Forty-three public software companies reported into the same AI macro and got four different answers. The split wasn’t random — and it wasn’t about growth. It was about whether AI arrives in your product or in your customer’s capex line.
In March we argued that AI would enhance enterprise software rather than replace it — and that the dividing line would be the domain moat. That was a thesis. Q2 2026 is the first earnings season in which management teams named AI as the direct cause of the quarter, on the record, on both sides of that line. This page is the audit: what forty-three companies disclosed, what the market paid for it, and what it changes for a limited partner and for a founder.
Between July 15 and August 12, thirty-nine software companies and four AI-capex bellwethers reported into an identical macro environment. Every one of them faced the same question from the same customers: what is AI going to do to my software budget?
The answers spanned sixty points of share-price reaction. Atlassian rose 35% on a quarter where AI deepened its platform. IBM fell 25% on a quarter where its customers diverted budget toward AI infrastructure. Same month, same sector, same macro. What separated them was not growth — it was position.
Approximate single-day move on the earnings print, in percent. The market read the split immediately: companies that are the AI platform were rewarded; companies whose customers’ budgets were frozen or raided were punished. Direction is carried by the sign as well as the colour.
This is the detail that makes the season more than noise. If the market were simply repricing decelerating software, the losers would be the slow growers. They weren’t.
| Company | Revenue growth | Stock | What the market was actually pricing |
|---|---|---|---|
| PEGA | +9% | −17% | Cloud engine fine (ACV +22%); the new-business engine seized. Total ACV +7%. |
| IBM | +1% | −25% | Software demand not dead — diverted. Budget raided for GPUs ahead of software. |
| DDOG | +36% | −19% | Beat and raise. Largest customer — an AI lab — is reducing usage after renewing. |
| TTD | +3% | −22% | Fully AI-native and still missed. AI-native alone earned no premium. |
| RNG | +5.9% | +25% | Slowest grower here, biggest gainer. The AI layer re-rated the whole story. |
| TEAM | +28% | +35% | Cloud re-accelerated to 31%; AI monetizing indirectly through seat expansion. |
The tell: the market did not price how fast you grew. It priced whether AI was adding to your number or subtracting from your customer’s budget.
The selloff of last winter has resolved into a sorting mechanism, not a sector-wide decline. Public software is no longer being marked as one asset class — it is being marked on AI position. That is the same axis along which a private vertical software portfolio is underwritten, which makes this season the first genuinely comparable read on our marks since the doctrine was published.
Growth alone stopped being the story this quarter. Two companies with similar numbers got opposite outcomes based on where AI sat in their model. Before your next board deck, answer the question the public market is now asking out loud: when your customer builds an AI budget, are you a line in it, or the line it comes out of?
Every customer in this cohort felt the same thing: uncertainty about what AI will cost and what it will replace. That single pressure resolved three ways, and which one you got was decided by whether you were the platform capturing the AI spend or the vendor whose budget it came out of.
Buyers can’t price the total cost of AI-era software, so they stop. It hits net-new and expansion; renewals survive. This is the horizontal-tooling risk.
Demand isn’t dead — it’s diverted. Dollars flow to GPUs and infrastructure first, deferring software deals to later quarters. A timing problem, not a demand problem.
When you are the agentic layer or the modernization prerequisite, AI budget lands on you. Expansion, higher ARPU, bigger deals, rising retention.
Management said this out loud, which is what makes the quarter unusual. Pegasystems described the freeze in plain language; ServiceNow and SAP described the opposite effect, from inside the same macro.
“People are just confused… it’s really almost a max confusion moment. This cost uncertainty is leading many organizations to freeze.”
“I have not seen any change [in cycles]. If anything, it’s on the positive.”
SAP put the mechanism most precisely: customers told them that “AI is going nowhere” without landscape and data modernization first — so AI pulls the ERP deal forward. When you are the prerequisite, AI is a demand generator. When you are optional, AI is the reason the deal slips a quarter.
These three states are a portfolio-level diagnostic, not a market opinion. Every software company we hold sits in exactly one of them, and the assignment is knowable from the workflow rather than from a forecast. Freeze risk concentrates in horizontal tooling; our exposure is vertical and workflow-embedded — the pull-forward column of this table.
You can move columns, and the move is not a rebrand. Freeze happens to vendors selling into an open-ended “what’s our AI strategy” decision. Pull-forward happens to vendors whose AI is attached to a specific outcome the customer can already price — days to lease, fraud caught, deals closed. That attachment is the work.
IBM said its customers’ budget was being diverted toward “a capex-sensitive area of the portfolio.” That dollar does not vanish. It is traceable, and this season traced it: it lands in cloud and GPU infrastructure first, and the market rewarded precisely the operators where that spend was visibly converting into revenue.
Read those four together and the market’s rule is legible: it rewards visible AI monetization and punishes open-ended AI spend — regardless of how fast the top line is growing. Amazon and CoreWeave were paid for capex that converts. Meta and SpaceX were penalized for capex that hasn’t yet.
Which puts application software in a specific position. It is not the loser of the reallocation — it is the deferred leg of it. The infrastructure gets bought first because it has to exist before anything can run on it. The software that runs on top is where the return is actually realized, and it is bought second.
The bear case last winter was that AI capex permanently crowds out software spend. The evidence says re-timing, not destruction — and the infrastructure leg is being built out at a pace that implies the application leg behind it. That is a duration argument for a fund with a five-to-seven-year hold, not a break in the thesis.
If deals slipped this year on “we’re spending on AI infrastructure first,” that is a sequencing objection with a natural expiry, not a lost deal. Track those separately from true losses in pipeline review, and build the follow-up cadence around the customer’s infrastructure timeline rather than your quarter.
The most common objection to the Enhancement Doctrine in March was that AI-in-software was a narrative without a number. That objection is now closed. Four companies disclosed AI revenue at four different scales, and none of them are pilots.
The RingCentral number is the one worth sitting with. A business growing under 6% was the largest gainer in a forty-three-name cohort because roughly an eighth of its ARR is now AI. The market is not paying for AI ambition. It is paying for AI that has a revenue line next to it.
There is now a public comparable set for the question “what is the AI layer worth inside a software business.” AI-validated names carry roughly a two-to-three-times multiple premium over AI-exposed laggards — ServiceNow near 7× revenue against Pegasystems near 2.5×. That spread is the quantified version of the doctrine, and it is the axis our portfolio is positioned along.
Disclosure is itself a value driver. Companies that could name an AI number were re-rated; companies with AI on the roadmap were not. Instrument the AI line before you need to defend it — AI ARR, AI attach rate, AI-influenced retention — because a diligence conversation eighteen months from now will open with that question.
This is the doctrine’s own test, and it deserves an honest answer rather than a favorable one. The claim in March was that vertical software — software with a domain moat — would be insulated from AI disruption and would capture AI upside. The season split that claim in two, and only one half generalized cleanly.
No vertical name in the cohort froze. Not one. AppFolio grew 19% with operating margin at 27.1%, up 90 basis points, on AI a property manager can measure: 5.7 days faster leasing, 69% more fraud caught. Manhattan Associates grew cloud 26% with RPO up 23% on record bookings. Procore grew 16% with RPO up 24% and shipped agentic products on consumption pricing.
“Operators are embracing AI that works because it knows their business and drives real performance outcomes.”
Insulation is not the same thing as re-rating, and this is where the doctrine picks up a caveat it did not have in March.
Construction. Agentic products on consumption pricing, cash-flow inflection, stock +2.3%. The template works when AI is a priced product.
Vertical fintech. Steady and freeze-insulated, but AI packaging is still to be decided. Stocks flat, multiples at multi-year lows. Upside promissory.
Government. SaaS revenue +21.7%, but total ARR growth halved to about 8% with no AI monetization story. Steady core, de-rated multiple.
And the sharpest correction to the March thesis: AppFolio — the cleanest vertical-SaaS template in the cohort, a beat and a raise with margin expansion and outcome-visible AI — is down roughly 30% year to date. The insulation is real. The market has not paid for it yet.
Two things at once, and both matter. The downside protection in the thesis is confirmed by disclosure — vertical workflow software did not freeze in the worst AI-budget quarter on record. And the upside is not yet in public prices, which is the definition of an entry point for private capital buying the same characteristics. AppFolio down 30% on the year, on a beat and a raise, is the opportunity stated plainly.
Being vertical protected everyone here. Being vertical re-rated only the companies that turned AI into something a customer buys. Tyler is the cautionary case: a strong core, a real domain moat, and no AI monetization story — steady, and de-rated. Differentiation has to be monetized, not merely built.
Early in the season it looked as though the whole cohort was rejecting token-based pricing. Across all twenty-four software names, the transcripts do not support that. There is no convergence.
An explicit, on-record move to outcome pricing. ServiceNow: “Customers aren’t paying us for tokens, they’re paying for resolutions.”
AI folded into existing seats and tiers as an uplift, with inference cost absorbed as cost of goods. The quiet majority position.
Where usage is the value, metering it is aligned rather than opaque. Q2 Holdings is openly weighing token-inclusive billing.
The new lesson is in what happened to the company that moved. HubSpot re-based its AI pricing on outcomes — credits per resolved ticket and per qualified lead, “pay when the agent works” — and reported a healthy quarter: revenue up 20%, margins up. The stock fell about 19%, because outcome pricing elongates sales cycles and resets near-term seat growth even as it aligns price to value.
Datadog supplied the mirror image: consumption can go down. Its largest customer, a leading AI lab, reduced usage after renewing — because AI had made them more efficient. A 36% beat and raise fell about 19% on that disclosure. AI does not only add to a meter.
Pricing model is now a diligence line item with a measurable market consequence, not a product detail. The HubSpot print prices the transition risk directly: even a correct pricing move can reset growth for several quarters. We underwrite that transition explicitly rather than assuming a clean step-up.
The takeaway isn’t “everyone rejects tokens” — it’s that you get to choose, and the transition itself is the risk. Where your customer can see the value in what they consume, metering is honest. Where they cannot, price the outcome. Either way, sequence the change deliberately: it will cost you growth optics before it pays.
AI has to land somewhere in a corporate budget. This season showed where: infrastructure first, and the software that makes it useful second.
The panic assumed AI would eat enterprise software. Forty-three companies have now reported into that assumption, on the record, and it did not happen. What happened instead is that AI became a sorting mechanism. Companies that are the place AI does its work — the agentic layer, the modernization prerequisite, the workflow an industry runs on — captured budget. Companies selling into an open-ended AI strategy decision watched their deals elongate.
Vertical software sits structurally on the right side of that sort. It is embedded in a workflow the customer cannot pause, attached to outcomes the customer already prices, and carrying data no general model has. Not one vertical name in this cohort froze. That is not a forecast — it is a disclosure.
The honest caveat is that the market has not finished paying for it. AppFolio ran the doctrine’s playbook to the letter and is down about 30% on the year. Insulation is priced in; upside is not. For an operator that is a mandate to make the AI value legible and monetized. For a capital allocator it is a window.
The SaaSpocalypse is repricing horizontal software while cash buyers pay up for what we own. Median public SaaS trades near 3.4× revenue; AI-positioned vertical players are transacting at 6–8× ARR, and roughly 72% of software M&A now references AI in the target. Constellation Software — the most disciplined software buyer in the world — deployed $732M in a single quarter buying companies that look like ours. The headlines are the wildcatters. We own the pipelines, and the toll is still getting paid.
Four questions this season says you should be able to answer by your next board meeting. One: when your customer builds an AI budget, are you in it or under it? Two: what specific outcome does your AI move, in your customer’s own units? Three: what is your AI number, and can you disclose it? Four: if you change pricing, how many quarters of growth optics will the transition cost? The companies that could answer these were re-rated. The ones that could not were not.
Golden Section invests in and lends to capital-efficient B2B vertical SaaS. We read every software earnings transcript this season and push the read-throughs into our portfolio.
Get in touch