There's a comforting story being told inside every B2B data company's all-hands right now. It goes something like this: AI is a tailwind. Agents will need our data. The platform shift is an opportunity. We're more relevant than ever.

It's a fairy tale.

What's actually happening is a structural collapse in the value of the thing these companies sell — buyer intent — and it's happening fast enough that you can already read it in the public market caps. The intent industry spent a decade building moats out of stuff that a college kid with Claude and a credit card can now replicate in a weekend. The moats aren't draining. They've already drained. Most of the players just haven't priced it in yet.

What "intent" actually was

Strip away the marketing and B2B intent data was always three things stitched together:

The whole industry was built on the bet that aggregating those three layers at scale was hard enough that it justified a six-figure ACV. And for fifteen years, that bet was correct. Bombora, ZoomInfo, 6sense, Demandbase, HG Insights, G2, Gartner Digital Markets — they all priced their data like it was rare because, mechanically, gathering it was rare. You needed a co-op of publishers, a scraping infrastructure, an enrichment waterfall, a sales team to sell it, and three years of trust before a Fortune 500 RevOps team would buy.

Now, every one of those layers is being commoditized in real time. Not by a competitor. By the model layer itself.

Apollo is the canary

Apollo is the cleanest example of what's coming for everyone. They built a great database — over 210 million contacts and 30 million companies — and stapled an engagement layer on top. The pitch was always "we have the data, and we have the workflow." That was a defensible position in 2022.

In 2026, it's a feature.

Apollo's own roadmap tells the story. They just shipped what they call "Apollo's AI Assistant" — natural language conversations that turn outbound intent into quick execution, telling you who to target, what to send, and what to do next. They expanded technographic coverage by using AI to extract data from 10M+ job postings, with technologies per organization nearly doubling. Read that again. The most-funded contact database in B2B is now using LLMs to scrape job postings — the exact same thing HG Insights has charged enterprise prices for since 2010.

The reason is obvious: the moat wasn't the data, it was the cost of acquiring the data. And that cost just collapsed.

Meanwhile the deficiencies pile up. Apollo's intent product, the thing supposed to justify upgrading to the Organization tier, provides account-level only signals — telling you "Company X is researching topic Y" but never identifying which contact at that company is showing buying behavior. So you're paying $119 per seat per month for resold Bombora data and a chatbot that scrapes LinkedIn. That's not an intent platform. That's a wrapper.

And here's the part nobody at Apollo wants to say out loud: a Claude or GPT instance with a browser tool can do the prospecting research portion of Apollo's job, today, for the cost of a few API tokens. Not perfectly. Not at the same fidelity. But well enough that the question shifts from "is the AI good enough" to "is the data 65-70% accurate" — which, by Apollo's own customers' admission, it already is.

The "intent lunch" isn't being eaten. It's being given away as a free side dish.

Gartner's stock chart is the rest of the story

Gartner is supposed to be the cathedral. The brand. The one B2B research firm that justifies a Magic Quadrant license fee because they have analysts and you don't.

The market disagrees.

That's not a soft quarter. That's investors pricing in the end of a business model.

The cause isn't subtle. Gartner's own predictions team forecasts that by 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges, and that traditional SEO and PPC will give way to agent engine optimization. Gartner is publishing the obituary for its own pricing power. The minute a CFO can ask an agent "show me the top six HRIS platforms ranked for a 2,000-person services firm with global payroll," and that agent returns a synthesized answer drawn from a hundred sources, the Magic Quadrant becomes wallpaper. It's still pretty. Nobody pays $30K to look at it.

This is the same dynamic that ate G2's traffic and forced them to consolidate Capterra, Software Advice, and GetApp from Gartner. Not strategy. Survival. When the chat interface replaces the comparison page, the only viable move is to become a data source for the chat interface — which means competing with every other data source on price and recency, not on brand.

HG Insights and the technographics commoditization problem

HG Insights built its empire on one genuinely hard claim: they reveal behind-the-firewall technology usage that conventional web-scraping or digital-signature methods cannot detect. They processed contracts, SEC filings, RFPs, earnings calls, and job postings with NLP pipelines that took years to build. That's why 90% of Fortune 500 technology companies rely on HG Insights data, with traditional access requiring expensive annual contracts often starting at $50,000+ per year with lengthy procurement processes.

Now ask yourself how hard that is to replicate today.

A reasonably competent engineer with a Claude or GPT API key, Firecrawl, and a Postgres instance can build an agent that does the following in an afternoon:

• Pulls a target company's careers page, indexed job posts on LinkedIn, and recent press releases.
• Extracts mentioned technologies, vendor names, version numbers, hiring patterns, and budget signals using a structured prompt.
• Cross-references SEC filings via EDGAR for any public parent.
• Outputs a clean technographic record with confidence scores.

That's not theoretical. That's a Tuesday on Base44 or Replit. The LLM does the NLP work that HG built a decade-long moat around. The "behind the firewall" detection is just pattern matching on signals — signals that any frontier model now extracts as a side effect of being asked nicely.

To HG's credit, they see it. Their entire 2026 pivot is an MCP server that connects HG Insights intelligence directly to external AI tools like Claude, ChatGPT, and Cursor. That's a smart move. It's also an admission. When your defense is "we'll be the data layer your agent calls," you've conceded that the agent — not the platform — is where the user actually lives.

The same logic applies to Clay, ZoomInfo, Cognism, LeadGenius, and everyone else in the technographic and firmographic space. The atomic data unit — "Company X uses Y technology, spends Z on it, contract renews in Q3" — is becoming a commodity input. The value migrates upward to the orchestration layer (which agent calls which data source when) and downward to the activation layer (what gets done with the answer). The middle, where most of the industry currently lives, gets squeezed.

Where the moat actually still holds

Now for the part of the industry that isn't on fire.

Netline runs the largest B2B content syndication network. They sell something genuinely hard for an agent to fake: a real human, at a real company, downloading a real piece of gated content, with a real form fill, on a property they don't own.

That's a first-party, permissioned, identity-resolved buyer signal. TechTarget's strength lies in its owned media portfolio of over 150 technology-focused publications. Because they own the sites where the content lives, their intent signals are genuinely first-party and highly reliable. Same architecture, same defensibility.

You cannot synthesize this with an LLM. You cannot scrape it. You cannot agent your way into it. To replicate Netline, you'd need to build a fifteen-thousand-publisher network from scratch and convince actual buyers to opt in to actual content registrations. That's not a software problem. That's a fifteen-year distribution problem, and AI doesn't compress distribution timelines.

The same is true for content delivery networks (CDNs) and the publishers themselves. The intent signal that travels through a Cloudflare or an Akamai pipe — a real session, a real device, a real cookie that hasn't been killed yet — is provably human and provably engaged. You can argue about freshness and resolution, but you cannot argue about its existence.

This is where I'd put real money in 2026: networks where the moat is human attention they actually own, not data they aggregate from public sources. Everything else is on the table.

The four predictions

Here's what the next 24 months look like if I'm right about any of this.

What's left

The honest answer about the future of the intent space: the data layer becomes a commodity, the orchestration layer becomes a frontier-model game (which only Anthropic, OpenAI, Google, and a few others can play), and the activation layer becomes a knife fight over the last truly defensible asset — actual human attention on actual content.

Everything in between gets compressed into a footnote on a CFO's expense report.

The companies that survive this will be the ones that stop pretending their data is a product and start treating it as ammunition for whatever workflow the buyer is actually trying to run. The ones that don't will keep printing slide decks about "AI-powered intent intelligence" until the renewals stop coming.

G2 saw this and bought their way into being unavoidable. Gartner is finding out what happens when you don't. Apollo is racing to wrap an LLM around a database before someone else wraps a better LLM around a better database. HG Insights is quietly turning itself into infrastructure. Netline is sitting on the part of the moat that AI can't drain.

The intent industry isn't dying. It's getting unbundled, repriced, and redistributed. The bonfire isn't the end of the market. It's just the end of the margins.