💥The Mark Up Game, Making of a… | Data Driven VC Summary

1. Key Themes

AI Infrastructure Is Eating the Series A Market

Capital concentration at Series A has become extreme. Across 1,082 US Series A rounds (July 2025–June 2026), Walker flags "AI Infrastructure, Semiconductors, Hardware and Cybersecurity as the sectors investors love most, against a median of $70.0M on a $14.1M median raise." Meanwhile, "Medical Devices has the lowest median valuations of any sector, Edtech has stayed flat despite the AI boom, and Food & Bev rounds are steady in count but low in valuation." The 80% figure is the sharpest signal: "So far in 2026, about 80% of companies building software products were AI companies, making AI the default starting position for founders."

The Founder Pipeline Has Structurally Shifted

The talent feeder network for unicorn founders has rotated decisively. Stanford's Ilya Strebulaev found that "among employers with 10+ unicorn founders, OpenAI alumni are 14.7x overrepresented versus a random sample, ahead of Check Point (11.2x) and D.E. Shaw (9.8x)." Simultaneously, legacy signals have collapsed: "PayPal's odds ratio as a founder-producing employer fell from 12.0x pre-2016 to 1.7x since." The relational dimension is equally notable: "Among 1,377 multi-founder US unicorns, 68.6% have co-founders who previously worked together or studied together, with shared employers (57.8%) more common than shared schools (33.3%)."

Venture Capital Must Become "Capital Entrepreneurship"

General Catalyst's Hemant Taneja argues the traditional VC value-add model is structurally inadequate for today's capital-intensive companies. "Taneja argues this generation of venture will win on a better-engineered capital stack, since infrastructure and hardware founders need financing that goes past the traditional Rule of 40." His historical precedent: Drexel Burnham Lambert's team "engineered PIK bonds and the high-yield debt market in the 1970s–80s before founding Apollo, Ares and Cerberus." The specific gaps he names include "capital for sales and marketing, GPUs, energy and power-leasing contracts, advance customer deposits, factories, and public-private infrastructure deals."

Analysis-First Capital Allocation Is Dead

Contrary's Kyle Harrison argues that the spreadsheet-driven investment framework has been made obsolete. He contends "the analyst's extrapolation-based playbook is dead, replaced by aggregating quality, narrative, leverage, or time." His case study: in 2018, "Harrison projected Figma could hit $200M ARR by 2022 from $4M ARR (valued ~$400M by Sequoia); an analyst dismissed it as too aggressive, yet Figma overshot even that projection." The four replacement strategies demand articulated worldviews that quantitative extrapolation cannot supply.

AI Agent Architecture Is Shifting from Loops to Graphs

A structural technical shift is underway in how autonomous AI systems are built. Peter Steinberger's viral post — "Are we still talking loops or did we shift to graphs yet?" — captured the field's recognized shift "from single loops to networks of loops ('graphs')." The essay identifies "four structural failures" of single-loop systems: Goodhart's law, blindness to target correctness, conflict between independent loops, and undetected measurement decay. The working model of a mature system is "a champion-challenger loop wired to drift-monitors and automatic rollback, with held-out eval sets the training loop is never allowed to see."

2. Contrarian Perspectives

The PayPal Mafia Is a Lagging Indicator — and VCs Still Rely On It The conventional wisdom in VC sourcing is that "mafia" alumni networks (PayPal, early Google, etc.) are durable signals of founder quality. The Stanford data flatly refutes this for the pre-eminent example: "PayPal's odds ratio as a founder-producing employer fell from 12.0x pre-2016 to 1.7x since," echoed by LinkedIn (6.0x to 1.1x) and Apache Software Foundation (9.8x to 2.2x). Most sourcing playbooks haven't caught up.

Patient Capital Can Be a Trap, Not Just a Virtue The conventional framing of long-horizon investing as inherently superior is challenged by the Pabrai case study. Despite a high-conviction, concentrated bet, "Pabrai held roughly 77% of his fund in Micron since 2017, then sold in 2023 right before Micron ran 15x+ in the AI-memory boom — an estimated $2B miss." Even elite patient-capital investors get punished when analyst instincts override the discipline of staying still. The implication: conviction without a governing worldview about why to hold is not enough.

The "Self-Improving AI" Pitch Has a Hidden Structural Flaw Many AI agent pitches center on autonomous self-improvement loops as a defensibility moat. The article surfaces a critical counter: "Even a good graph fails without ground-truth anchors." A single self-improvement loop suffers from Goodhart's law — "an optimized metric stops measuring what it should" — and the problem compounds in networked systems without independent audits. The implication for diligence: the quality of the eval set, not the sophistication of the loop, is the real defensibility question.

3. Companies Identified

Figma Design software company. Cited as a case study for how analysis-first investing fails: "Harrison projected Figma could hit $200M ARR by 2022 from $4M ARR (valued ~$400M by Sequoia); an analyst dismissed it as too aggressive, yet Figma overshot even that projection."

Micron Semiconductor/memory company. Cited as the asset in Mohnish Pabrai's costly exit: "Pabrai held roughly 77% of his fund in Micron since 2017, then sold in 2023 right before Micron ran 15x+ in the AI-memory boom — an estimated $2B miss."

OpenAI AI research and deployment company. Cited as the highest-overrepresentation employer for unicorn founders: "Among employers with 10+ unicorn founders, OpenAI alumni are 14.7x overrepresented versus a random sample."

Vanta Security compliance company. Cited as evidence that storytelling is a real hiring category: "Vanta [is] hiring a head of storytelling for up to $274,000."

General Catalyst Venture capital firm. Featured for Hemant Taneja's thesis that VCs must become capital entrepreneurs: "Taneja argues this generation of venture will win on a better-engineered capital stack."

Contrary Venture capital firm. Featured for Kyle Harrison's argument that analysis-first investing is dead and must be replaced by one of four aggregation strategies.

Drexel Burnham Lambert 1970s–80s investment bank. Cited as historical precedent for financial innovation: the team "engineered PIK bonds and the high-yield debt market in the 1970s–80s before founding Apollo, Ares and Cerberus."

Apollo, Ares, Cerberus Private equity/credit firms. Named as the institutional descendants of Drexel's financial engineering culture, offered as the model for what VC "capital entrepreneurship" could produce.

Founders Fund Venture capital firm. Named as an example of the "Contrast" storytelling framework for its "stance against incrementalism."

Y Combinator Startup accelerator. Named as an example of the "Insight" storytelling framework: "Y Combinator's thesis that great startups don't look great early."

Indie.vc Venture firm. Named as an example of the "Path" storytelling framework for its "alternative to raise-scale-exit."

Granola AI meeting transcription tool (newsletter sponsor). "Granola transcribes directly from your computer or phone audio. It works across any meeting tool: Zoom, Google Meet, Microsoft Teams."

4. People Identified

Peter Walker Head of Insights, Carta. Cited for breaking down Series A sector valuation data across 1,082 US rounds: "Walker flags AI Infrastructure, Semiconductors, Hardware and Cybersecurity as the sectors investors love most."

Kyle Harrison Investor, Contrary (previously Kickstart, TCV, Coatue, Index Ventures). Author of the "Capital Allocation Is Dead" thesis: "Harrison…argues the analyst's extrapolation-based playbook is dead, replaced by aggregating quality, narrative, leverage, or time."

Ilya Strebulaev Professor, Stanford GSB. Led the study mapping prior employers of 4,357 unicorn founders: "Strebulaev…finds the feeder pipeline has shifted hard toward Google, Facebook, and OpenAI."

Hemant Taneja Managing Director, General Catalyst. Author of the "Stop Playing the Mark Up Game" thesis, arguing VCs must engineer capital structures: "Taneja argues this generation of venture will win on a better-engineered capital stack."

Laurie Owen Founder/principal, Refinery Media. Author of the VC storytelling framework breakdown: "Owen cites Vanta hiring a head of storytelling for up to $274,000 and LinkedIn 'storyteller' job postings doubling over the past year as proof the function is now a real hiring category."

Peter Steinberger Engineer/technologist. Author of the viral post triggering the AI agent architecture discussion: "Steinberger's post, 'Are we still talking loops or did we shift to graphs yet?', gathered thousands of likes."

Mohnish Pabrai Value investor and fund manager. Case study in the limits of patient-capital conviction: "Pabrai held roughly 77% of his fund in Micron since 2017, then sold in 2023 right before Micron ran 15x+ in the AI-memory boom — an estimated $2B miss."

Andre Retterath Author, Data Driven VC newsletter. Curator and analyst synthesizing the above research for investors.

5. Operating Insights

For VCs: Commit to One of Four Allocator Strategies — and Name It Harrison's framework isn't just conceptual; it's a forcing function for self-definition. The four strategies — aggregating quality, narrative, leverage, or time — each require an articulated worldview. The practical takeaway: "Each of the four allocator strategies demands an articulated worldview, something a spreadsheet can't supply on its own." Funds that can't name their strategy are likely running an undefined hybrid that wins by accident.

For Fund GPs: Pick One Storytelling Frame and Repeat It Relentlessly VC storytelling fails when it tries to be everything. Owen's four frameworks — Contrast, Insight, Solution, Path — are mutually exclusive positioning choices: "Committing to one of the four frames is the practical takeaway for any GP rewriting their positioning." The structural difference from company marketing: "VC storytelling has to zoom out to how value accrues across a decade, since fund performance is a lagging indicator." LPs and founders respond to consistent character and conflict, not portfolio lists.

For Sourcing Teams: Remap Your Talent Radar to OpenAI, Google, and Facebook The Stanford data provides a direct operational update to pre-seed and seed sourcing: "For talent scouts and pre-seed investors, Google, Facebook and OpenAI are the new proxy worth watching." With OpenAI alumni at 14.7x overrepresentation and legacy signals like PayPal at 1.7x, the sourcing filter needs to be updated — and updated now, before the rest of the market prices it in.

6. Overlooked Insights

Co-Founder Relationships Are More Predictive Than University Networks The Stanford data surfaces a specific and underappreciated finding: among multi-founder unicorns, shared prior employers (57.8%) are more predictive than shared schools (33.3%). The headline narrative around "university alumni networks" as a founder sourcing signal may be systematically overstated relative to workplace networks. The implication for accelerators, angels, and pre-seed investors: tracking team formation through company alumni channels (not just university alumni associations) may be the more reliable pipeline.

LinkedIn "Storyteller" Job Postings Have Doubled in One Year Owen's data point on the professionalization of storytelling is easy to skip past but has real implications: "LinkedIn 'storyteller' job postings [have doubled] over the past year." This signals that a function once treated as informal (founder narrative, content, thought leadership) is being institutionalized at the operator level — with compensation to match ($274,000 at Vanta). For startups and funds alike, the window to hire this capability cheaply may be closing.