Article by Universe Bo Mingren, Shenchao TechFlow
In February 2026, the technology stock market is experiencing a systemic collapse referred to by some media as the "SaaSpocalypse."
Salesforce's stock has dropped nearly 40% from its 2025 high; ServiceNow plunged over 11% in a single day after its quarterly earnings report, simply because management mentioned on the earnings call that "AI agents are complicating the visibility of seat growth"; Workday fell more than 22%; together, the entire S&P 500 Software & Services index lost nearly $1 trillion in market value within the first six weeks of 2026.
The market logic is straightforward: AI agents have already replaced a large number of manual tasks; when companies use AI to accomplish what previously required 100 people, they no longer need 100 software licenses. The SaaS business model based on per-seat pricing is widely regarded as having reached its historical endpoint.
Amid this wave of panic trading sweeping through the industry, Stephen Bersey, Head of Technology Research at HSBC US, released a research report with a highly provocative title: “Software Will Eat AI.”
His core point, in one sentence: Market panic is a misjudgment.
A report emerging against the trend
The market's concern that AI will replace enterprise software is misplaced.
He wrote at the beginning of the report. In his view, AI will not eliminate software; rather, it will be absorbed by software, becoming a capability layer embedded within enterprise software platforms. Software is not AI’s adversary—it is the vehicle through which AI reaches the real world.
This logic flips the entire narrative framework of the current market. The market's fear is that "AI will replace software," while Bersey's view is that "software will tame AI."
He cited a historical analogy from the internet era: when the internet first exploded, initial value accumulation focused on physical infrastructure—servers, fiber-optic cables, and data centers. Massive capital flowed into hardware infrastructure, yet the struggling early internet companies ended up capturing the long-term value. Software, he noted, was the ultimate destination of internet value.
According to Bersey, the evolution of AI is replaying the same script. 2024 and 2025 are the years of infrastructure building—computing power, models, and code integration—all paving the way for a surge in the software layer. 2026 is the year when the engine truly ignites.
Software will be the primary mechanism through which AI spreads across the world's largest enterprises. We believe 2026 will be the year software monetization takes off.
Why can't foundational models replace enterprise software?
The most compelling argument in the report is a step-by-step deconstruction of the logic that "AI will directly disrupt software."
Critics' arguments appear compelling: large language models can already write code, vibe coding—generating functional software directly through natural language descriptions—is emerging, and AI model companies are increasingly experimenting with application-layer solutions. So why do businesses still need expensive, traditional software systems like Oracle, SAP, and Salesforce?
Bersey's response unfolds on three levels.
First, the base model has inherent flaws.
The report explicitly states that the foundational models "have inherent flaws" and are incapable of performing a "complete replacement" of core platforms for large enterprises. While they perform well in narrow scenarios—such as image generation, small-scale application development, and text processing—they are "not realistic" for high-fidelity, enterprise-grade core platforms.
The root cause lies in the limitations of training data. LLMs are trained on publicly available internet data, but the proprietary architecture knowledge, business logic, and operational standards accumulated over decades within enterprise software systems—core intellectual property—are not accessible online, making it impossible for AI to learn or replicate them. The competitive moat of systems like Oracle and SAP cannot be overcome by writing code alone; it is built over time through accumulated business scenarios.
Second, the capabilities of Vibe Coding are severely overestimated.
The report directly highlights Vibe Coding's critical weakness: it places the entire responsibility and burden of design on the developer. You tell the AI, “I want a system that handles global supply chains,” and the AI can generate code—but decisions such as “how to define the system’s architecture, how to handle exceptions, and how to ensure it doesn’t crash under extreme stress” still require human judgment.
More importantly, Bersey points out that the major AI model companies “have almost no experience building enterprise software.” They are entering an extremely complex environment from scratch. Enterprise software, after decades of iteration, has evolved to meet a baseline of “nearly zero errors, high throughput, and high reliability”—a standard that AI newcomers cannot match in the short term.
Third, the switching cost for businesses is a real and significant barrier.
Even if we assume AI could write code of equal quality, the cost for companies to replace their core systems remains extremely high—risks of revenue disruption, productivity losses, system compatibility issues across IT environments, and the accumulated trust in vendor brands and service capabilities—all of these are real switching costs that do not disappear just because AI can write code.
Enterprise-grade software demands proven, industry-tested 99.999% uptime and flawless operation across complex IT environments. This trust is earned over time, not built through code alone.
Who will truly benefit from AI monetization?
If the first half constitutes a defensive argument, the second half of the report presents an offensive strategy.
Bersey's core judgment is that the largest share of the AI value chain will ultimately flow to the software layer, rather than the hardware and chip layers.
We believe AI is the primary source of value creation in the software stack, and the largest share of long-term value will belong to software, not hardware.
He also noted that hardware scarcity, including GPU shortages, power constraints, and data center bottlenecks, will persist for the next several years. This scarcity underscores the strategic importance of software platforms: only software platforms can transform AI capabilities into scalable, repeatable business value.
The report points to agentic AI as the specific vehicle for monetization.
Bersey predicts that by 2026, task-oriented, workflow-embedded AI agents will be widely deployed in Fortune 2000 enterprises and small-to-medium businesses. However, his characterization of agents differs sharply from the dominant narrative in the market; he does not view agents as disruptive replacements for software, but rather as entities that must operate within parameters and permissions defined by software—this “bounded agent” approach is precisely what meets enterprises’ needs for AI risk management.
In other words, businesses do not need an all-powerful, unrestricted AI; they need an AI that can be governed, audited, and operated within a compliance framework—and only agents deeply embedded in enterprise software systems can achieve this.
Software is the key pathway for enterprises to use AI in a controlled manner.
At the same time, the report predicts that inference demand will gradually surpass training demand, becoming the primary driver of compute consumption growth. This means that as agents become more widespread, compute consumption will not decline but continue to grow, further supporting the entire software and infrastructure ecosystem.
Opportunity or trap?
At the time of the report’s release, the software sector’s overall valuation had fallen to a historic low. Bersey’s assessment is that the combination of low valuations and the impending year of monetization represents an opportunity to enter, not a signal to exit.
Software valuations are at historical lows, despite the industry being on the brink of massive expansion.
In terms of specific stock recommendations, HSBC’s logic is clear: software companies that have established strong data moats, possess embedded AI agent capabilities, and do not rely solely on headcount-based pricing models will be the biggest beneficiaries of this wave of AI monetization. The buy-rated list includes Oracle, Microsoft, Salesforce, ServiceNow, Palantir, CrowdStrike, and Alphabet—covering nearly all the key players in enterprise software.
Notably, HSBC also downgraded IBM and Asana, and placed Palo Alto Networks on a "sell" rating—not all software companies can weather this storm safely; the key lies in becoming the infrastructure for AI agent deployment, rather than merely an artificial interface that agents bypass.
Bersey's report is logically rigorous and perfectly timed; its contrarian stance itself carries strong viral potential.
But there’s one issue the report doesn’t directly address: if AI agents can truly operate efficiently within enterprise software frameworks, might the demand for software “seats” still be quietly declining? While the value of software as an AI vehicle may hold true, whether the “per-user” business model can justify current valuations remains an open question.
Will software devour AI, or will AI devour software? Every earnings report in 2026 will be new evidence in this debate.