Salesforce’s latest earnings did more than rescue a stock that had spent much of the past year under pressure. They reopened a much bigger market question: what if artificial intelligence does not hollow out enterprise software, but instead makes the strongest software platforms more valuable?
That possibility cuts directly against one of the most persistent trades of the 2025-2026 AI cycle. Software was treated as the vulnerable layer. Semiconductors owned the scarcity. Data-center infrastructure owned the spending boom. SaaS companies, meanwhile, were discounted on the fear that AI agents would bypass traditional interfaces, reduce seat counts and weaken the economics of subscriptions.
The market priced that threat aggressively. The S&P 500 Software & Services group fell more than 33% from its October 2025 peak to the April 2026 low, according to the material supplied for this analysis. The relative chart between the iShares Expanded Tech-Software Sector ETF, IGV, and the semiconductor ETF SOXX showed software underperforming chips for more than 374 days before the ratio began turning higher.
Salesforce has now put a fundamental catalyst behind that technical change.
Its fiscal second-quarter results showed revenue of $11.3 billion, up 11% year over year, subscription and support revenue of $10.8 billion, up 12%, and current remaining performance obligations of $33.5 billion, up 14%. Free cash flow rose 81% to $1.1 billion. More importantly for the AI debate, Agentforce and Data 360 annual recurring revenue reached nearly $3.9 billion, up more than 210% year over year, while Agentforce ARR itself exceeded $1.5 billion.
That is the number I would keep coming back to.
AI is no longer just a product demo sitting beside the core Salesforce business. It is becoming a reported revenue stream, a usage stream and, increasingly, a reason customers stay inside the platform.
Data: company filings and investor-relations releases. Editorial chart.
Salesforce just gave the market a different answer
For much of the past two years, the bear case on enterprise software has been straightforward. If AI agents can interact directly with databases, workflows and internal systems, why should enterprises keep paying for layers of application software and thousands of individual user seats?
The answer from Salesforce is that the application layer may become more valuable precisely because enterprises need somewhere to govern what agents are allowed to see and do.
The company’s new Claudeforce partnership with Anthropic makes that point unusually clear. Salesforce in Claude launches with 37 prebuilt sales skills that allow Claude to reason over live Salesforce information, prepare sellers for meetings, review pipelines and take governed actions. The important part is not the number of skills. It is the architecture underneath them.
Claude provides the reasoning layer. Salesforce provides enterprise data, permissions, business rules, workflows and transaction history.
That flips the old disruption story around.
Instead of AI bypassing Salesforce, Salesforce is trying to become the trusted control plane through which AI acts. The user interface may change. The economic moat may not.
I think that distinction matters more than the chatbot demos that dominated the first phase of the generative-AI trade. Enterprises do not merely need models that can generate text. They need models that understand which customer records an employee may access, which discount limits a salesperson can approve, which transactions require escalation and which actions must leave an auditable trail.
That is software territory.
And it helps explain why the strongest incumbents may have an advantage over smaller AI-native challengers. The incumbent already owns the permissions, the workflow history, the customer data model and the integration surface.
Data: company filings and investor-relations releases. Editorial chart.
The real change is that AI is showing up in the numbers
The reason Salesforce’s quarter matters is not that management said “AI” more often. Every software company says AI now.
The difference is that the commercial evidence is becoming measurable.
Agentforce and Data 360 ARR increased from about $3.4 billion in the prior quarter to nearly $3.9 billion in Q2. Agentforce ARR moved from more than $1 billion to above $1.5 billion. Salesforce also reported 3.2 billion Agentic Work Units in the quarter, up 97% sequentially, while Data 360 ingested 104 trillion records, up 355% from a year earlier.
Those metrics are not conventional SaaS seat counts. That is exactly the point.
Software economics are beginning to move from a simple “users times seats” model toward consumption, data processing, agent activity and premium workflow packages. If that transition works, AI does not merely protect existing revenue. It can create an entirely new billing layer.
What I see in Salesforce’s numbers is the early shape of that shift.
There is still plenty to prove. Agentforce remains small relative to Salesforce’s total revenue base. Some of the company’s reported growth also includes the contribution from Informatica, so investors should not confuse acquisition-driven revenue with organic acceleration. But the direction is much harder to dismiss than it was six months ago.
The company also raised full-year FY27 revenue guidance to $46.1 billion-$46.4 billion from $45.9 billion-$46.2 billion. The midpoint increased by $200 million.
That is not a dramatic reset on its own. In the context of a stock that had been priced for stagnation and AI disruption, it matters.
Data: company filings and investor-relations releases. Editorial chart.
The bigger signal is the basket, not one stock
Salesforce can squeeze 20% or more after earnings and still fail to produce a durable software rotation.
That is why the IGV/SOXX ratio in the supplied analysis is more useful than CRM alone.
IGV holds 106 stocks and includes large positions in companies such as Palantir, Microsoft, Palo Alto Networks, CrowdStrike, Salesforce, Oracle, ServiceNow and Adobe. As of August 26, the fund traded at roughly 37 times earnings. That is not cheap, especially if long-term Treasury yields move higher again.
The relative chart in the source document is therefore doing more work than a normal price chart. Software underperformed semiconductors for more than a year, then reclaimed its 50-EMA band in late July. If that ratio continues rising, it means capital is rotating toward software even when the absolute market is choppy.
That is a regime signal.
The important phrase is “if it continues.”
One earnings-driven squeeze can look like a structural turn for several days and then completely unwind. A genuine rotation needs breadth. Salesforce cannot carry it alone. CrowdStrike, ServiceNow, Workday, Adobe, Okta and other large software names need to keep participating.
This is where I would resist the temptation to call a bottom based on one spectacular CRM session. The better claim is narrower: Salesforce has given the rotation thesis a credible fundamental catalyst. The market still has to confirm it.
Peer results are making the AI monetization case harder to ignore
The strongest supporting evidence is coming from other enterprise-software companies.
CrowdStrike reported fiscal Q2 revenue of $1.47 billion, up 26%, while net-new ARR reached a record $333 million, up 51% year over year. The security angle is particularly important because every new AI agent expands the number of identities, permissions and machine actions that enterprises must monitor.
Workday gave an even more direct signal. Management said AI drove more than 25% of new annual contract value in Q2, and more than 5,500 customers are now using at least one of its organic agents. That customer count increased by more than 35% from the previous quarter.
The common thread is not “AI software” as a marketing category.
It is enterprise control.
Salesforce controls customer workflows and revenue data. CrowdStrike controls endpoint, cloud and identity security. Workday controls finance and human-capital workflows. Okta controls identity. ServiceNow controls enterprise service workflows. Adobe controls creative production and document ecosystems.
These companies sit at choke points inside large organizations.
AI agents need access to those choke points if they are going to do useful work.
This is why the original SaaS-pocalypse thesis may have been too broad. AI can absolutely destroy weak software products whose only advantage is a thin interface over commodity data. But platforms that own proprietary workflow context, permissions and system-of-record data may become the rails that AI agents run on.
That is a very different outcome.
Data: company filings and investor-relations releases. Editorial chart.
The seat-count bear case is not dead
There is still a serious counterargument.
If AI agents become dramatically more capable, enterprises may need fewer human users touching traditional software screens. A sales organization with 10,000 CRM seats could eventually operate with fewer people doing more work through agents. That creates pressure on conventional per-seat pricing.
Salesforce seems to understand this. The company has been introducing new editions, consumption measures and agentic-work units that move monetization away from pure seat growth.
But transitions like this are rarely clean.
Customers may resist paying both for traditional subscriptions and for new AI consumption layers. Procurement departments will demand evidence that AI reduces labor costs or increases revenue before accepting higher software bills. The more vendors add agent fees, data fees and premium AI packages, the more likely customers are to rationalize their software stacks.
That means the winners may consolidate spend rather than simply expand it.
For Salesforce, that could still be favorable. A large enterprise may decide to pay more to one trusted platform while cutting smaller point solutions. But the industry-level implication is less bullish than the company-level implication.
AI can expand the value of dominant platforms and still shrink the total number of viable software vendors.
That nuance matters.
Valuation is the next obstacle
The technical setup has improved, but software is no longer obviously cheap.
IGV’s official portfolio data showed a price-to-earnings ratio just above 37 on August 26. That is a demanding multiple for a group that remains sensitive to bond yields.
The entire rotation therefore has a macro dependency.
If long-term Treasury yields rise sharply, software’s improving earnings story may be overwhelmed by valuation compression. The companies most exposed would be those where free cash flow remains years away or where investors are paying heavily for future AI monetization.
Salesforce is somewhat better positioned than it was during its high-growth era because the company has become a serious cash-return story. Q2 free cash flow rose to $1.1 billion, and the company has been aggressively reducing its share count through buybacks, including the large accelerated repurchase announced earlier this year.
That changes the stock’s character.
Salesforce is no longer a company that must grow 25%-30% to justify itself. It is increasingly a mature platform combining low-double-digit revenue growth, expanding AI revenue, strong margins and capital returns.
That can attract a very different shareholder base.
Still, the valuation argument only works if AI proves additive. If Agentforce growth slows sharply or customers push back on pricing, the market can quickly return to the old narrative that mature SaaS companies are ex-growth assets with AI risk layered on top.
What would prove this is a real rotation?
I would use five tests.
The first is relative performance. IGV/SOXX needs to hold above the reclaimed trend zone rather than collapsing back into the prior downtrend.
The second is revenue evidence. AI-linked ARR, ACV, consumption or workflow metrics must continue appearing in quarterly results. Product announcements are not enough anymore.
The third is technical behavior on pullbacks. Software names that have reclaimed medium- or long-term moving-average bands need to hold them when the market has a bad week. A breakout that survives stress is much more informative than a gap after earnings.
The fourth is breadth. The move has to spread beyond Salesforce and CrowdStrike. ServiceNow, Adobe, Okta, Workday and adjacent software groups need to participate.
The fifth is rates. If the long end of the Treasury curve breaks materially higher, the valuation headwind can overwhelm improving fundamentals.
None of these requires perfect conditions.
But the more boxes that remain checked simultaneously, the stronger the case that software is moving from an AI victim trade into an AI monetization trade.
Data: company filings and investor-relations releases. Editorial chart.
Salesforce is the headline; IGV/SOXX is the scoreboard
The market spent most of the early AI cycle rewarding the companies that built the infrastructure: chips, networking, data centers and power.
The next phase may be about who turns that infrastructure into enterprise revenue.
Salesforce’s quarter suggests the answer will not belong exclusively to AI-native startups. The incumbents that own enterprise data, permissions, security and workflows have something the foundation models do not: embedded business context.
Claudeforce makes that logic visible. Claude can reason. Salesforce can tell it what the company allows, what the customer bought, what the sales process requires and which action should happen next.
That combination is much more valuable than another chatbot tab.
I would not call the SaaS-pocalypse over. Weak products can still be disintermediated. Seat counts can still come under pressure. Pricing friction can still slow adoption. And at 37 times earnings for the broader IGV basket, investors are not being paid to ignore execution risk.
But the burden of proof has changed.
Six months ago, software had to prove that AI would not destroy its economics.
Now some of the largest platforms are beginning to prove that AI can increase their economics.
CRM is the headline.
IGV/SOXX is the scoreboard.
Research note: The article expands the thesis and chart framework in the supplied Salesforce document. Current company figures were checked against Salesforce, Anthropic, CrowdStrike, Workday and iShares disclosures available through August 28, 2026.
