The $660 Million Question: Is AI Eating Software's Lunch?

IBM just proved the budget-crowding-out thesis. But the market already priced it in. Now ServiceNow and IBM report Wednesday, and the whole SaaS narrative hangs in the balance.

On July 14, IBM dropped 25% in a single day. The company missed consensus revenue by $660 million, and the reason was a sentence no software investor wanted to hear: customers diverted their quarterly budgets to AI servers and memory, leaving software purchase orders unsigned. CEO Arvind Krishna described a sudden late-June pivot — clients scrambled to secure servers, storage, and DRAM before anticipated price hikes, and software contracts got left on the table.

The market had been sniffing this for months. Accenture is down 46% this year. ServiceNow is off 51% from its 52-week high, trading at $104. Salesforce has lost 35% of its value. Workday, 32%. The BVP Emerging Cloud Index, the benchmark for SaaS, is down 44% over five years, as Benchmark's Everett Randle noted on 20VC.

The question is no longer whether the AI budget-crowding-out thesis is real. IBM answered that. The question is whether the fear is already priced in, or whether Wednesday's earnings from ServiceNow and IBM will prove the damage is worse than anyone expects.

The Mechanism: How AI Eats a Software Budget

The argument has two layers.

The first is budget displacement. Enterprise IT spending is not infinite. When a company allocates $50 million to GPU clusters and AI inference, that money comes from somewhere. IBM's Krishna made it concrete: a global memory supply crunch drove enterprises to pre-buy hardware, and software purchase orders went unsigned. This is the same mechanism Citrini Research flagged in February: "Capital is rotating out of bits and into atoms. What happens to your 30x revenue multiple when new technology poses a potentially existential threat?"

The second, and more structural, is seat-based pricing disruption. Traditional SaaS charges per user. If an AI agent replaces five knowledge workers, the company needs five fewer Salesforce seats. The revenue base shrinks without anyone canceling a contract. As a16z's recent piece put it: "For the first time in history, humans are cheaper than software."

The market has drawn a sharp line. According to a16z's data, since ChatGPT launched in November 2022, Palantir — the AI-native transformation play — is up 79%. The SaaS basket is down 2.9%. That is not a rotation. That is a verdict.

The Split: Who Gets Hurt, Who Gets a Pass

Not all software is suffering equally. The market is sorting companies by whether AI is a threat or a tailwind.

Cybersecurity and data infrastructure are the clear winners. Palo Alto Networks is up 108% over the last 63 trading days at 85x forward earnings. Datadog is up 108% at 91x. Fortinet is up 96% at 47x. The logic is straightforward: every AI deployment needs security and observability. These are the picks and shovels of the AI gold rush, and they trade at AI-premium multiples.

The seat-based application SaaS companies are the losers. Salesforce trades at 11x forward earnings. Adobe at 8.5x. Workday at 11.6x. These are companies growing revenue at double-digit rates, throwing off enormous free cash flow, trading at multiples usually reserved for declining industrials. The 10x valuation gap between Palo Alto Networks (85x) and Salesforce (11x) is the market saying it knows exactly who gets disrupted. Whether it is right is the question.

The Revenue Reality: Growth Has Not Broken Yet

Here is the uncomfortable fact for the bears: the actual P&Ls are still printing double-digit growth, and companies are beating estimates.

Salesforce reported Q1 fiscal 2027 results in late May. Revenue hit $11.13 billion, up 13% year over year. Non-GAAP operating margin reached 34.8%. EPS came in at $3.88, a 24% beat against the $3.13 consensus. The company's Agentforce AI product crossed $1 billion in annual recurring revenue and processed 28.6 trillion tokens, up 152% quarter over quarter. Marc Benioff called it "the biggest growth opportunity since we brought CRM into the cloud."

ServiceNow reported Q1 2026 numbers in April. Subscription revenue grew 22% to $3.67 billion. Current remaining performance obligations, the best forward indicator, rose 22.5% to $12.64 billion. New logo annual contract value accelerated to 50% year over year. Now Assist, the company's AI product, is on track to exceed the $1 billion target for 2026. CEO Bill McDermott's message: "Customers are moving past experimentation into full-scale enterprise-wide AI investment."

These are not dying businesses. They are businesses in transition, and the market is pricing the risk that the transition fails.

The $500 Billion Time Bomb Nobody Is Talking About

There is a second, less visible risk that sits in private credit markets rather than public equities.

A June 2026 research paper from CrossVol Research, flagged by Larry Swedroe, documents that direct lending to software issuers grew from roughly $8 billion in 2015 to more than $500 billion by the end of 2025. Those loans were underwritten against EBITDA multiples of roughly 30x. Those multiples have since compressed to approximately 16x. The collateral value implied by the original underwriting has roughly halved without a single contractual default.

Bloomberg Intelligence identifies $280 billion of "interface-heavy app spending" at direct risk as AI agents reconfigure the application software industry. The Pluralsight case — equity to zero, BDCs still carrying senior debt near par — is the template for what happens when the music stops.

This is a credit-market risk, not an equity-market risk. But if business development companies start marking down SaaS loans, the cascade effect could hit equity valuations regardless of fundamentals.

The Wednesday Test

On July 22, after the market closes, both ServiceNow and IBM report earnings. Consensus EPS for IBM is $3.02; for ServiceNow, $0.40. The stakes are unusually high.

For IBM, the question is whether the July 14 preannouncement already aired all the dirty laundry, or whether the full Q2 results reveal deeper damage. The stock is trading at 16x forward earnings with an RSI of 27, deep in oversold territory. Any sign that the worst is over could trigger a sharp relief rally.

For ServiceNow, the question is whether the AI disruption narrative finally shows up in the numbers. The stock is down 51% from its 52-week high, trading at roughly 21x forward earnings. Q1 was solid: 22% subscription revenue growth, 22.5% cRPO growth, and the Now Assist business on a $1 billion trajectory. As Motley Fool put it: "July 22 Will Show Who's Right."

The real signal is not the earnings themselves. It is the guidance. Management teams see the pipeline before the market does. If ServiceNow's forward commentary on deal cycles, budget cycles, and AI displacement risk is cautious, the market will hear it. If it is confident, the market will hear that too.

What to Watch

Three things will determine whether the SaaS fear trade continues or reverses.

First, the ServiceNow print on Wednesday. A beat and raise on subscription revenue and cRPO would force the market to reconsider whether the AI disruption narrative is overshooting. A miss or a guide-down would validate every bearish thesis in the book.

Second, the second-quarter earnings wave for the broader SaaS cohort. Workday, Salesforce, and Adobe all report in late August. If they collectively show deceleration, the narrative hardens. If they hold the line, the valuation gap starts to close.

Third, the private credit shoe. The CrossVol paper is a warning, not a prediction. But if a major BDC marks down its software loan book, the signal will travel fast.

The market has already priced in a lot of bad news. The gap between the narrative and the numbers is now the widest it has been in a decade. Something has to give.

More research at bargo.ai/research.