In a report published 18 months ago, we warned that “AI challenges were emerging and posing fundamental risks to entire segments of the software industry, suggesting that one would need to be very careful when investing in the sector.” Since then, we have remained underweight enterprise software—and that stance has paid off, with our AI and Big Data strategies up 14% YTD despite what has become a true software massacre. We see no reason to revisit our positioning, as competitive pressure from AI is moving from theoretical to very real.
In just a few days, two AI model releases—Anthropic’s Claude and Google’s Genie—sent shockwaves through the software ecosystem. In our view, rightly so. Anthropic unveiled new tools for its Claude Cowork agent that automate complex tasks across legal (contract review, legal briefs), sales, marketing, and data analysis.
Meanwhile, Google released Genie 3, a model capable of generating and simulating interactive 3D worlds in real time from simple prompts. While still imperfect—particularly around interactions—the progress since the first Genie model just two years ago is striking: from rudimentary 2D graphics to near console-quality 3D environments. It is easy to envision a refined version of Genie in the next couple of years evolving into a full-fledged game engine, directly competing with incumbents like Unity. More broadly, this would dramatically lower barriers to entry in video gaming, intensifying competition—even if studios with strong IP and franchises would remain relatively insulated in our view.
While much of the debate today around AI leaders and hyperscalers’ massive capex focuses on monetization and future profitability, we believe the software opportunity is massive in terms of revenue and margins and still largely underappreciated by the market. AI leaders are indeed rapidly tailoring their models to specific industry needs, with almost every industry likely to attract their attention. We should soon see domain-specific models for instance for drug discovery (posing risks to players like Schrödinger), as well as for financial analysis and modeling—challenging incumbents such as Bloomberg or Thomson Reuters.
To illustrate the point, OpenAI has reportedly assembled a team of over 100 former investment bankers from firms including Goldman Sachs, JPMorgan, and Morgan Stanley to train its models on financial workflows such as IPOs, restructurings, and valuation—essentially replicating core analyst tasks.
Even in a benign scenario where AI does not directly replace a software vendor, another risk remains: if AI reduces white-collar headcount, the number of software seats billed will inevitably decline. That alone puts long-term pressure on revenues and valuations.
In short, it is becoming increasingly clear that AI giants will move aggressively into territories historically dominated by traditional software, with the opportunity to replace companies whose combined valuation is in trillions (Salesforce, Adobe, ServiceNow, Bloomberg…). This calls into question the long-term outlook—and current valuation multiples—of much of the sector. Against this backdrop, we remain far more comfortable being overweight hardware and semiconductors, especially as new devices such as humanoids, robotaxis, and smart glasses are poised to open a new hardware growth cycle.
