ServiceNow beat every number and fell 27% anyway

I renewed my Jira license last month. Same price. Same features. But our team had already moved three workflows to an AI agent. We were paying for seats nobody sat in.

That's what I kept coming back to when I saw what happened to ServiceNow on Thursday.

ServiceNow reported Q3 earnings on October 28. Subscription revenue came in at $3.99 billion, up 21% year over year. The company beat its own guidance on every line. It raised its AI contract-value target from $1.5 billion to $1.7 billion.

The stock fell 27% in a single day. The worst day in ServiceNow's history.

Why? Because customers renewed at 98%… but seat growth slowed. The renewal rate looks healthy. It cannot see the freeze underneath.

Gartner has a term for it: agentic arbitrage. An AI agent does work across several enterprise applications without a human touching each one. Every agent that picks up a task is a seat that doesn't get renewed next year.

This is not just ServiceNow. Salesforce is down 13% in 2026. Workday is down 9%. Atlassian reported its first-ever decline in enterprise seat counts. Monday.com cut 630 jobs—20% of its staff—in July, after rebranding itself as an “AI Work Platform.”

The BVP Nasdaq Emerging Cloud Index, the benchmark for cloud software stocks, is down roughly 29% from its 2021 highs. The S&P 500 Software Index fell 5.7% in a single session on February 3, the day after Anthropic shipped a legal plugin for Claude Cowork—an AI desktop tool that handles the kind of work SaaS companies built their businesses on.

Wall Street calls it the SaaSpocalypse. I don't love the name. But the math is right. Enterprise software stocks have lost roughly $2 trillion in market value over the past twelve months.

So here is what I find interesting. The per-seat license was the foundation of the cloud software business. Every company in the sector priced its product the same way: per user, per month, forever. Growth meant adding seats. Seats meant headcount. And headcount kept growing… until it didn't.

Now AI agents do work that used to need a human login. Salesforce launched its own AI agent platform, Agentforce, with “Flex Credits” at about 10 cents per action. That's a different model. It means the company that wins the AI transition still loses per-seat revenue on the way there.

Both sides of this trade are losing right now. The companies building AI are spending themselves into negative cash flow. The companies AI replaces are watching their seat counts shrink. I have no idea which side recovers first. But both can't win at the same time. That's the anomaly worth watching.

The AI disruption eating through enterprise software runs on infrastructure that somebody has to pay for. This week we saw what that bill looks like.

Alphabet reported third-quarter 2026 earnings on October 28. Revenue rose 25% to $127.6 billion. Google Cloud more than doubled, reaching $30.7 billion. On paper, a strong quarter.

But capital expenditures hit $51.4 billion in a single quarter, up 114% from a year ago. Operating cash flow fell 12.6% to $42.3 billion. So the company spent $9 billion more than it earned. Free cash flow came in at negative $9.07 billion—compared with positive $24.5 billion a year earlier.

That is a $33.5 billion swing in cash generation in twelve months.

Full-year 2026 capex guidance stays at $195 to $205 billion, and management told analysts to expect a “significant increase” in 2027. Google Cloud's backlog sits at $561 billion, most of which Alphabet expects to recognize within two years. So the spending isn't slowing down.

The stock fell 12.6% on October 29—the largest single-day drop in Alphabet's history.

The companies building AI are burning cash. The companies AI replaces are losing customers. Both sides of the trade are underwater at the same time.

And here is where AI-driven volatility meets a tighter plumbing system.

The Treasury General Account—the government's checking account at the Fed—crossed $1 trillion on October 28 for the first time in months. It reached $1.016 trillion, up $70.5 billion in a single week, as the Treasury settled a heavy schedule of bill and coupon auctions ahead of month-end.

Every dollar that flows into the TGA has to come from somewhere. Normally some of it drains from the Fed's Overnight Reverse Repo facility. But that buffer is gone. The ON RRP balance sat at $176 million on October 28—essentially zero.

So the TGA rebuild drained commercial bank reserves dollar for dollar. Weekly average reserves fell to $2.894 trillion, per the Fed's H.4.1 release.

A February 2026 research paper estimated that the banking system needs more than $3 trillion in reserves to operate comfortably under the current framework. We are below that number now.

The stress shows up in funding rates. SOFR—the overnight rate banks charge each other—traded 4 to 5 basis points above the interest rate the Fed pays on reserves. That premium has widened over the past two weeks. Discount Window borrowing rose $566 million to $9.5 billion.

Wednesday comes the Treasury's Quarterly Refunding Announcement, which will set coupon auction sizes for the rest of the quarter. If the Treasury keeps auction sizes large, the drain continues. We'll see.

Three more things worth keeping track of this week.

Christine Lagarde resigned as ECB president on October 29. She will step down on February 7, 2027—nearly nine months before her term was scheduled to end in October. She said she wants “freedom of speech” and “freedom of expression.” The resignation opens a succession fight at the ECB right as the euro zone runs a 2.50% deposit rate, France's fiscal deficit sits above 5% of GDP, and bond yields across the continent remain elevated. The last time the ECB changed presidents mid-crisis was 2011, when Mario Draghi replaced Jean-Claude Trichet just months before the “whatever it takes” pivot. The next president will inherit a harder set of choices. We'll track the shortlist.

About $76.6 billion in CMBS loans mature by the end of 2026, per Trepp, with 36% likely to face refinancing friction. Office is the worst. Trepp's Q3 data shows $22.2 billion in office CMBS debt hitting hard maturity in 2027, and three-quarters of Class A office loans need fresh equity to refinance at current rates. We covered office CMBS delinquency rates earlier this fall. The maturity schedule is the next chapter. When loans stop extending, losses stop being hypothetical.

The FOMC held rates at 3.75–4.00% on October 28 with a 10–2 vote. Governor Lisa Cook and Cleveland Fed President Beth Hammack both dissented in favor of a quarter-point rate hike. That's the first time since 1992 that a sitting Fed governor and a regional bank president dissented together on the same side. Cook has been at the center of the Trump administration's legal efforts to remove her from the Board. Some economists called her dissent a “protest vote.” Regardless of the motivation, two hawkish dissents six days before midterm elections is not a normal event. We'll see.

I keep a list of companies that once seemed permanent.

In 1999, Fortune magazine named Siebel Systems the fastest-growing company in the United States. Deloitte recognized it for 782,978% revenue growth over five years. Siebel sold customer relationship management software—CRM, the same category Salesforce dominates today—and by 2002, it held 45% of the global CRM market.

Revenue passed $2 billion. The stock reached about $119 per share. Tom Siebel, the founder, was on magazine covers. The company's product was in every large enterprise on earth.

But the product ran on-premise. It required armies of consultants to install. It cost millions upfront. And a small company in San Francisco had a different idea.

Salesforce launched in 1999 with one pitch: CRM in the browser, paid per seat, per month. No servers. No consultants. No upfront cost.

Siebel's management said the cloud model wouldn't work for serious enterprise buyers. They said large companies would never trust their data to someone else's servers. They said the product was too simple. That it lacked features. That it was a toy.

The stock fell more than 90% from its peak. Revenue stalled. In September 2005, Oracle bought Siebel for $5.8 billion at $10.66 per share—roughly a tenth of what the company was worth five years earlier.

The per-seat, cloud-based SaaS model that replaced Siebel went on to create $2 trillion in market value. Salesforce alone is worth nearly $200 billion today.

And now the model that killed Siebel is facing its own version of the same disruption. AI agents don't need seats. They don't log in. They don't click through menus. They do the work that per-seat software was designed to support, and they do it at 10 cents per action instead of $125 per user per month.

Siebel's managers said the threat was a toy. Salesforce's managers say Agentforce is the future—and they are building it themselves. That's smarter. But it still means the old revenue model has to die for the new one to grow.

In 1999, the fastest company in America sold on-premise software for millions per deal. By 2005, it sold for pennies on the dollar to a rival. The thing that disrupted it was the per-seat cloud model. And the per-seat cloud model is now what's breaking.

I have no idea how long this takes. We'll see.