AI will not lift all software stocks—the winners own the workflow, not the seat
AI is not a rising tide for software: platforms embedded in critical workflows and usage-linked infrastructure have a clearer path to monetization than undifferentiated seat-based applications. The latest earnings reactions show investors are already separating those models, making software selection—not software exposure—the trade.
The contrarian view is becoming the obvious one: AI will not lift all software stocks. The market is moving past the lazy assumption that every application vendor benefits when companies deploy more agents, because agents can also reduce the number of seats customers need. The winners will own the workflow, the data layer, or the infrastructure that agents must use; the losers will sell access to a screen whose value is increasingly delivered by an automated worker. Datadog and HubSpot falling sharply after solid results, while investors continue to reward selected platform and infrastructure names, is the early evidence of that split.
The Aug. 16 debate over whether AI agents expand software demand or undermine the SaaS subscription model is no longer theoretical. On Aug. 6, Datadog and HubSpot fell more than 15% after earnings, with Salesforce and ServiceNow also slipping in sympathy. That reaction matters because neither company suddenly stopped growing; rather, investors questioned where future software value will accrue and whether the existing pricing model can capture it. The market is not rejecting software. It is rejecting the idea that software is one homogeneous AI beneficiary group.
Datadog shows why even usage-based infrastructure is not automatically insulated. The company reported Q2 2026 revenue of $1.12 billion, up 35.6% year over year, and adjusted earnings per share of $0.65 against roughly $0.59 expected. Yet management said its largest customer would reduce usage beginning in the third quarter, implying growth of roughly 28% to 29% ahead. That is a healthy growth profile, but the stock still carries a 21.14 price-to-sales multiple in the supplied market data. When revenue depends on consumption, AI can increase workloads—but a single optimization cycle, customer pullback, or shift in usage can also expose the operating leverage in reverse. Usage is a better alignment than a fixed seat in some cases, but it is not the same as pricing power.
ServiceNow is the cleaner expression of the workflow thesis. In Q2 2026, subscription revenue rose 24.5% year over year to $3.877 billion, while current remaining performance obligations rose 21% to $13.20 billion. More important than either figure, ServiceNow said AI had crossed $1 billion in annual contract value. That is the distinction investors should focus on: AI is not merely an assistant added to a product menu; it is being attached to a platform that already sits inside IT, employee, and enterprise operating workflows. When the software owns the process through which work is approved, routed, and measured, it has a stronger chance of charging for outcomes or expanding the platform than an application that simply charges for another login.
The contrast with seat-based SaaS is visible in the valuation and performance dispersion. HubSpot has grown revenue 19.2%, yet its shares are down 37.2% year to date and the stock trades at a 3.56 price-to-sales multiple. Salesforce is a more established counterexample to the bear case, with a 4.00 price-to-sales multiple, 18.7% net margin, and 21.9% EPS growth in the supplied data. But Salesforce also illustrates why scale and workflow depth matter: its AI products can be sold into an installed customer base rather than requiring a new seat category to be invented from scratch. The market is asking whether AI will expand revenue per customer, not simply whether a vendor can add an AI feature.
Snowflake and CrowdStrike sit on the infrastructure side of the barbell, though neither is a valuation free pass. Snowflake’s product revenue grew 32% year over year, it reported 654 customers generating more than $1 million in trailing 12-month revenue, and its Dynamic Model Routing launch explicitly targeted better AI economics. That is a data and usage-layer proposition, not a conventional seat sale. CrowdStrike’s Falcon platform spans 33 cloud modules across identity, cloud, SIEM, and managed services, giving AI more surfaces on which to expand security operations. But the market data still assigns Snowflake a 22.92 price-to-sales multiple and CrowdStrike a 38.37 multiple. Platform ownership can improve monetization durability; it cannot erase the need for growth, margins, or reasonable expectations.
Yes, the AI bulls have a credible answer. Agents can increase software demand by creating more implementation work, more data consumption, and more automated activity across existing systems. Salesforce has said its Agentforce and Slack products have delivered 2.4 billion Agentic Work Units, while more than 40% of Data Cloud and Agentforce bookings came from existing-customer expansion in the prior fiscal year. That is evidence that AI can expand budgets rather than merely cannibalize seats. But the comparison still misses the key issue: additional activity accrues disproportionately to the vendor that controls the workflow, data, or execution layer. More agent actions do not guarantee more value for every application vendor if the agent can complete the task with fewer paid users.
This is also why the current selloff should not be dismissed as simple multiple compression. Valuation clearly matters: investors can pay for growth only when they believe the growth will persist and convert into cash generation. Yet the post-earnings reactions suggest a deeper model question. A seat-based company with AI features must prove that automation raises revenue per account faster than it reduces the need for seats. A workflow platform must prove that AI increases module adoption, contract value, or control over a mission-critical process. An infrastructure vendor must prove that usage growth is broad enough to offset customer concentration and optimization risk. Those are different tests, and the market is beginning to price them differently.
The cloud transition offers the useful historical parallel, but not the comforting conclusion that all software wins. The durable winners owned recurring workflows or a usage layer that customers could not easily bypass; generic applications faced commoditization as the stack changed. AI accelerates that sorting process because agents make the seat itself less sacred. We therefore prefer to judge each company by what it owns when the agent is doing the work: the system of record, the system of action, the proprietary data, or the infrastructure bill. A polished interface without one of those anchors is a weaker AI asset than its software label suggests.
The bottom line is a barbell, not a blanket software allocation. ServiceNow’s AI annual contract value and workflow position show what monetization can look like, while Datadog’s customer-usage warning shows that even attractive infrastructure growth can be fragile. Snowflake and CrowdStrike offer platform exposure, but their rich price-to-sales multiples demand execution; Salesforce offers a lower-multiple, scaled test of whether an installed workflow can turn AI activity into expansion.
We would change our mind if seat-based vendors consistently demonstrated that AI raises revenue per customer without relying on more seats, or if usage-based platforms showed that concentration and optimization risks were immaterial. Until then, investors should stop asking which software stocks have AI and ask which ones still own the economic workflow after AI does the work.
Our take, not advice. This is opinion commentary — informational only, not personalized investment recommendations. Markets carry risk. Do your own research and consider your own situation before any trade.
▌The Daily Briefing · Free
A new stock idea, every evening.
One stock worth watching each weekday, plus the analysis behind it. Free, in your inbox.
