In the span of a few weeks, software markets have gone through one of their sharpest repricings in years. Broad software indices have fallen, valuation multiples have compressed and the narrative has taken hold that AI is beginning to consume traditional SaaS budgets.
Per Bobby Molavi, 164 stocks across software, financial services and asset management shed roughly $611 billion in market value in a single week. A Goldman Sachs basket of software stocks dropped to around 21x estimated profits, down from peaks >100x in 2021.
What is happening underneath all this is nuanced.
Take Salesforce, the pioneer of SaaS. Salesforce now trades at roughly 14x forward profits compared with a decade-long average near 46x. The company itself is not in distress - it recently reported ~$12.9 billion in trailing free cash flow, +22% YoY, with operating margins near 35%. It recently signed a $5.6 billion contract with the US Army!
Clearly what has shifted is not earnings today, but investor confidence in how durable and predictable future cash flows will be. Rather than underwriting growth several years into the future, investors are focusing on what they believe is certain and discounting the rest more aggressively.
In practice, public SaaS growth rates have been declining every quarter since the 2021 peak and should arguably have triggered a valuation reset earlier, but AI has acted as the catalyst that made this slowdown impossible to ignore.
AI has clarified enterprise technology budget allocation. Jason Lemkin summarises the dynamics: AI budgets +100% YoY, overall budgets +8% YoY, app count: flat, new net customers: declining, seat counts: under pressure.
In other words, AI agents are not replacing systems like Salesforce tomorrow, but they are absorbing most of the incremental budget. When total IT spending grows modestly but AI spending accelerates rapidly, expansion dollars inevitably shift away from traditional seat-based software models.
The story unfolding in 2026 is not the death of SaaS (or human labour for that matter), but certainly a redistribution of value. The challenge is understanding where durable value will settle as this transition continues.
The transition phase
The idea that AI and vibe-coding will simply replace enterprise software misses a basic reality: building a prototype is easy, but running, securing and integrating mission-critical software for thousands of employees is not. AI dramatically speeds up building new tools, but it does not replace decades of development, compliance, integrations and operational reliability embedded in systems of record. This is why a more pragmatic transition model is emerging: software combined with a lean human operational layer (often associated with Palantir’s FDE model).
Instead of selling pure software, companies increasingly deliver automation plus human support, replacing large volumes of manual back-office work while keeping oversight where needed. It is effectively “software + operators,” particularly powerful in government, logistics, healthcare and other historically labour-intensive or technologically lagging sectors. Palantir is essentially Accenture with a software multiple, but the model works because enterprises are layering automation on top of existing systems rather than replacing them outright.
AI delivered well into underserved, labour-heavy sectors may be one of the most compelling investment opportunities in this transition.
Phase II
Looking further ahead, vertical integration is incredibly compelling. Owning both the asset and the software stack can be more powerful than selling software alone. Agents and automation increasingly commoditise interface and workflow logic, but deep operational expertise embedded in hospitals, logistics fleets, manufacturing systems or energy infrastructure is difficult to replicate.
Operators who deploy proprietary AI and software internally can create durable advantages. Over time, SaaS multiples are likely to normalise toward those of other industries, with companies growing in the low-to-mid teens and generating solid margins trading closer to EBITDA multiples than historical lofty revenue multiples.
The next wave of businesses may combine tightly coupled software with owned assets - running hospitals, logistics operations or industrial infrastructure with proprietary AI systems that allow them to operate more efficiently than competitors. AI makes building and maintaining bespoke operational software far easier than a decade ago, allowing value to shift toward companies that control real-world assets while using software as a performance multiplier.
Uncertainty
Through this transition, different groups must track different signals.
Technology investors will focus on how net revenue retention evolves, how much growth comes from pricing versus usage or seat expansion, how churn differs across pricing models, whether pricing units shift toward transactions or workloads and whether sales efficiency holds as AI-enabled products scale.
Policymakers will monitor how automation changes employment patterns, how quickly AI adoption spreads across sectors with public consequences and whether regulation and labour policy keep pace with economic change.
As individuals we face our own recalibration - tracking which skills remain valuable, how quickly roles shift from manual execution to AI-assisted work and tracking which pockets of the economy benefit or suffer in the revised human labour <> technology union.