Q3 Thesis: The Data Toll Booth

Among media and software companies, the line between Q3 winners and losers won't be drawn by sector — it'll be drawn by who owns a proprietary data corpus that AI companies need and customers will pay to have turned into intelligence. Two new monetization rails have opened in the last 1-2 years, getting paid by foundation labs for training/grounding data, and getting paid by customers for AI-powered intelligence built on data nobody else has. Companies sitting on top of both are becoming toll booths. Companies with neither are becoming free inputs.

Why this is happening

A two-sided market for proprietary data has formed faster than most generalist investors have repriced for it.

On the media side, the licensing economy is real and growing, not theoretical. News Corp signed a deal worth more than $250 million over five years with OpenAI and roughly $50 million a year with Meta. Reddit's licensing agreements with Google and OpenAI are collectively worth over $200 million, and the company has become the most-cited source across AI models. Three times more than Wikipedia . This is why Reddit is now pushing both partners toward dynamic pricing that scales with how essential its content proves to be in AI outputs, rather than flat annual fees. The whole market is shifting in that direction: live-access and grounding deals, as opposed to one-time training payments, have grown from roughly 2 in 2023 to a projected 34 in 2026 — meaning the best-positioned publishers get paid on a recurring, usage-linked basis, not a single check.

The flip side is just as stark. Referral-traffic collapse from AI search has hit small, undifferentiated publishers roughly three times harder than large ones. If your archive is generic and interchangeable, AI labs simply take it as free training substrate while your search and social traffic evaporates.

On the software side, the same logic is playing out around what's being called the "SaaSpocalypse." In February 2026, roughly $285 billion in software market value was erased in a single session, with names like ServiceNow, Salesforce and LegalZoom down repricing lower, as investors priced in agents replacing per-seat tools wholesale. But the dispersion underneath that selloff is the actual signal: a16z points to companies like Bloomberg, Abridge, OpenEvidence, and VLex as proof that proprietary, hard-to-replicate data becomes more valuable, not less, once AI makes it possible to do far more with it.

Why Q3 specifically

Two things converge this quarter. First, licensing renegotiations (Reddit/Google, Reddit/OpenAI, and the broader shift toward usage-based "grounding" deals) are landing in real time, and will start showing up as a distinct, durable revenue line rather than a one-time headline. Second, Q2/Q3 enterprise software earnings are the first real test of whether "data moat" companies actually hold retention and pricing power through the agent-disruption scare, versus point solutions that don't. The market is currently pricing this distinction crudely — punishing "software" and rewarding "AI-native" as categories, rather than discriminating on the one variable that actually matters: who owns data nobody else can get. That mispricing is the trade.

The one-line version

Q3 separates companies that are getting paid to feed the AI economy from companies that are feeding it for free, and the variable that determines which side you're on isn't "media vs. software," it's whether you own data nobody else can get.