Software stocks are in a bear market. Not in a “rotation” or a “correction” — a real, sustained decline that has wiped out two years of gains and got a name on the trading desks: the SaaSpocalypse. The iShares Expanded Tech-Software Sector ETF has had its worst stretch since 2008. Adobe is down ~32% year-to-date. Salesforce is down ~31%. ServiceNow is down ~25–30%. Even Microsoft has dropped 23%.
Layer two stories on top of that — tariff chaos rattling Wall Street and a Goldman Sachs strategist publicly comparing the future of software to the future of newspapers — and you have the conditions for the most uncomfortable conversation in tech: what if the AI boom, working exactly as advertised, triggers a white-collar recession?
This article walks through what is actually happening to software valuations, why the AI-disruption thesis is suddenly being taken seriously, and what the warning signs look like for engineers and operators.
What the numbers actually say
The 2026 selloff hit pure-play software harder than any other tech segment. A few of the headline data points:
- SaaS index down nearly 40% year-to-date.
- Adobe: −32%
- Salesforce: −31%
- Microsoft: −23%
- Shopify: −26%
- ServiceNow: −25 to −30%
- iShares Expanded Tech-Software Sector ETF: worst stretch since the 2008 financial crisis.
This is happening to companies still posting double-digit revenue growth. The compression is in multiples, not earnings — markets are repricing what software companies are worth as a function of revenue, not punishing a temporary blip in cash flow.
That distinction matters. A multiple compression of this size means investors have changed their long-term view of the asset class itself.
The Goldman warning
Ben Snider at Goldman Sachs published a note in early 2026 arguing that the software selloff may be “the end of the beginning” of the decline, not the beginning of the end. The analogy he reached for was unusually direct: software is now in the position newspapers were in around 2007. Disruption that takes a decade to fully play out can cause prolonged stock-price declines well before revenue meaningfully turns.
That framing is what gives traders pause. Newspapers in 2007 still had revenue. They still had subscribers. They still had what looked like reasonable financials. Their stocks were already down 40–60% from peak — and continued to decline for another decade.
If software in 2026 is at the 2007 newspaper point of the cycle, that implies the bear market is not over.
Why “AI working as advertised” is the bear case
The standard bull case for AI was: AI lifts software. Productivity tools become smarter, customers do more, ARPU goes up, the SaaS flywheel accelerates.
The bear case being priced in now is the inversion of that:
1. Agentic AI removes the UI
A traditional SaaS product’s moat is partly the user interface — the dashboard, the workflow, the muscle memory operators build over years. Agentic AI doesn’t use UIs. A coding agent does not need a Salesforce dashboard; it makes API calls. A customer-service agent does not need a Zendesk console; it reasons over the same data structures directly. As more work is delegated to agents, the surface area where SaaS UI moats matter shrinks.
2. Customer count is at risk, not just per-seat pricing
Per-seat SaaS pricing assumes a roughly linear relationship between headcount and software seats. If AI agents replace meaningful slices of white-collar headcount, customers buy fewer seats. The Oracle layoffs (30,000 cut on March 31 to fund AI capex) are the most visible instance of a wider pattern — and every cut seat is a SaaS line item that doesn’t renew.
3. The “AI premium” is now an AI discount
In 2024–2025, software companies could earn a multiple expansion just by mentioning AI on earnings calls. In 2026, the same companies are being asked the harder question: how does your business look when the AI you’re integrating starts replacing your customers’ employees? If the answer is “we lose seats,” the premium becomes a discount.
The tariff overlay
On top of the structural AI thesis, 2026 has been a uniquely turbulent year for trade policy. “Pure tariff chaos” is how multiple analysts have described it: Supreme Court rulings on tariffs, EU trade-deal reversals, and shifting policy from week to week. For software companies that earn 30–60% of revenue internationally, that uncertainty is a direct headwind on multiples — even before AI is in the picture.
Tariffs do not directly tax SaaS revenue. But they raise the cost of GPUs, hardware, and the data-centre buildout that software companies are now buying into for AI. They also raise the discount rate the market applies to long-duration cash flows. Both pull software multiples down.
The white-collar recession scenario
The scenario the bears are quietly modelling looks like this:
- AI agents continue to mature through 2026–2027, with measurable substitution for entry- and mid-level white-collar work in coding, customer support, sales operations, finance, and legal review.
- Large enterprises follow Oracle’s playbook — layoffs explicitly framed as “funding AI,” cutting 10–20% of headcount in non-AI divisions.
- Reduced white-collar employment causes a consumer-spending slowdown, particularly in services and discretionary categories that rely on professional-class income.
- Reduced spending feeds through to S&P 500 earnings, triggering a broader correction.
- The same AI agents that drove the productivity gains become the catalyst for a recession.
It is not a base case for most firms. But it is now plausible enough that strategists are publicly modelling it. That is itself a regime change from a year ago.
What this means if you build software
Three concrete things, regardless of whether the worst case plays out.
1. The “more seats” growth story is closing
If your business model assumes linear customer-headcount growth at your customers, model what happens if that flatlines or reverses. The companies that will do well are the ones whose pricing aligns with value delivered, not seats provisioned — usage-based, outcome-based, agent-call-based.
2. Build for agents as customers, not just humans
If agentic AI is increasingly the consumer of your service, make sure your service is something an agent can use well. Stable APIs, structured outputs, well-described tool schemas, predictable error semantics. Products optimised entirely around a human dashboard are increasingly stranded.
3. Watch capex announcements
The companies cutting hardest now are the ones spending most on AI infrastructure. That trend continues. Capex commitments in the tens of billions are forward indicators of the next layoff round at the same firm — not always, but often enough to be worth tracking.
Is this the bottom?
Honest answer: nobody knows. The same conditions that made this selloff feel structural — AI eroding software moats, white-collar substitution, capex-funded layoffs — also create the possibility of a sharp rebound if any of the assumptions are wrong, or if AI productivity gains turn into new software businesses that didn’t exist before.
What is clearly true is that the easy “software always wins” thesis of the last decade is not the operating assumption anymore. Investors, executives, and engineers are all repricing what software is worth in a world where the AI it integrates can also replace the people buying it.
That is the conversation now. Whether or not you find the bear case persuasive, the multiples have already moved.