HSBC "Contrarian Statement": Software will devour AI, now is a good time to bottom fish

Can AI coding disrupt SaaS? HSBC's answer is quite the opposite.

According to the Wind Trading Desk, on February 24, HSBC's Head of U.S. Technology Research, Stephen Bersey, and his team released a report titled "Software Will Eat AI Stocks," throwing out a "counter-market statement."

In the current climate of "AI disruption and panic trading," HSBC clearly pointed out that software will not face extinction; rather, it is the key way for the world's largest enterprises to "control and utilize AI."

HSBC summarized its judgment with a contrasting statement: "Hardware/semiconductors are strong enough, but software will be better." The logic is that what enterprises truly need is not a "talking model," but a system capability that is controllable, auditable, and repeatable, which is precisely the strength of software platforms.

• Enterprise software will not be threatened by AI; on the contrary, AI will be embedded into software platforms.
• Enterprise software vendors have already completed heavy lifting work such as design, intuitive programming, and embedded intelligent agent testing.
• The valuation level of the software sector is at a historical low, even though the industry is poised for large-scale expansion.

The "Inherent Defects" of Large Models and Enterprise Barriers

The biggest concern in the market currently is that AI writing code (Vibe-coding) will significantly lower the software development threshold, allowing startups to easily disrupt existing SaaS giants.

HSBC firmly refuted this. The report pointed out that from a technical perspective, foundational AI models have "inherent defects." AI is essentially non-deterministic and may provide different answers or even make mistakes when faced with the same problem.

This is fatal in enterprise applications. "The world has become accustomed to software platforms that are repeatable, auditable, and error-free in daily operations, while foundational models lack these attributes." HSBC emphasized that it is unrealistic to expect "overall migration and replacement" using AI for high-fidelity enterprise-level platforms.

Moreover, enterprise software has reached extremely high throughput and reliability after decades of evolution. Behind this lies a vast amount of critical and proprietary intellectual property (IP) that cannot be trained on AI in the public internet. HSBC bluntly stated: "If you don't know what code you are writing, Vibe-coding is almost useless."

It's like a pharmaceutical company wouldn't design chips or smelt steel for its own use. Enterprises abandoned writing their core IT systems decades ago because it contradicted basic economic principles.

These companies quickly realized that developing and maintaining these systems internally and staffing them is very expensive; spending huge sums to write a massive platform that only spreads costs across one use case (i.e., for their own use) is extremely uneconomical. In contrast, purchasing products from software vendors with expertise in development, maintenance, and staffing is much more economical, as these costs can be spread across thousands of customers

Who Will Write the Best AI Software? It's the Traditional Software Giants

Since startups and large model providers lack the experience to build "enterprise-level" complex architectures, who is best suited to use AI to generate better software?

HSBC provides a very definitive answer: "Of course, it's the software vendors themselves."

The logic is very clear: Established software giants like Salesforce, Oracle, ServiceNow, and Microsoft have deep domain expertise, solid sales channels, and customer trust. More importantly, they have been embedding refined intelligent agents into their extensive platforms using the same AI programming tools.

The role of AI is to be "dimensionalized" and "tamed." HSBC made a vivid analogy: AI is responsible for creatively analyzing and producing intelligent data, but this data must be processed, stored, checked, and executed by deterministic software technology stacks.

"The vast majority of enterprise software will not be threatened by AI; instead, AI will be domesticated within the application technology stack through agents, creating tremendous value in the process."

2026: The Year of Software Monetization, Valuations at Historical Lows

From an investor's perspective, the technological logic ultimately needs to translate into performance guidance and market space.

HSBC provided a clear timeline: Major software giants will have begun the heavy work of designing and beta testing embedded AI agents by 2024, and the technology is now mature and starting to be promoted to large global clients.

"We believe that 2026 will be the kick-off for monetization of software," HSBC pointed out, noting that this will be the main mechanism for the world's largest enterprises to consume AI, driving exponential growth in AI inference demand.

Regarding the investment rhythm of the market, HSBC made a resounding conclusion: "As good as Hardware/Semi has been, Software will be better."

HSBC believes that AI is a technology, but "enterprises rarely buy technology; they buy solutions to business problems," and these solutions can only come from an infinitely flexible software technology stack. In this ecosystem that creates over $100 trillion in global GDP, traditional software giants are the core beneficiaries of unleashing AI's value potential.

Currently, the total addressable market (TAM) of the software industry is on the eve of a large-scale expansion cycle lasting 5-10 years. However, the misalignment in market perception has led to the software sector's valuations currently being at historical lows. HSBC suggests that now is the right time to establish or expand positions in the software sector before a valuation reassessment.