Jensen Huang Says: The Software Sell-Off Makes No Sense!

Recently, automated tools developed by AI unicorns such as Anthropic have triggered a wave of selling, sending global software stocks sharply lower.

However, Nvidia CEO Jensen Huang said the sell-off in software stocks “makes no sense,” adding that time will ultimately prove the market wrong. In his view, AI will rely on existing software tools to accomplish tasks rather than reinventing them from scratch. “Would you use a screwdriver,” he asked, “or invent a new screwdriver every time?”

Huang specifically highlighted ServiceNow, SAP, Cadence, and Synopsys, calling them standout companies within the software sector.

He added that Nvidia itself has widely adopted a range of software tools, significantly improving employee productivity and allowing teams to focus on the company’s core mission: designing semiconductor chips and the systems that support them.

Deep Moats Still Protect Software Leaders: System Depth, Licensing, and Networks Are Beyond AI

Huang’s argument is not without merit. Leading software companies continue to enjoy deep competitive moats. The first is system depth. Take cloud software leader Salesforce as an example: AI tools may be able to replicate a superficial “shell,” but that accounts for perhaps 2% of Salesforce’s true value. What about the remaining 98%? Who maintains and upgrades the system? Who handles security audits? Who integrates with more than 500 upstream and downstream enterprise tools?

The moat of enterprise software lies not in simply “writing the code,” but in long-term expansion, compliance, and maintenance—capabilities Salesforce has built over two decades. Claims that AI agents will rapidly disrupt SaaS do not withstand scrutiny when tested against real-world operations.

The legal software sector, which was hit particularly hard in this sell-off, is protected by real-case data and licensing barriers. Thomson Reuters has deep roots in the legal industry through Westlaw, its proprietary database containing over a century of U.S. case law and nearly two billion legal documents. Investors widely view Westlaw as the crown jewel of its legal business—an asset AI simply does not possess.

In highly specialized professions such as law, licensing is mandatory, and AI does not hold such credentials. Anthropic itself has cautioned that AI-generated legal analysis must be reviewed by licensed attorneys before being used in legal decisions. Beyond law, industries such as medicine and finance also require professional certification, and AI currently holds no regulatory licenses in any of these fields.

Why AI Won’t Overturn the Gaming Software Industry Anytime Soon

Gaming software stocks have previously experienced AI-driven sell-offs, but near-term disruption remains unlikely. A Bernstein report notes that generative AI models are fundamentally “probabilistic,” predicting the next pixel or frame based on statistical patterns. Video games, by contrast, must be “deterministic,” governed by strict rules to deliver consistent player experiences.

When playing an action game, players demand precise feedback and reliable mechanics—not randomly generated “surprises.” The moat in game development runs far deeper than producing visually appealing 3D assets.

The report emphasizes that “fun” is an extraordinarily complex concept. AI-generated visuals alone cannot solve whether a narrative is engaging, combat feels satisfying, or character progression is balanced. In multiplayer games, complexity increases exponentially, encompassing in-game balance, anti-cheat systems, matchmaking, and live-service tuning—factors that generative AI cannot meaningfully understand or replace.

Believing that AI can rapidly displace traditional game engines ignores the reality that games are complex, interactive media ecosystems.

AI has undoubtedly lowered barriers to entry, resulting in a surge in supply. In 2025, more than 20,000 new games were released on Steam, partly due to AI-assisted development. Yet truly high-quality titles remain scarce: only 137 games received more than 10,000 user reviews.

This suggests that AI has increased the supply of low-cost, one-off experiences, but not the supply of distinctive, premium content. Human dopamine responses diminish with repetitive experiences, and flooding the market with homogenized AI-generated content does not lead to sustainable commercial success.

Ultimately, IP strength, brand recognition, and emotional attachment remain decisive. Whether Nintendo’s iconic franchises or the Grand Theft Auto series, these long-established emotional connections cannot be replicated overnight by AI.

Game companies also possess formidable operational expertise. Bernstein argues that long-term live operations have no “one-size-fits-all” solution. Even similar titles like Battlefield 6 and Arc Raiders can exhibit vastly different player retention outcomes.

Sustaining a game’s lifespan requires continuous, context-specific adjustments based on player feedback—dynamic decision-making for which there is currently no evidence AI can outperform elite human teams. If even industry veterans cannot guarantee success, expecting AI to autonomously manage complex online communities is unrealistic.

Goldman Sachs echoes this view, noting that commercially viable games must simultaneously satisfy four conditions: repeatable and controllable gameplay logic; long-term balance of progression systems; sustained content updates and live operations; and mature acquisition, retention, and monetization mechanisms. AI currently possesses none of these capabilities.

The Real Risk for Software Stocks: Budget Reallocation, Not AI Disruption

The true risk facing software stocks is not near-term AI disruption, but customer spending cuts.

Consider the data: enterprise AI budgets are growing more than 100% year over year, while overall IT budgets are up only 8%. New customer growth is slowing, and software seat counts are under significant pressure.

If AI budgets double while total IT spending grows only modestly, where does the extra funding come from?

It doesn’t materialize out of thin air—it is squeezed directly out of SaaS budgets: fewer purchased seats, canceled new application rollouts, and reduced add-on modules.

Since 2021, growth rates among publicly traded SaaS companies have declined quarter after quarter. Recent “growth” has largely come from price increases on existing contracts and deeper monetization of current customers—not from net new customer additions.

Put plainly, this is not true growth; it is harvesting the existing user base. Without new customers, pricing power eventually hits a ceiling—and churn risk rises.

The core question for software companies is whether customers will continue spending on their software, or reallocate budgets toward AI investments. With fixed total budgets, every additional dollar spent on AI is one less dollar for Salesforce seats, Workday modules, or ServiceNow components.

Moreover, as AI replaces certain human tasks, enterprises require fewer employees—and therefore fewer software licenses.

Earnings season confirms this pressure. Bloomberg data shows that among S&P 500 software companies that have reported results, only 71% exceeded revenue expectations, compared with 85% across the broader technology sector.