👋 Hi, I’m Andre and welcome to my newsletter Data Driven VC which is all about becoming a better investor with data and AI.
Upcoming events:
DDVC Investor Summit at Bits & Pretzels 2026 Sep 28-30
DDVC x Bits & Pretzels Breakfast 30th Sept in Munich
Brought to you by Granola – AI Copilot for Back2Back Meetings
You know those meetings where a random bot joins and suddenly everyone’s distracted? Granola works differently.
There are no meeting bots. Nothing weird joins your call.
Granola transcribes directly from your computer or phone audio. It works across any meeting tool: Zoom, Google Meet, Microsoft Teams. And even for in-person conversations just click to start.
Our team has been using it for years and couldn’t live without it anymore. Now you can test it 1 month for free with code "DATA100OFF"
Welcome to another Data Driven VC “Insights” episode where we cover the most interesting research and reports about startups, GPs, LPs, AI & automation.
The Cost of AI Talent
Adrian Radu, at Khosla Ventures, published a Cost Per Headcount analysis that tracks reported acquihire and acquisition deal values per employee from 2013 to 2026, showing the top end of the market moving from ~$15M/head to nine figures.
Classic Era Ceiling Was ~$15M/Head: Google's 2013 acquisition of DNNResearch (the Hinton/Krizhevsky/Sutskever spinout) for a reported $40-44M implies roughly $15M per head, the top of the market through 2017.
MosaicML Reset the Mid-2020s Benchmark: Databricks' ~$1.3B acquisition of MosaicML (~60 employees) put the implied value of talent at roughly $21M per head.
Frontier-Era Deals Reach $67-118M/Head: 2024-2026 "reverse acquihires" (licensing technology and hiring key people without buying the company) reportedly reached $87M/head for Character.AI, $67M/head for Groq, and $118M/head for io Products.
✈️ KEY TAKEAWAYS
The going per-head rate for elite AI researchers now approaches what a whole founding team commanded a decade ago. For VCs, this reprices early-team risk at pre-seed AI labs and raises the bar for what "talent density" needs to look like to justify a valuation.
The State of Pre-Seed
Carta's State of Pre-Seed Q2 2026 report, shared by Hamza Shad, shows how widely post-money valuation caps (val caps) disperse for SAFE rounds of the same size.
Median Val Cap of $15M for a $2M Round: For a $2M pre-seed SAFE (the $1M-$2.4M round-size bucket), the median post-money val cap is $15M, which Carta calculates as roughly 13% dilution.
90th Percentile Reaches $40M, 10th Percentile Just $8M: Top-decile founders in that same round-size bucket see a $40M val cap (5% dilution), versus $8M (25% dilution) at the bottom decile, a 5x spread in dilution outcome for an identical round size.
$100M Caps Now Appear on $4-6M Rounds: For rounds above $2.5M, the 90th-percentile val cap reaches $100M, even though the actual round sizes at that percentile are typically only $4-6M.
✈️ KEY TAKEAWAYS
Dilution outcomes at pre-seed now vary enormously by founder profile instead of round size alone. Investors benchmarking pre-seed terms should compare against the relevant decile, since a median-only comparison risks mispricing deals for standout teams.
Join 1925+ investors in our free Slack group as we automate our VC job end-to-end with AI. Live experiment. Full transparency.
How to Own Your Intelligence
Sequoia's Own Your Intelligence: A How-To Guide, by Sonya Huang, lays out when and how portfolio companies should build their own AI capability instead of renting frontier APIs.
Build When Renting Stops Making Sense: Sequoia flags four triggers: usage costs eating into margins, small fast models beating slower ones on latency, sensitive data staying in-house, and wanting control over how the product learns.
A Four-Step Buildout, in Order: Define evals to test the system, build a harness governing the model's context and tools, apply the lightest post-training fix that moves the score, then feed usage data back in to improve.
Match the Fix to the Actual Problem: Missing knowledge calls for context or retrieval, bad formatting calls for fine-tuning, taste calls for preference tuning, a narrow task calls for reinforcement learning, and speed or cost issues call for distillation.
✈️ KEY TAKEAWAYS
The four triggers work as a filter: if none apply, renting is probably still the better call. If one does, follow the order: eval, harness, then the smallest post-training fix that solves the actual problem.
Rise and Fall of Top Employers over Time
Stanford GSB professor Ilya Strebulaev's analysis compares the prior employers of 2,633 unicorn founders (2015 or earlier) against 1,194 founders of unicorns founded in 2016 or later.
Google Alumni Now 9.1% of Unicorn Founders: Google's share of prior-employer credit for unicorn founders rose from 4.4% (pre-2016 cohort) to 9.1% (post-2015 cohort), close to one in eleven founders.
OpenAI Went From 0% to 1.8%: New-generation employers entered the ranking entirely in the post-2015 period: OpenAI (0% to 1.8%), Flagship Pioneering (0.2% to 1.6%), and Dropbox (0.04% to 1.3%).
Legacy Tech Giants Declined: IBM fell from 2.8% to 2.1% of founders' prior employers, Accenture went from 1.2% to 0.6%, and Hewlett-Packard from 1.4% to 0.3%; Juniper Networks and Siebel Systems produced no unicorn founders in the post-2015 sample.
✈️ KEY TAKEAWAYS
The founder pipeline is shifting toward frontier AI labs and modern platforms as higher-signal alumni networks, at the expense of legacy enterprise-software incumbents. For sourcing, this is a reason to weight recent AI-lab experience more heavily when scouting.
Upgrade your subscription to access our premium content & join the Data Driven VC community
Tech Is The Economy Now
Stephen Messer's piece, “Tech Is Not an Asset Class Anymore. It Is the Economy.”, argues traditional VC is being absorbed by much larger capital pools now moving into private tech.
VC Is a $1.3T Niche Inside a Much Larger Pool: Global VC AUM sits at roughly $1.3 trillion, versus $11.5 trillion at BlackRock, ~$8 trillion in US private equity, and ~$4.5 trillion in hedge funds.
a16z's Latest Fund Alone Was 18% of 2025 US Venture Dollars: Andreessen Horowitz's January 2026 $15 billion raise accounted for more than 18% of all US venture capital deployed in 2025.
Non-Traditional Capital Is Moving Down the Stack: The piece cites Point72's $1B+ private-credit fund targeting tech borrowers, sovereign funds (Mubadala, GIC, Norges Bank) taking direct private-tech positions, and Goldman Sachs' 2024 acquisition of secondary specialist Industry Ventures as evidence.
✈️ KEY TAKEAWAYS
As hedge funds, sovereign wealth funds, and banks move directly into private tech, traditional VCs gain new co-investors and competitors that can move faster and write larger checks. Funds that can't differentiate on speed, network, or check size risk losing allocation to these larger pools
Technical Ownership Drives AI Adoption
The DDVC Landscape Report 2026 surveyed globally leading VC firms on who owns their internal AI/data stack across four firm archetypes.
50% of Fullstack VCs Have a Tech Lead Own the Stack: This is the highest technical-ownership share of any archetype, versus 17% at Workflow VC firms (build their own automations without engineers) and 7% at Productivity-focused firms.
Workflow VC Concentrates Ownership at the Partner Level: GPs/Partners (44%) and Investors (36%) jointly own the stack in 80% of Workflow VC firms, versus a combined 44% at Fullstack VC firms (build data infrastructure and internal tech stacks with engineers).
The Transition to Technical Ownership: As firms move from off-the-shelf tools to proprietary infrastructure, ownership shifts away from GPs and Investors toward dedicated technical hires.
✈️ KEY TAKEAWAYS
For investment firms, the practical read is to audit who currently owns the AI/data stack and whether that owner has the mandate and time to drive adoption forward. The most advanced firms have dedicated technical C-Level owners, whereas the least advanced have non-technical Partners own the stack.
That’s it for today!
Stay driven,
Andre
PS: Join the DDVC Investor Summit at Bits & Pretzels virtually or physically in Munich during Oktoberfest Sep 28-30
