Khan Capital | January 2026
Key Takeaways
- The iShares Software ETF fell 20% YTD by mid-February 2026, with Microsoft losing $400 billion in a single session and enterprise names like Salesforce (-30%), ServiceNow (-28%), and Intuit (-34%) punished severely.
- The sell-off was driven by AI agents threatening the per-seat licensing model and “capex burnout” anxiety over the gap between hyperscaler infrastructure spending and revenue monetisation.
- Capital rotated aggressively into industrials and financials: Caterpillar surged 28% as a “Physical AI” beneficiary, while JPMorgan’s market cap exceeded $900 billion.
- Bank of America called the sell-off logically inconsistent, while J.P. Morgan noted it was “indiscriminate” and Goldman Sachs warned it may be “the end of the beginning.”
- The AI trade has entered Phase Two: differentiation between infrastructure winners, software survivors, and AI-native disruptors, replacing the undifferentiated enthusiasm of 2023-2025.
In the final week of January 2026, Microsoft lost approximately $400 billion in market capitalisation in its worst trading session since March 2020. Azure cloud growth decelerated. Capital expenditure reached a record $37.5 billion for the quarter, a 66% increase year-over-year. And Wall Street, which had spent two years cheerfully funding the AI infrastructure buildout, suddenly demanded to know: when does the spending become the earning?
Microsoft’s reckoning was the catalyst, but the sell-off that followed engulfed the entire software sector. By mid-February, the iShares Expanded Tech-Software ETF had plummeted by 20% year-to-date. Salesforce was down 30%. ServiceNow had fallen 28%. Intuit had lost over 34%. Even Palantir, 2025’s market darling, shed 22%. In Europe, the Stoxx Software and Computer Services index dropped over 5% in a single session. The carnage was indiscriminate, punishing quality names alongside speculative growth. Analysts began calling it the “Saaspocalypse.”
| Stock | 2026 YTD Decline | Key Vulnerability |
|---|---|---|
| Microsoft (MSFT) | -$400bn in single session | Azure deceleration; $37.5bn quarterly capex |
| Salesforce (CRM) | -30% | Per-seat CRM model under existential threat |
| ServiceNow (NOW) | -28% (-34% at trough) | Beat earnings but sold off 14% in a week |
| Adobe (ADBE) | -27% | Generative AI challenges creative suite |
| Intuit (INTU) | -34% | AI automation of tax/accounting workflows |
| Palantir (PLTR) | -22% | Still at 97x forward P/E after sell-off |
| Caterpillar (CAT) | +28% | “Physical AI” beneficiary; record backlog |
| JPMorgan (JPM) | Market cap past $900bn | Safe harbour for capital exiting tech |
The Catalyst: AI Agents Threaten the Per-Seat Model
The proximate trigger for the sell-off was Microsoft’s earnings report, but the underlying anxiety had been building for months. The emergence of autonomous AI agents, software systems capable of navigating interfaces, executing workflows, and performing tasks without human intervention, raised a fundamental question about the business model that underpins the entire enterprise software industry: if AI agents can do the work that previously required human employees, what happens to per-seat licensing revenue?
The traditional SaaS model charges customers based on the number of users (“seats”) who access the software. More employees equals more licences equals more revenue. But the AI agent paradigm threatens to sever the link between headcount and software spend. If a single AI agent can perform the work of multiple human operators, the number of required software licences declines even as the value delivered increases.
Salesforce, whose entire business model is built on per-seat CRM licences, was among the hardest hit precisely because its revenue model is most directly exposed to this dynamic. Adobe, ServiceNow, and other enterprise platforms face similar structural questions. The market was not merely repricing near-term earnings; it was repricing the terminal value of business models that may be fundamentally disrupted by AI automation.
The Capex Paradox: Spending More, Earning Less
Microsoft’s earnings report crystallised a paradox that the market had been grappling with since DeepSeek’s emergence in January 2025: the companies spending the most on AI infrastructure were not yet generating proportionate returns. Microsoft’s $37.5 billion quarterly capex was a record, but Azure revenue growth was decelerating. The return on investment from AI spending remained, at best, an article of faith.
Wall Street’s patience with the “build it and they will come” narrative was exhausting. For two years, investors had accepted the argument that massive AI infrastructure investment was necessary and that monetisation would follow. The DeepSeek shock of January 2025, which demonstrated that competitive AI models could be built at a fraction of the cost, had already planted doubts. Microsoft’s Q2 results were the confirmation that those doubts were justified.
The concept of “capex burnout” entered the market vocabulary. Investors were not questioning whether AI was transformative; they were questioning whether the current level of capital expenditure was justified by the timeline to monetisation.
The Rotation: From Digital to Physical
As capital fled software, it found a new home in sectors that the market had largely ignored during the AI euphoria of 2024-2025. Caterpillar surged 28% in the first six weeks of 2026, reimagined by investors as a “Physical AI” play: the company whose heavy machinery builds the data centres, lays the fibre optic cables, and upgrades the power grid that AI infrastructure requires. Its order backlog reached a record $39.9 billion. JPMorgan’s market capitalisation pushed past $900 billion as a steepening yield curve and M&A activity revival attracted capital fleeing tech volatility.
The rotation reflected a deeper intellectual shift. The “AI premium” was migrating from the application layer (software companies deploying AI) to the physical infrastructure layer (companies building the data centres, power plants, and networking equipment that AI requires). The market was, in effect, deciding that the picks-and-shovels of the AI revolution were not software licences but concrete, copper, and kilowatt-hours.
What the Market Is Misunderstanding
The sell-off has been indiscriminate, and that creates opportunity. The market is treating all software companies as equally vulnerable to AI disruption, but the reality is far more nuanced. J.P. Morgan noted that the sell-off had been “indiscriminate” and that emerging evidence suggests AI is “more likely to be additive to software workflows in the near term rather than a substitute.” Quality software businesses trading at 15-20x forward earnings, down from 30-40x, may represent compelling value for investors willing to look past the headline panic.
The per-seat model is evolving, not dying. Several software companies are already transitioning from per-seat to consumption-based or outcome-based pricing models that align revenue with the value AI delivers rather than the number of human users. This transition will be disruptive in the near term, creating uncertainty about revenue trajectories, but it ultimately positions software companies to capture a larger share of the value their products create.
Nvidia’s CEO may be right about the “most illogical” narrative. Jensen Huang argued publicly that the notion that the software industry is being replaced by AI is “the most illogical thing in the world.” His logic: AI agents will use existing software platforms as tools, increasing rather than decreasing their value.
Goldman Sachs’ warning deserves attention. While J.P. Morgan counselled that the sell-off had overshot, Goldman Sachs warned that the downturn in software may be “the end of the beginning, rather than the beginning of the end.” If AI agents genuinely reduce the number of human workers performing software-mediated tasks, the revenue implications for per-seat models could be far more severe than the current 21% drawdown implies.
Bank of America called the sell-off logically inconsistent. BofA’s Vivek Arya argued that investors were simultaneously pricing two mutually exclusive scenarios: that AI capex would deteriorate (hurting infrastructure stocks) while also being so transformative that it would make existing software obsolete. Both cannot be true simultaneously.
Structural Interpretation: The AI Trade Enters Phase Two
The 2026 software sell-off marks the transition from Phase One to Phase Two of the AI trade. Phase One (2023-2025) was characterised by undifferentiated enthusiasm: everything connected to AI went up. Phase Two is the differentiation phase, where the market begins to distinguish between AI’s winners and losers within the technology sector. Infrastructure providers are being repriced based on actual demand trajectories. Software companies are being repriced based on their vulnerability to AI disruption. AI-native companies are being valued separately from companies that are merely deploying AI features in existing products.
This differentiation is healthy but painful. Historically, every major technology cycle, from the internet to mobile to cloud, has experienced a similar transition from undifferentiated hype to differentiated reality. The winners that emerge from this phase tend to be the dominant franchises of the next decade.
Implications for Investors
Software stock selection has replaced software sector allocation. The era of buying software as a thematic basket is over. Individual company analysis, focused on pricing model vulnerability, competitive moat durability, AI integration capability, and free cash flow generation, is now essential.
The “builders vs. adopters” framework is useful. Companies building AI infrastructure (semiconductors, cloud, networking) face different dynamics from companies whose products may be disrupted by AI. Investors may wish to evaluate their tech exposure through this lens.
Value and industrial stocks may deserve increased allocation. The rotation into Caterpillar, JPMorgan, and other “Old Economy” names reflects more than a temporary flight from tech volatility. If the AI buildout is a multi-year physical infrastructure project, the companies that supply the materials, equipment, and financing for that project have a structural earnings tailwind.
Watch the Q1 earnings cycle for pricing model evidence. The next round of software earnings reports will be scrutinised for evidence of AI-driven revenue impact, pricing model transitions, and capex efficiency improvements. Any company that demonstrates credible progress toward outcome-based pricing will be rewarded disproportionately.
Conclusion
The 2026 software sell-off is not the death of the software industry. It is the market’s belated recognition that AI is a disruptive force that will reshape software business models as profoundly as cloud computing reshaped them a decade ago. The transition from per-seat licensing to AI-augmented, outcome-based pricing will create winners and losers. For investors, the opportunity lies in distinguishing between the two while the market treats them as one.
Sources: Fortune / Bank of America, Nasdaq, Motley Fool, DeVere Group / J.P. Morgan / Goldman Sachs, Morningstar, CNBC
Related Reading
The 2026 tech volatility followed the disruption covered in DeepSeek Shock: China’s AI Challenger. For the AI investment cycle that preceded this correction, see AI Capex Boom: $600 Billion and Nvidia’s AI Supercycle. For the previous tech selloff, see Growth Stock Carnage: The 2022 Tech Wreck. For how the AI disruption narrative spread into private credit and triggered a redemption crisis at Blue Owl and beyond, see The Private Credit Crackup: Blue Owl, Redemption Gates, and the Liquidity Illusion. This selloff became the opening salvo of the Q1 2026 market correction, the worst first quarter for US equities since 2022. For how volatility continued to shape the financials complex, see our Q1 2026 bank earnings analysis.
Related Reading: see Tesla Q1 2026 and the Mag 7 capex reckoning. For continuing coverage on this theme, see our analysis of The Great Tech Divergence: Software Breaks While Silicon Soars.
Leave a Reply