While everyone is selling off software stocks, HSBC says you’re wrong.
TechFlow Selected TechFlow Selected
While everyone is selling off software stocks, HSBC says you’re wrong.
Market panic is a misjudgment.
Navigating Web3 tides with focused insightsBy Universe Boruto, TechFlow
In February 2026, the tech stock market is undergoing a systemic collapse dubbed the “SaaSpocalypse” by some media outlets.
Salesforce’s stock price has fallen nearly 40% from its 2025 peak; ServiceNow plunged over 11% in a single day following its quarterly earnings release—triggered merely by management noting on the earnings call that “AI agents are complicating visibility into seat growth”; Workday dropped over 22%; and the entire S&P 500 Software & Services Index lost nearly $1 trillion in market capitalization within the first six weeks of 2026.
The market’s logic is straightforward: AI agents can now replace vast amounts of manual work. Enterprises use AI to accomplish tasks previously requiring 100 people—and therefore no longer need licenses for 100 software seats. The per-seat SaaS business model is widely believed to have reached its historical endpoint.
Amid this wave of panic-driven trading sweeping the entire industry, Stephen Bersey, Head of U.S. Technology Research at HSBC, released a provocatively titled research report:“Software Will Eat AI”.
His core thesis, summarized in one sentence:The market’s panic is a misjudgment.
A Counter-Cyclical Report
“The market’s concern that AI will displace enterprise software is misplaced.”
He writes in the report’s opening. In his view, AI will not eliminate software—it will instead be absorbed by it, becoming an embedded capability layer within enterprise software platforms.Software is not AI’s rival; software is the vehicle through which AI reaches the real world.
This logic flips the prevailing market narrative on its head. While the market fears “AI replacing software,” Bersey asserts “software taming AI.”
He draws a historical analogy from the internet era: when the internet first exploded, initial value accumulation centered on physical infrastructure—servers, fiber-optic cables, data centers. Massive capital flooded into hardware infrastructure, while early internet companies—many struggling at the time—ultimately captured long-term value.Software was where internet value ultimately landed.
Bersey believes AI’s evolution is replaying the same script. 2024 and 2025 were infrastructure-building years—focused on compute, models, and code integration—all laying the groundwork for an explosion at the software layer. And 2026 is the year the engine truly ignites.
“Software will be the primary mechanism for AI diffusion across the world’s largest enterprises. We believe 2026 marks the launch year for software monetization.”
Why Foundation Models Cannot Replace Enterprise Software
The report’s most substantial argument is a step-by-step deconstruction of the logic behind “AI directly disrupting software.”
Critics’ arguments appear compelling: large language models (LLMs) can already write code; vibe coding—generating functional software directly from natural-language descriptions—is gaining traction; and AI model companies are increasingly venturing into application layers. So why would enterprises still need costly legacy systems like Oracle, SAP, or Salesforce?
Bersey responds on three levels.
First, foundation models have “inherent limitations.”
The report explicitly states that foundation models “possess intrinsic flaws” and cannot realistically undertake “full replacement” of large enterprises’ core platforms. They perform well in narrow domains—image generation, small-scale app development, text processing—but for high-fidelity, enterprise-grade core platforms, such wholesale substitution “is not realistic.”
The root cause lies in training data constraints. LLMs are trained on publicly available internet data, whereas decades of proprietary architectural knowledge, business logic, and operational protocols accumulated within enterprise software systems—core intellectual property—simply do not exist on the public web. AI cannot learn or replicate them. Oracle’s and SAP’s moats aren’t built by writing code—they’re forged through time and deep immersion in real-world business scenarios.
Second, the capabilities of vibe coding are severely overestimated.
The report directly calls out vibe coding’s fatal weakness: it places the entire burden of design responsibility squarely on developers. You tell the AI, “I want a system to manage global supply chains,” and it may generate code—but decisions about “how to architect that system, how to handle exceptions, how to ensure it won’t crash under extreme load” remain entirely human responsibilities.
More critically, Bersey points out that leading AI model companies “have virtually no experience building enterprise-grade software.” They are entering an extremely complex domain from scratch. Meanwhile, enterprise software—refined over decades—has evolved to deliver “near-zero error rates, high throughput, and exceptional reliability”—a benchmark that AI newcomers cannot match in the near term.
Third, enterprise switching costs constitute a very real, formidable barrier.
Even assuming AI could produce code of equivalent quality, the cost for enterprises to replace core systems remains prohibitively high—risks of revenue interruption, productivity loss, cross-IT-environment compatibility issues, and the accumulated trust in vendor brands and service capabilities… these are tangible switching costs that won’t vanish just because AI can write code.
Enterprise software demands proven, multi-year uptime of 99.999%, flawless operation across diverse and complex IT environments. That trust is earned over time—not built line-by-line in code.
Who Will Truly Benefit from AI Monetization?
If the first half of the report serves as a defensive rebuttal, the second half constitutes an offensive strategic roadmap.
Bersey’s central thesis: the largest share of the AI value chain will ultimately flow to the software layer—not the hardware or chip layer.
“We believe AI is the primary source of value creation within the software stack, and the largest share of long-term value will accrue to software—not hardware.”
He further notes that hardware scarcity—including GPU shortages, power constraints, and data center bottlenecks—will persist for years to come. This scarcity, paradoxically, strengthens the strategic position of software platforms:Only software platforms can translate AI capabilities into scalable, repeatable commercial value.
And the specific vehicle for monetization? The report points to AI agents (agentic AI).
Bersey forecasts that 2026 will see widespread deployment of task-oriented, workflow-embedded AI agents across Fortune 2000 enterprises and SMEs. Yet his characterization of agents diverges sharply from mainstream narratives: he does not view agents as disruptive replacements for software, but rather as entities operating strictly within parameters and permissions defined by software. It is precisely this “bounded intelligence” that satisfies enterprises’ risk-management requirements for AI.
In other words, enterprises don’t need an omnipotent, unbounded AI running freely—they need an AI that is governable, auditable, and compliant with regulatory frameworks. And only agents deeply embedded within enterprise software systems can fulfill that requirement.
“Software is the critical pathway for enterprises to use AI controllably.” This is the report’s most pivotal conclusion.
Simultaneously, the report predicts inference demand will gradually surpass training demand as the primary driver of compute consumption growth—meaning that as agents proliferate, compute consumption will not shrink but continue rising, 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 already slumped to historic lows. Bersey’s assessment is:Low valuations combined with the imminent monetization inflection point represent an entry opportunity—not a signal to exit.
“Software valuations are at historic lows, even though the industry stands on the cusp of massive expansion.”
In terms of specific stock recommendations, HSBC’s logic is clear: software companies that have already established deep data moats, possess strong capabilities to embed AI agents, and do not rely solely on per-seat pricing models will be the biggest beneficiaries of this AI monetization wave.Buy-rated names include Oracle, Microsoft, Salesforce, ServiceNow, Palantir, CrowdStrike, and Alphabet—covering virtually all core enterprise software players.
Notably, HSBC downgraded IBM and Asana, and placed Palo Alto Networks on its “reduce” list—not all software firms will safely navigate this transition. The key differentiator lies in whether a company becomes the infrastructure enabling AI agent deployment—or gets bypassed as a mere human interface.
Bersey’s report is logically rigorous, impeccably timed, and its contrarian stance itself carries strong viral potential.
Yet one question remains unanswered in the report: if AI agents truly operate efficiently within enterprise software frameworks, won’t demand for software “seats” quietly continue to erode? While software’s value as an AI delivery vehicle may hold, whether the per-seat business model can sustain current valuations remains an open question hanging in the air.
Will software eat AI—or will AI eat software? Every earnings report in 2026 will serve as new evidence in this debate.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
