suggests that this transition is beginning. Customers executed 1.6 billion Agentic Work Units in Q1 FY27, more than twice the previous quarter. Salesforce defines an AWU as a discrete task completed by an agent—for example, making a decision, updating a record or triggering a workflow. Live customer agents have now delivered 3.8 billion AWUs, while AWUs inside Slack grew approximately 350% quarter over quarter.Source: Salesforce Q1 FY27 investor presentationSalesforce is attempting to position itself across the entire agentic stack:Slack can become the conversational layer where people and agents coordinate work.Agentforce provides the agents and orchestration.Data 360 and Informatica provide trusted context, identity and governance.MuleSoft connects agents to applications and exposes permissioned actions.Customer 360 remains the system of record where actions are executed and documented.Salesforce’s Headless 360 strategy makes this particularly interesting. Customers do not necessarily need to interact with Salesforce through a Salesforce interface. They can use Slackbot, Claude, ChatGPT, Cursor or another front end while Salesforce continues to provide the data, permissions, workflows and write-back layer underneath.In other words, Salesforce is preparing for a world in which customers use Salesforce constantly without regularly opening Salesforce.1.1 Where will the economics accrue?The debate is often framed too simply: Will Agentforce succeed?The more important question is where value accrues as enterprise software shifts from human navigation to machine execution.Traditional SaaS monetizes access. A company employs 1,000 people, purchases 1,000 seats and those employees click through a graphical interface.Agentic software monetizes work. A much smaller human team may instruct agents to perform thousands—or millions—of actions across numerous systems.That threatens software companies whose value resides primarily in their interface or human seat count. But it may benefit companies that control proprietary data, permissions, business logic and the right to execute actions.Salesforce spans both sides of this divide. Its browser interfaces and seat-based licenses are exposed. Its customer data, metadata, automation and workflows could become increasingly valuable toll roads.I see three broad scenarios.1. The backend winnerSalesforce remains the trusted system of record and action. External agents call its APIs, use Data 360 for context and write their actions back into Customer 360. The Salesforce interface becomes less important, but consumption, data usage and premium editions more than offset declining human seats.2. The balanced transitionSales and Service seats grow slowly or decline modestly. Slack, Data 360, Industry Clouds and Agentforce consumption compensate for the pressure. Revenue continues to compound in the high single digits while margins gradually expand.3. The commoditized databaseMicrosoft, OpenAI, Anthropic, ServiceNow or AI-native competitors own the user relationship and orchestration layer. Salesforce remains an expensive database behind the scenes. Customers simplify their software stacks, and consumption revenue fails to compensate for declining seats and weaker pricing.The share price appears to discount the third scenario. The investment case depends on whether Salesforce can remain the trusted data and action layer.2. What the reported numbers already saySalesforce produced $41.5 billion of FY26 revenue, up 10%, and $14.4 billion of reported free cash flow. Subscription and support revenue was $39.4 billion. The old product presentation showed Sales at $9.0 billion, Service at $9.8 billion, Platform/Slack/Other at $8.9 billion, Marketing/Commerce at $5.4 billion and Integration/Analytics at $6.2 billion.The Q1 FY27 results looked stronger at first glance: revenue rose 13% to $11.1 billion, subscription revenue rose 14%, current RPO rose 14%, and Agentforce ARR reached $1.2 billion, up 205%. Yet $444 million of quarterly revenue came from Informatica. Excluding that acquisition, total revenue grew about 8.7%; subscription revenue excluding Informatica grew about 9.3%Salesforce also changed its product disclosure in Q1. Instead of five service offerings, it now reports only two:The new presentation is strategically coherent but analytically less useful. It hides the standalone trajectories of Slack, Tableau, MuleSoft, Marketing and Commerce precisely when investors need to distinguish AI winners from losers. The 25% growth in the second bucket is not evidence that the whole data/platform portfolio is growing at 25%; most of the acceleration came from acquired Informatica revenue.There are nevertheless real positive signals. Agentforce and Data 360 ARR reached nearly $3.4 billion, including $1.1 billion of Informatica cloud ARR and $1.2 billion of Agentforce ARR. Premium Agentforce One and Agentforce-for-app bookings grew nearly 60%. Sales and Service still recorded positive seat growth. The ten customers with the greatest Q1 Agentic Work Unit usage increased their total Salesforce spending by an average of 1.5 times over the preceding year.The contrary evidence is equally important. Management explicitly cited continuing weakness in Marketing and Commerce and increasing softness in Tableau bookings and renewals. 3. The AI winners and losers inside the portfolioSalesforce is a collection of products with very different exposure to AI.Treating the entire company as either an “AI winner” or an “AI loser” misses the central tension. AI can strengthen Salesforce’s data and workflow infrastructure while simultaneously weakening its interfaces, human-seat model and implementation ecosystem.I assess each business using four questions:Does AI increase demand for the product’s underlying function?Does Salesforce own proprietary data, permissions or workflows that an external agent still needs?Can Salesforce monetize agent activity sufficiently to replace any lost seat revenue?Does AI lower the product’s barriers to entry or make it easier to bypass?On that basis, the portfolio divides into three broad groups.The infrastructure winners are Slack, Data 360, Informatica, the Salesforce Platform and Industry Clouds. These products provide context, governance, coordination and the ability to execute actions—functions that become more important as agents perform more enterprise work.The battlegrounds are Service Cloud, Sales Cloud and MuleSoft. AI creates obvious new use cases, but it also threatens existing seats, pricing or implementation economics. These businesses can win, although the path from rising agent usage to rising owner earnings is not automatic.The exposed products are Tableau, Marketing Cloud, Commerce Cloud and professional services. AI can improve each product, but it can also commoditize or bypass a meaningful part of what customers currently pay for.The most promising place to begin is also perhaps the most misunderstood acquisition in Salesforce’s history: Slack.Below: the product-by-product scorecard, reconstructed owner earnings, and why my base case compounds above 25% a year.