AI Killed $2 Trillion of Software. Jensen Huang Disagrees.

The Year the Market Decided Software Was Dead

On February 3, 2026, software investors got a day they now call Black Tuesday. Roughly $285 billion of software market cap evaporated in a single session - the S&P 500 software index fell 5.7%, its worst day since March 2020. Nobody had missed earnings. Nothing had been recalled. What happened was a thesis: AI agents had gotten good enough to do the work that software licenses are priced on.

The selloff had started three weeks earlier, when Anthropic launched Claude Cowork - described as "Claude Code without the code" - and it kept rolling for months. A second wave hit in April when Anthropic shipped Claude Managed Agents, a production stack for running agents without buying SaaS tooling. By spring, press tallies put the cumulative damage across the software complex at roughly $2 trillion. The sector's relative drawdown - about 22 percentage points behind the S&P 500 - was described by analysts as the worst software has ever experienced, exceeding the dot-com bust, the financial crisis, and the 2022 rate shock.

And standing in the middle of the rubble, repeating himself at every microphone he could find, was the CEO of the most valuable company in the world. Jensen Huang's verdict on the software-is-dead trade: "I think the market's got it wrong."

The stock he kept defending by name was ServiceNow - down, at its April low, 66% from its all-time high while still growing revenue 22%.

One of these two views is badly mispriced. We used Barebone to pull the tape on the entire software complex, ServiceNow's filings and price history, the analyst record, and every congressional disclosure that touched the stock - to figure out which.

The Wreckage

First, the damage. Here is the software complex year-to-date, against the S&P 500:

Ticker	Company	2026 YTD return
HUBS	HubSpot	-50.5%
MNDY	monday.com	-43.9%
TEAM	Atlassian	-41.0%
WDAY	Workday	-34.8%
CRM	Salesforce	-33.7%
ADBE	Adobe	-32.2%
NOW	ServiceNow	-30.1%
IGV	Software ETF	-12.7%
SNOW	Snowflake	+9.0%
-	S&P 500	+7.4%

This is not a correction. It's a repricing of an entire business model. Software traded at a forward P/E of 84x at the 2021 peak; by March 2026 the sector multiple had compressed to roughly 22x - within a point of the S&P 500 itself. The market is saying, in plain numbers: software is no longer a growth industry.

The logic of the bear case is seductive because it's simple. SaaS is priced per seat. Seats are humans. If AI agents do the work, the humans - and the seats - go away. And if an agent can assemble software on demand, why pay for somebody else's?

The Bear Case Deserves Respect

This isn't a strawman thesis from retail traders. It arguably started at the very top. Microsoft CEO Satya Nadella, back in December 2024, said the quiet part out loud: "SaaS applications or biz apps - the notion that business applications exist, that will probably collapse in the agent era." The business logic that lives inside today's apps, he argued, migrates to an AI tier that operates across every database underneath.

The most concrete version of the bear case came from UBS analyst Karl Keirstead, who downgraded ServiceNow on April 10 - cutting his target from $170 to $100 - and knocked the stock down 7.6% in a day. His evidence was unusually specific: over half of his enterprise customer conversations now included anecdotes of companies containing non-AI software spend, as AI infrastructure crowds out the rest of the IT budget. He cut his exit-2026 growth estimate for ServiceNow's contracted backlog from 20% to 16%. Notably, ServiceNow had been UBS's only Buy-rated application software stock.

Then the tape made its own argument. On April 22, ServiceNow reported a first quarter that beat on essentially every line - and the stock fell 17.8% the next day, the worst single day in company history. The complaints: backlog growth guidance decelerating, margin dilution from an acquisition, and several large Middle East deals delayed by the war with Iran (a 75-basis-point headwind, per the CFO). When a stock gets destroyed on a beat, the market isn't grading the quarter. It's grading the thesis.

The Most Important Man in AI Disagrees

Jensen Huang has now made the opposite argument at four separate venues in four months - at GTC in March, at ServiceNow's Knowledge conference in May, in TV interviews in between, and at GTC Taipei on June 1, where his keynote dismissing the agents-kill-software trade single-handedly sparked a sector-wide rally.

His core claim, delivered at Knowledge 2026:

"For the first time, service is software. Software is service, and the service industry is 100x larger than the software industry."

Unpack that. Companies pay software vendors tens of billions to track work - tickets, claims, onboarding, approvals. They pay humans trillions to do that work. Agents collapse the distinction: the moment software performs the task instead of recording it, software vendors stop competing for the IT budget and start competing for the labor budget. In Huang's framing, agents don't shrink software's market. They expand it a hundredfold.

But an agent can't just show up and start working, any more than a new hire can. It needs credentials, permissions, access to systems, and your company's rules. The agent is the worker. The platform is the office. And Huang has been explicit about which office he expects the Fortune 500 to use - at Knowledge 2026 he called ServiceNow "destined to be the best platform, the operating system of enterprise AI agents," and told the audience, "I came because I want my ServiceNow." He went further: factory robots, he said, will "run on top of ServiceNow, and you manage them just like employees."

This is not charity. NVIDIA and ServiceNow ship joint products - an autonomous desktop agent (Project Arc), a secure agent runtime, and a benchmark suite for enterprise agent work - and selling the agentic future is very much Huang's book. He is, however, the single best-informed person on earth about where enterprise AI spending is actually going. When he says the market has it wrong, it's worth checking the price.

Down 66%, Growing 22%

So what does the market's verdict look like against the company's actual numbers?

The stock peaked at $239.62 (split-adjusted) on January 28, 2025. By mid-April 2026 it touched $81.24 - a 66% drawdown. As of June 9 it sits at $106.97, still 55% below the high and down 30% this year alone.

Now the business over the same stretch:

Metric (Q1 FY2026, reported April 22)	Value
Revenue	$3.77B, +22.1% YoY
Subscription revenue	$3.67B, +22%
Contracted backlog (cRPO)	$12.64B, +22.5%
Total backlog (RPO)	$27.7B, +25%
Non-GAAP operating margin	32%
AI product ACV target for 2026	$1.5B - raised 50% from $1B
Fortune 500 penetration	more than 85%

Revenue growth didn't slow during the crash - Q1's 22.1% came in faster than either of the two quarters before it. Customers paying $1M+ annually for ServiceNow's AI products grew 130% year over year. Management raised its AI bookings target by half a billion dollars during the selloff. And in April the company closed the $7.75 billion acquisition of Armis - its largest ever - months ahead of schedule, betting that securing AI agents and connected devices becomes part of the same platform sale.

The disconnect lives in the multiple, and it's worth being precise about, because both bulls and bears quote it selectively. On trailing GAAP earnings, ServiceNow trades at 64x - roughly double the S&P 500's 32x, and the honest version of "this is still an expensive stock." On forward earnings, it's 24.7x against the index's 21.1x - a 17% premium for a business growing revenue three times as fast. And on enterprise-value-to-forward-revenue, the multiple compressed from a historical average around 13x to 5.6x at the April lows. The same company, described three ways. What actually changed in 2026 wasn't the business. It was which description the market chose to believe.

Six Weeks of Whiplash

If you want proof this stock now trades on narrative rather than numbers, look at what it did in the six weeks after the April crash.

From $87 in mid-May, ServiceNow ripped +56% to $135.86 by June 1. The catalysts were all thesis-level, not company-level: Bank of America initiated coverage with a Buy ("ServiceNow is so embedded in the workflow… they could actually turn it into revenue growth" - the stock jumped 8.8% that day). Snowflake posted a +36% single day - its best since IPO - after showing AI was accelerating its demand. ServiceNow announced agentic-AI data partnerships with Experian and Boomi. Then Huang's Taipei keynote added another +9.2% in a session.

And then it gave a fifth of it back. From June 1 to June 9 the stock fell 21% - profit-taking, renewed worries about cloud costs and Armis margin dilution, and a hot May jobs report (+172K vs ~85K expected) that pushed rate-cut hopes out and hit every long-duration growth multiple.

+56% up, -21% down, in six weeks, on essentially no new fundamental information. That's not a market analyzing a company. That's a market fighting over a story.

What the Smart Money Sees

Two groups worth checking: Congress and the sell side.

Since the start of 2025, eleven members of Congress have disclosed 31 separate purchases of ServiceNow stock, per public disclosure aggregators. The largest came after the crash was well underway: Representative Tony Wied disclosed a buy of $1,000,001 - $5,000,000 on March 9, 2026. Ro Khanna, Byron Donalds, Josh Gottheimer, and Senator Markwayne Mullin all disclosed 2026 purchases as well. The usual caveats apply double here: disclosure ranges are wide, some of the same names also sold, and congressional trades are a noisy signal. But noisy isn't the same as empty - these are people whose day job is seeing policy and procurement before the market does, and they kept buying through the drawdown.

The sell side, meanwhile, never capitulated: of 48 analysts, 34 rate it Strong Buy and exactly one rates it Sell. The average target is $141.86 - 33% above the current price. The street-low target is $85; the high is $236.

The most interesting datapoint is the low one. In April, the stock traded below the lowest price target on Wall Street - under $85, beneath even the number published by its most bearish analyst. Markets occasionally hand you a clean capitulation marker. A stock trading under its street-low target while growing 22% is about as clean as they come.

Where the Thesis Breaks

The bull case is not a free lunch, and intellectual honesty requires the other ledger.

If agents flatten the application layer, distribution stops being a moat. ServiceNow's 85% Fortune 500 penetration is only valuable if the platform remains the place work happens. Nadella's scenario - business logic migrating to an AI tier that treats every app as a database - is precisely the world where incumbency stops compounding. Microsoft has both the incentive and the install base to force that experiment.

The budget evidence is real. UBS's observation - that half of enterprise conversations now involve containing non-AI software spend - is the most concrete bearish datapoint anyone has produced this year. AI infrastructure is eating IT budgets, and ServiceNow sells to the part being eaten. A 16% growth exit rate instead of 20% doesn't sound catastrophic, but at a premium multiple, deceleration is how 30% drawdowns become permanent.

The trailing multiple still demands the story come true. At 64x trailing GAAP earnings, ServiceNow's reported profits have years of catching up to do. The forward multiple only looks reasonable if the estimates hold - and the whole bear case is that they won't.

And Jensen is talking his book. NVIDIA sells the chips agents run on and co-sells products with ServiceNow. He may be the best-informed voice in AI; he is not a disinterested one.

What This Means

Strip the noise and this is a single, testable disagreement. The market is pricing software - ServiceNow included - as the casualty of the agent era. Huang is arguing software platforms are the infrastructure of the agent era, competing for a labor budget 100x the size of the IT budget. Both views cannot be right at a 5.6x revenue multiple.

The useful move isn't picking a side on vibes. It's knowing what would prove each side right:

The agent-era platform test. For any software company in the blast radius, ask three questions. Does it have distribution - is it already wired into the Fortune 500's systems and contracts? Is it cross-departmental - a system of record agents must touch to do real work, not a single-team tool? And does it own governance - the permissions, audit, and rules layer that lets a CISO say yes to a thousand agents? ServiceNow clears all three. That's Huang's entire argument compressed - and it's also why the companies that fail those tests (single-department tools, seat-priced point solutions) deserve their multiple compression.

The signals to watch. Backlog growth (cRPO) at the July 22 earnings - the company guides around 19.5% in constant currency; UBS says 16% by year-end. The $1.5B AI bookings target - if it's raised again, agents are monetizing; if it's quietly missed, the bears were early, not wrong. And the sector's forward multiple against the S&P 500 - software trading at an index multiple is either the buying opportunity of the decade or the new permanent regime.

For thirty years, "nobody got fired for buying software" was the safest trade in growth investing. In 2026, the market decided it was the most dangerous. The most important man in AI spent the entire spring saying the market has it backwards - and the data, for now, says the business is growing like he's right while the stock is priced like he's wrong. The next two earnings reports decide who blinks.

---

Data: Barebone | Sources: ServiceNow Q1 FY2026 results (SEC filings), ServiceNow Knowledge 2026 and NVIDIA GTC 2026 remarks, UBS and Bank of America published research, public congressional trading disclosures | Prices as of June 9, 2026 close