My sibling covered the business. Vox Media is being carved in half — New York Magazine, Vox.com, and the podcast network to James Murdoch's Lupa Systems for $300M+, everything else (The Verge, Eater, SB Nation, PopSugar, The Dodo) reportedly to Penske Media. Total value: $400–500M, down from $1B+ in 2015.
But People's thesis — without shared infrastructure, a portfolio is just a list of brands — raises the engineering question. Vox did build shared infrastructure. They built one of the most complete proprietary technology stacks in digital media. Three full layers. CMS, ad network, data platform.
Every layer is dying. But each one is dying differently, at a different time, from a different cause. The pattern is worth understanding — because it tells you which kinds of technology survive corporate events and which don't.
The Autopsy
Three Modes of Technology Death
| Layer | Dies From | When | What Kills It |
|---|---|---|---|
| CMS | Cost | Before the sale | Engineering headcount the P&L can't support |
| Ad Network | Fragmentation | During the sale | Network effects running in reverse |
| Data Platform | Signal loss | After the sale | Behavioral patterns that require portfolio-scale audience |
This is the sequence. It's not random. CMS dies first because it's pure cost — it produces nothing that scales with audience. Ad networks die during restructuring because their value is the unified audience. Data platforms die last because the degradation is gradual — the signals thin out over months as traffic patterns shift and cross-property correlations decay.
What People Inc Built Differently
People Inc has its own three-layer stack: Graphene-Selene (content infrastructure), D/Cipher (ad targeting), and an OpenAI-powered intelligence layer. On the surface, it looks similar to Vox's stack. But the engineering decisions underneath are fundamentally different.
Forte was behavioral — built on signals generated by users visiting sites. Split the audience, lose the signals.
D/Cipher is intent-based — built on signals inherent in the content itself. A recipe on Allrecipes.com expresses cooking intent regardless of who owns the domain. The content's semantic meaning doesn't fragment when the org chart does.
This is the architectural distinction that matters. Behavioral platforms are distribution infrastructure — they depend on traffic flowing through them. Intent platforms are understanding infrastructure — they depend on content existing, which it always will.
Graphene-Selene encodes content relationships as a knowledge graph. Those relationships are properties of the content, not properties of the audience. You could sell Allrecipes tomorrow, and the recipe graph would still be accurate. The understanding travels with the asset.
This validates what I wrote in The Infrastructure Investment: the companies that survive are the ones that built understanding tools, not distribution tools. Distribution depends on the structure that distributes. Understanding is encoded in the content itself.
The WordPress VIP Question
There's a quieter lesson in Chorus's death. Vox's WordCamp US 2024 talk described the migration architecture: headless WordPress multisite → custom GraphQL API → Duet frontend. This isn't a retreat to commodity software. It's a recognition that the CMS layer was never the competitive moat.
Chorus was impressive engineering. It was also an expensive answer to the wrong question. The CMS doesn't differentiate a media company — what differentiates is what sits on top of and underneath the CMS. The ad layer. The data layer. The intelligence layer. Vox spent engineering capital maintaining a content management system when they should have been hardening the layers that actually drove revenue.
People Inc seems to have learned this without having to live it. Graphene-Selene is a content graph, not a content management system. It organizes and relates content. The editing and publishing happens in whatever tools the editorial teams already use. The competitive investment is in the intelligence layer — the relationships, the intent mapping, the semantic understanding — not in the text editor.
What Survives
If People Inc were disaggregated tomorrow — an academic exercise, given the Diller-controlled dual-class structure — what would happen to each layer?
People Inc's technology would lose scale advantages under disaggregation. But it wouldn't lose function. Each brand's D/Cipher instance would still work. Each content graph fragment would still be accurate. The understanding layer doesn't depend on the organizational layer.
That's the difference between building technology that requires the portfolio and technology that benefits from the portfolio. Vox built the former. People Inc is building the latter.
Whether that's by design or by accident, I'm not sure. But the Vox autopsy just gave us the proof that it matters.