Idea: TalentGraph — AI Researcher Intelligence for Venture Capital

Trigger: DeepMind exodus — Jumper (Nobel ’24, AlphaFold) to Anthropic, Shazeer (LaMDA) to OpenAI in 48 hours. Google shares down 5%+. Silver to Ineffable Intelligence. Talent pipeline has structurally reversed. Source Research: Priya report 2026-06-25

The One-Liner

A data intelligence platform that tracks where top AI researchers land, models lab-level “talent gravity” as an investment signal, and surfaces deal flow for VCs before valuation resets happen.

Customer & Problem

Primary customer: Partners at AI-focused VC firms (a16z, Sequoia, Lux, Index, Air Street, Matrix, LocalGlobe, Elaia).

The daily reality: A VC partner wakes up, reads that Jumper left DeepMind for Anthropic, and asks three questions:

“Who else is going to leave DeepMind?”
“Which under-the-radar labs are quietly stacking top talent?”
“Which startups have a talent density that signals they might be the next breakout?”

Right now, answers come from: LinkedIn feeds, personal networks, Signal chats, and hunches. None of it is systematic.

The signal: When DeepMind loses Jumper + Shazeer + Silver in weeks, that’s a 5% stock drop and a structural signal about who will win the next 24 months. VCs would pay for early warning on these flows.

Solution

TalentGraph — an AI researcher intelligence platform purpose-built for venture capital.

Core Product

1. Talent Gravity Index A weighted score per lab/organization combining:

Citation weight (semantic scholar): research impact, adjusted for recency
PhD pedigree (which programs, graduation years, advisor lineage)
Patent density (AI-related patents filed)
Exit velocity (rate of researcher departures vs. arrivals, 90-day rolling)
H-index concentration (top-decile researcher density)

Output: A single “Gravity Score” per lab. When it falls 15% in a quarter, that’s an early warning. When a small lab builds a score that rivals big labs? That’s deal flow.

2. Researcher Flow Monitor

Real-time departure/arrival tracking for 5,000+ tracked AI researchers
Alerts: “3 top-100 cited researchers left DeepMind this week” or “Anthropic’s talent gravity just passed OpenAI on the inference track”
Historical migration maps: “Where did every departing Google Brain researcher go since 2023?”
Compensation proxy estimates based on role, lab prestige, and funding rounds

3. Deal Signal Engine

Surfaces labs/startups with rising talent gravity scores that lack proportional venture funding
“Undervalued talent clusters” — 3+ top researchers gathered in a pre-seed startup no VC has backed
Cross-reference with: funding history, press mentions, patent filings
Output: curated deal memo with talent data as the lead metric

4. Portfolio Talent Health Dashboard

For funds that have already invested: track whether their portfolio companies are gaining or losing talent density
Early warning: portfolio lab’s gravity score is declining while a competitor’s rises

Why Now

DeepMind exodus (this week) — Jumper + Shazeer + Silver in 30 days. The talent map just redrew itself. VCs who caught this early could have invested in Anthropic before the valuation re-rate.
AI talent is the scarcest resource in tech — There are ~10,000 top-cited AI researchers globally. Every VC wants to know where they are. The market for this intelligence is unserved.
VC competitive pressure — AI specialist funds (Air Street, Elaia) are already doing this manually. Generalist funds that can’t afford the manual research are the real target market.
Public market signal — Google dropped 5% on Jumper+Shazeer departures. That’s a $100B+ market cap swing. Talent data has direct investment return correlation.
Startup formation wave — Fired DeepMind researchers are starting labs (Ineffable Intelligence). VCs need lead indicators to find these before the seed round.

Wedge

Start as a private weekly newsletter/briefing for 50 VC partners. “The Talent Flow” — curate the week’s top-5 researcher moves with analysis. Build the data infrastructure behind the scenes to power the newsletter. After 10-15 issues, start charging for the full dashboard.

Pricing:

Newsletter: Free (demand gen)
Dashboard access: $25K/year per fund (5-seat license)
Enterprise: $100K/year for multi-fund shops (Sequoia, a16z, Index) with API access and custom alerts

TAM: ~300 AI-focused VC funds globally. At $50K average, that’s $15M ARR. Feasible target: 50 funds in year 1 = $2.5M ARR.

Competition & Moat

Direct: PitchBook/Crunchbase (funding data, no talent flows). Signal (hunches, not systematic). No direct competitor.

Adjacent:

LinkedIn Talent Insights — HR-focused, not VC- or research-weighted. No AI-specific researcher tracking.
Semantic Scholar / Google Scholar — citation data, no org-level aggregation or VC-specific deal signals.
The Information’s tech talent tracker — media, not structured data.

Moat:

Data network effect — The more VC firms track via TalentGraph, the more researcher movements get verified and cross-referenced (funding announcements, internal referrals).
Gravity model IP — The citation-PhD-patent weighted scoring isn’t trivial to replicate. Each weighted dimension takes 6+ months to build accurately.
First-mover with VCs — Once a VC firm embeds TalentGraph in their sourcing workflow, switching costs are high.

Risk

Small market — 300 funds is a niche. Upside requires expanding into corporate venture, government research labs, and executive recruiting.
Data accuracy — Researcher moves aren’t always public. Self-reported data + LinkedIn scraping + funding announcements leave gaps. Former DeepMind researchers joining Anthropic isn’t always announced.
Privacy concerns — Tracking individual researchers could face pushback. Product must be positioned as organization-level intelligence with individual flows as supporting data.
VC seasonality — Funds may churn after a slow quarter. Annual contracts required.
AI researcher paywalls — If top researchers stop publishing moves publicly (e.g., Anthropic’s stealth culture), the model loses signal.

File

~/Library/CloudStorage/Dropbox/AI/Obsidian/Resources/Ideas/Backlog/2026-06-25-talent-graph.md