Petit Rich | iStock | Getty Images
Financial analysts and experts offer endless elaborate and simplistic explanations for complex market movements. Over the years, the causes have ranged from El Niño and climate change to the World Trade Organization, rock and roll music, and water fluoridation. Many of these explanations have proven to be wrong, but commentators continue to reach for overly simplistic narratives to explain market movements. The latest example is the decline in software stocks.
Over the past year, the market has evaporated nearly $2 trillion in software value by asking the wrong questions about AI. The question behind so-called “SaaSpocalypse” — can an AI agent do what this software does? — is misguided because it treats software companies as nothing more than a bunch of tasks waiting to be automated.
The right questions to ask, and the clearest tests we know, to tell the difference between an AI winner and its car crash are different. Does this company own something that the AI agent cannot function without?
Salesforce’s recent explosive earnings show why the balance of power is shifting away from the frontier model and back toward the software companies that investors prematurely abandoned.
The story of “Saa Spocalypse” was very nonsensical.
marc benioff
Salesforce CEO
Analysts have assumed for months that Frontier’s large-scale language models would capture an ever-larger share of the value created by AI, and that the valuations of those companies would soar proportionately. But intelligence is exactly what is being commoditized before our eyes. Now, Frontier models are popping up one after another every few months, becoming not only more and more capable, but also more and more compatible.
AI may also be approaching RSI (recursive self-improvement), which could further accelerate the improvement cycle. Meanwhile, the price of raw intelligence plummets amid intense competition from well-funded rivals and China’s increasingly sophisticated open source models.
So, in this new world, if LLM is a commodity, where is the moat? We believe the answer is data. As Wells Fargo analysts put it, “As the cost of intelligence falls, the value of existing data increases.” That’s Economics 101. When an input becomes abundant and cheap, value shifts to its scarce complement. And for AI agents, the rare complement is trusted, proprietary data.
Agents who close sales still need a place to research customers, record interactions, save contracts, and customize terms. Without reliable proprietary data, AI agents are much less useful. As the old saying goes, “Junk in, junk out.”
Salesforce is one of the world’s largest repositories of enterprise customer data, and far from benefiting from lucky performance or temporary deviations, it could be positioned as a long-term structural winner from the commoditization of AI. Data keeps flowing in because agents can’t work without it. Salesforce’s Data 360 ingested a staggering 104 trillion customer records this quarter, an increase of 355% year-over-year. Meanwhile, the AI agent itself generates even more data, all of which must reach a trusted place. Salesforce delivered 3.2 billion units of agent work this quarter, nearly double the previous quarter.
As the accompanying chart shows, this flywheel (agents create jobs, jobs create data, data deepens the moat, and AI agents become more valuable with each successive generation) is why Salesforce’s annual combined AI and data recurring revenue has reached $3.9 billion, more than tripling in one year. Agent Force alone saw ARR jump from $100 million to more than $1.5 billion within 18 months of launch, a 240% year-over-year increase.
“The ‘SaaSpocalypse’ story was so much nonsense,” Salesforce CEO Marc Benioff said on CNBC with Jim Cramer, as the stock soared more than 20% after Wednesday’s earnings report. Growth is the strongest in four years, skeptics say seats will decline, Agentforce sales and service and Slack both increased seats year over year, and attrition rates are near record lows, with bookings quarter over quarter. We’ve more than doubled, and contract length has improved across all segments. Agents are using Salesforce more than ever.”
In fact, the numbers show why his argument is noteworthy. Pricing rights were to be evaporated and profit margins to be compressed. Instead, non-GAAP operating margins reached 34.1%, adjusted earnings of $5.90 per share nearly doubled, beating the consensus of about $3.27, revenue rose 11% to $11.35 billion, bookings grew even faster with a 14% increase in current remaining performance obligations, and management raised its full-year guidance to $46.4 billion.
But the most conclusive evidence came from the customers who were in the best position to know. If AI agents really can run without Salesforce, then Frontier AI Labs won’t be intermediating it. In return, they’re paying for it, with nine of the top 10 AI companies now running on Salesforce and Slack, and their combined spending is up 435% year over year. And we are partnering with the company, as exemplified by the debut of Claudeforce. In Benioff’s words, “Anthropic, the world’s No. 1 AI, and Salesforce, the No. 1 CRM, are working together.”
Companies building supposedly all-conquering models have concluded, as Benioff puts it, that those models “rely on CRM, not replace CRM.” This is what he means when he says “Salesforce is first and foremost a data business,” and why he’s betting on that claim with a $25 billion stock buyback, the largest in the company’s history.
As we argued, this test generalizes far beyond Salesforce. Not all software companies will thrive in this new world. Companies with proprietary data valuable to AI agents should be in a better position to prosper, while software companies that offer little more than functionality and have little else in customer stickiness may struggle. Balance sheets will also be important. Companies with strong free cash flow and little leverage will have more room to reinvest in AI transitions, while companies with high debt may be forced to use scarce cash to pay down debt instead.
Nevertheless, the direction of power in the AI economy is becoming clearer. We’re going from companies whose primary advantage is manufacturing intelligence, which is becoming more abundant month by month, to companies that own rare assets that cannot function without intelligence. That increasingly means data.
Jeffrey Sonnenfeld is the Lester Crown Professor of Leadership Practice at the Yale School of Management and director and founder of the Yale Institute for Chief Executive Officers. Steven Tian is director of research at the Yale University Institute for Chief Executive Officers and a former quantitative analyst at Rockefeller Capital Management. Stephen Henriquez is a senior fellow at the Yale Institute for Chief Executive Officers and a former McKinsey consultant.
