The punishment has been fast and indiscriminate.
The iShares Expanded Tech Software Sector ETF (IGV) has declined more than 21% year to date in 2026 and sits roughly 30% below its September 2025 peak.
Software price to sales multiples have compressed from a 5.6 times average at end 2025 to approximately 4.2 times by mid March, levels last seen in the mid 2010s.
Estimates of lost market capitalisation run to two trillion dollars. The term that now circulates in every investment note is "SaaSpocalypse", shorthand for the fear that artificial intelligence agents do not merely compete with traditional software but make it structurally redundant.
To understand why that fear is both legitimate and overstated, start with the two most recent signal events: Oracle's Q4 FY2026 results and Accenture's Q3 FY2026 report, both released in the second week of June. Neither company delivered a disaster. Both stocks fell double digits. That disconnect is the entire story.
Table of content
- Why the Short-Term Pain Is Not Irrational
- Oracle: A Tale of Two Balance Sheets
- Accenture: The Consulting Model Under Siege
- Why the Long Term Still Belongs to Software
Why the Short Term Pain Is Not Irrational
The pricing of software is not purely sentiment. There are three overlapping structural forces compressing near term revenues that investors are correctly identifying, even if the market is overextrapolating them into perpetuity.
The first is seat compression. The per seat subscription model that built companies like Salesforce, Workday, Atlassian, and Monday.com assumed that revenue growth was anchored to headcount growth.
When a company hired more employees, it bought more licences. AI inverts that relationship. When agents execute workflows that previously required three analysts, customers require three fewer seats.
Workday itself announced layoffs of 8.5% attributed to AI efficiency gains, an AI vendor firing people because of the AI it sells. The irony is instructive.
Net revenue retention figures, which investors relied on as the gold standard of SaaS quality, are being hollowed out from below as seat counts contract even while surviving customers expand their AI tier purchases.
The second is budget redirection. The five largest hyperscalers — Amazon, Microsoft, Google, Meta, and Oracle — plan to spend between $660 billion and $690 billion on AI infrastructure in 2026, nearly double 2025 levels.
Approximately 75% of that sum targets AI compute and data centres. Corporate IT budgets are finite.
The marginal dollar that once went to a new enterprise software subscription now goes to an AI model API call or a GPU cluster. Traditional SaaS vendors are competing for budget against the very infrastructure they depend on.
The third is model substitution. AI agents can now perform tasks that dedicated SaaS tools were built to perform: project tracking, CRM updates, support triage, meeting scheduling, document review, report generation.
The critical distinction is that the agent does not sit inside the software. It replaces the need for the software.
When an AI agent creates a Jira ticket from a Slack conversation without a human touching Jira, the project management licence becomes overhead. This is not a hypothetical; it is being measured in Q1 and Q2 2026 earnings calls across the sector.
Oracle: A Tale of Two Balance Sheets
Oracle beat on both lines. Q4 FY2026 earnings per share came in at $2.11 against a consensus of $1.95, an 8% upside surprise.
Revenue reached $19.2 billion, just ahead of the $19.1 billion estimate, up 21% year over year. Full year FY2026 revenues hit a record $67.4 billion, with cloud revenues rising 39% to $34 billion.
Cloud infrastructure alone grew 93% in the quarter, powered by AI workload demand. The stock fell 10% in extended trading.
The market was not punishing Oracle for its operations. It was punishing the bill for those operations.
Capital expenditure for FY2027 is expected to reach $70 billion on a net basis. The company simultaneously announced plans to raise $40 billion through a combination of debt and equity financing, including a $20 billion share sale, after having already raised $43 billion in debt and $5 billion in equity in FY2026.
Free cash flow turned deeply negative in the December quarter. The arithmetic of aggressive AI infrastructure expansion is compressing near term returns even as the forward contract book has soared to $455 billion, up 359% in a year.
Accenture: The Consulting Model Under Siege
Accenture's story is structurally different and arguably more instructive.
The firm reported Q3 FY2026 revenues of $18.72 billion, growing 6% in US dollar terms but only 3% in local currency. EPS of $3.80 beat the $3.71 forecast by 2.4%.
The miss was on revenue — $60 million below consensus — and on new bookings, which came in at $19.3 billion against expectations for 7% growth.
The company also narrowed its full year FY2026 revenue growth guidance to 3% to 4% in local currency, cutting the top end of the prior 3% to 5% range.
The structural headwind is threefold.
- First, the US federal business, which Accenture flagged as contributing a 1% to 1.5% revenue drag has slowed sharply under budget rationalisation.
- Second, corporate enterprise clients are pausing large scale consulting engagements while they determine what their AI transformation should look like; spending money on a McKinsey or Accenture to redesign workflows feels redundant when the workflows might be redesigned by an AI agent.
- Third, and most importantly, Accenture's core labour arbitrage model is exposed to exactly the productivity displacement that AI promises. The consulting industry sold complexity for decades. AI threatens to commoditise that complexity.
Why the Long Term Still Belongs to Software
The bear narrative prices in disruption without crediting survival.
Gartner estimates that by 2030, 35% of point solution SaaS tools will be replaced or absorbed by AI agent ecosystems. The inverse of that figure — the 65% that survives — barely makes headlines.
More importantly, the companies that are omitted from that 65% are mostly horizontal, narrow function, low switching cost tools.
The companies that remain are the systems of record, the compliance platforms, the data repositories, the workflow orchestration layers. These are not footnotes in the software industry. They are its structural backbone.
Consider the competitive moat that two decades of enterprise software deployment have built.
A Fortune 500 company running SAP for its ERP, Salesforce for its CRM, and ServiceNow for its workflows has not simply purchased software. It has encoded its business logic, its approval chains, its exception handling, and its institutional knowledge into those platforms across years of customisation.
An AI agent can perform a task within that workflow. It cannot, in any practical near term timeframe, replace the workflow architecture itself. The switching cost is not a product feature. It is years of embedded human decisions.
As PwC notes in its 2026 analysis of software M&A, what is hard to replicate is not the code but the accumulated domain expertise, regulatory understanding, and customer relationships that make enterprise software sticky.
The pricing model evolution also matters more than it is being credited. The per seat model is not dying. It is transforming. Gartner reported that as of early 2026, 40% of enterprise SaaS contracts include outcome based elements — charging for a resolved customer ticket, a completed compliance review, or a processed transaction rather than a human licence.
That figure was 15% two years prior. This shift does not shrink the addressable market; it expands it. When a software vendor converts a labour cost into a software cost, the total contract value grows.
Procore, for instance, is experimenting with pricing its AI capabilities on construction dollar volumes rather than employee headcount, insulating its revenue from the seat compression that is plaguing horizontal tools.
ServiceNow is perhaps the clearest demonstration of the thesis. The company guided $15.5 billion in subscription revenue for 2026 at 19% to 20% growth, with a 32% operating margin and 36% free cash flow margins.
Rather than betting on a single AI model, ServiceNow has positioned itself as the workflow orchestration layer through which any AI model operates within the enterprise, capturing value regardless of whether customers use OpenAI, Anthropic, or Google's models.
That is not a company at war with AI. It is a company that has decided AI will flow through its pipes.
Microsoft's numbers provide the macro confirmation. More than 80% of Fortune 500 companies are already running workloads on Azure AI services. Copilot is now in full enterprise rollout with deployments in excess of 500,000 users at single organisations in healthcare.
The AI transition is not hollowing out Microsoft's software revenues. It is expanding them because the most defensible software companies are not fighting AI; they are becoming the governance, compliance, and orchestration layer on top of it.
