Software stocks have completely exploded over the past two days.
Over the last two trading sessions, the software sector has outperformed the market (the S&P 500 Index) by more than 10 percentage points. This marks the largest two-day excess return in 25 years. Among them, Snowflake has risen 60% in the last three sessions, and Datadog has risen 56%.
Just six months ago, Wall Street was collectively bearish on software stocks.
In May this year, Goldman Sachs analyzed the holdings of over 1,000 active hedge funds. As of the end of the first quarter, the allocation to software stocks was only about 6%, the lowest level since 2019.
But half a year later, the sector that was initially 'sentenced to death' by AI has become one of the best-performing sectors in the U.S. stock market.
From being avoided by everyone to frantic capital rebalancing, what exactly happened to software stocks? Was the initial market judgment that 'AI will disrupt software' wrong from the start?
01 Surging 21% in a Single Month, the Largest Gain in Nearly 5 Years
In May that just passed, U.S. software stocks completely took off.
The iShares Expanded Tech-Software Sector ETF (IGV), an ETF focused on North American software stocks, rose about 21% in May, achieving its best monthly performance since October 2021. By June 1st, it rose another 6%, turning its year-to-date gain positive again.
Looking at individual stock performance is even more dramatic. For example, Snowflake rose 87% in a month, Datadog rose 58%, and Figma rose 40%. Across the entire software sector, monthly gains of 20% or more are not uncommon.
This surge in software stocks boils down to just two logics.
First, earnings disproved the panic about 'AI impacting software'.
Over the past two years, there has been a market fear: as OpenAI, Anthropic, and others continuously enhance model capabilities, AI itself can write code, analyze data, and generate reports. So what value do software companies like Snowflake, Datadog, and Salesforce still have?
But the earnings season gave the opposite answer.
Snowflake became one of the most important triggers for this rebound.
On May 27th, the company not only raised its full-year product revenue guidance but also signed a $6 billion long-term cooperation agreement with AWS, focusing on generative AI and Agent infrastructure.
After the news was announced, Snowflake's stock surged over 36% in a single day.
More crucially, the management clearly stated that more and more enterprises are deploying AI workflows onto the Snowflake platform. Initially, the market feared AI would bypass software. The result shows that AI actually needs more software.
A similar story happened with identity and access management service provider Okta. In Q1, the company's revenue was $765 million, higher than the market expectation of $752 million. Adjusted earnings per share were $0.91, also beating the expected $0.85. On that day, Okta's stock price soared 30%.
In fact, this viewpoint was mentioned by someone in March, but it didn't receive enough attention.
At the time, Deutsche Bank believed that although the market kept discussing how AI would hurt software companies, to date, they hadn't found any major software company expecting AI to have a material negative impact on this year's revenue.
On the contrary, the U.S. software industry's profit growth remains near 30%, and next year's profit expectations are still being revised upwards.
The second logic is that institutional positioning in software stocks is simply too low.
Not long ago, Goldman Sachs analyzed the holdings of over 1,000 active hedge funds. As of the end of Q1, semiconductors already accounted for nearly 10% of portfolios, but the allocation to software stocks was only about 6%, the lowest level since 2019.
When earnings proved software wasn't destroyed by AI, these underweighted funds were forced to rebalance their positions. Thus, the rebound quickly evolved into a short squeeze.
What's more interesting is that as stock prices rose, market views on software stocks also began to change.
Recently, Goldman Sachs publicly stated that the AI hardware hype has peaked, and profits are starting to shift from hardware to software.
Goldman's logic is mainly two points: Software stocks are starting to find their business models and will accelerate monetization next; AI profits will shift from hardware to software; Cloud giants, considering cash flow issues, will cut capital expenditures in the future, impacting hardware.
So the question is, from the beginning-of-the-year fear that 'AI will eat software' to the current collective software stock rebound, how should we reinterpret AI's impact on the software industry?
02 AI Also Needs Software
The market's past panic was built on an assumption: when Agents become powerful enough, people will no longer need software.
But what's happened over the past half-year increasingly points in another direction.
AI Agents aren't reducing software usage; they might instead become new users of software.
Dan Shipper, founder of Every, proposed a very interesting viewpoint: past software mainly served humans; future software will likely serve both humans and thousands of Agents simultaneously.
In the past, an employee might click a software interface only dozens of times a day; in the future, an Agent might call APIs, query databases, and execute workflows every minute.
Software hasn't disappeared; its usage frequency might actually become higher.
Dan Shipper mentioned that even in a highly AI-driven company like Every, SaaS spending continues to grow.
Okta is a classic case. The market previously thought that as Agents become smarter, the importance of identity authentication and access management would decline.
But reality is precisely the opposite. Employees need identity management, and Agents also need identity management.
In the future, an enterprise might have 1,000 employees while running 10,000 Agents. Which data can these Agents access, which systems can they call, what operations can they perform, and how to trace issues when they occur—all require new governance systems.
In other words, the Agent era hasn't weakened Okta's value; it has actually expanded its market space.
This is also why Jensen Huang recently emphasized repeatedly that Agents won't eliminate software companies.
The reason is simple: Agents themselves need software—databases, CRM, identity management systems, payment systems, monitoring systems, and various industry tools. The task of future software companies will no longer be just serving human users but also becoming infrastructure that Agents can call upon and collaborate with.
03 The Distance from Intelligence to Results is Software's Moat
Even if model capabilities continue to improve, large model companies may not directly consume the entire application layer. This is a point recently raised by Joe Schmidt, a partner at a16z.
He believes that model companies like OpenAI and Anthropic will enter more and more application scenarios. Especially in areas like code generation, writing, and image generation, the stronger the model, the better the product experience tends to be.
But the enterprise software world is far more complex than these scenarios. Many enterprise processes aren't simply calling a few tools; they involve multiple systems, multi-person collaboration, approval workflows, historical rules, industry experience, and regulatory requirements.
This is especially true in industries like legal, insurance, finance, healthcare, and sales operations. Vast amounts of critical knowledge are embedded in workflows formed from long-term operations, exception handling, human judgment, and historical feedback.
For now, there remains a considerable distance between general-purpose models and real-world business. And this distance represents the opportunity for AI application companies.
This distance primarily comes from three aspects.
First is experience. The most valuable knowledge in many industries circulates within real business processes.
Why was an insurance application denied? Why did a sales lead ultimately convert? Why must a particular customer service issue be escalated? This experience only solidifies into systemic capabilities after accumulating a large number of real cases.
A system that has handled thousands of insurance underwriting cases and a new product just entering the industry obviously won't understand problems at the same level.
Second is cost. Real enterprises won't call the strongest, most expensive model for every task. Mature AI applications typically employ multi-model orchestration, scheduling based on business needs.
For example, using large models for complex tasks, medium models for standard tasks, and smaller, cheaper models for repetitive tasks.
In this process, large model companies provide general intelligence, while AI application companies are responsible for translating this intelligence into profitable business processes.
Third is governance. The closer to core business operations, the more enterprises care about controllability. For instance, healthcare has privacy requirements, finance has regulatory requirements, and law has professional norms.
Enterprises care not only whether AI can complete tasks but also what data it accessed, what operations it performed, and how to assign responsibility when problems arise.
Therefore, what many AI application companies ultimately deliver is not just a set of model capabilities but a complete set of operational mechanisms that enterprises can accept and trust.
Sales is a typical example. On the surface, AI sales is just finding clients, writing emails, sending messages. But once truly implemented, it quickly evolves into a complex set of processes.
Client screening, information completion, background research, channel selection, outreach cadence, and result feedback—each link affects the final conversion rate. The true value of AI applications lies in connecting these links and continuously optimizing them.
So, when we re-examine AI's impact on the software industry, we might discover an interesting phenomenon:
Enterprises won't pay just because a model is smarter. What truly makes enterprises pay is the ability to reliably transform intelligence into results. And between intelligence and results, there still lies a gap of complex business processes, industry experience, and organizational rules.
Half a year ago, the market panicked and asked: Will AI kill software?
Looking back now, the answer is already clear. AI won't kill software, but it will redefine software. And those companies that can complete this redefinition first will become the biggest winners in the next cycle.
This article is from the WeChat public account "Silicon Valley Observer Pro", author: Silicon-based Analyst
