Venture Capital’s Incentives⚠️…

1. Key Themes

VC fund structure incentivizes scale over performance

The article surfaces a structural critique that even AI models flag as anomalous: the fee/carry model rewards fund growth rather than investor returns.

"LLMs analyzing today's VC structure tend to optimize for AUM and fees rather than backing strong companies or generating cash returns. The result points to a deeper issue: incentives can favor capital concentration and fee growth over long-term company outcomes."

The IRR/DPI gap is widening — paper gains, not cash

Fund performance data shows a growing disconnect between reported value and actual liquidity for LPs, even years into a fund's life.

"Top-decile IRRs for 2017 to 2019 VC vintages have fallen while TVPI climbed, showing a wider gap between headline value and realized cash. Megafunds now capture 64% of capital raised, while DPI remains just 9% for those vintages even nine years in, keeping liquidity a central LP concern."

AI is compressing both cost curves and defensibility

Two separate insights point to the same shift: AI is making inference dramatically cheaper while simultaneously eroding traditional software moats.

"New models make simple classification decisions in milliseconds by scoring probabilities instead of generating full responses. On production email data, they roughly doubled accuracy while cutting inference costs by as much as 100x."

"Switching costs, data lock-in, and long contracts are becoming weaker as AI makes migrations easier and buyers favor flexible pricing. With median gross revenue retention falling from 88% to 84%, product velocity is becoming a more important defense."

Startup geography is decentralizing

Innovation hubs are no longer concentrated in a handful of legacy cities.

"The Bay Area led Global Champions, Mumbai topped Rising Stars, and smaller hubs such as Cambridge and Boulder ranked among the strongest for startup density."

2. Contrarian Perspectives

VC's own incentive structure is arguably broken — even by AI's own assessment. The conventional narrative is that VCs are aligned with LPs and founders through the promise of outsized returns; this piece argues the actual mechanics reward fee-generation and capital concentration instead.

"incentives can favor capital concentration and fee growth over long-term company outcomes."

Software moats — long considered a durable competitive advantage — are dissolving. Rather than continuing to praise "sticky" enterprise software with high switching costs, the article argues speed of shipping is replacing lock-in as the real defense, evidenced by falling retention metrics.

"median gross revenue retention falling from 88% to 84%, product velocity is becoming a more important defense."

"Pre-empts" are mostly relationship-driven, not opportunistic sprints. Common founder lore treats a pre-empt as a lucky surprise offer; the reality described here is that it's the product of long-cultivated investor relationships and formal documentation.

"A real pre-empt starts with a written term sheet and usually follows months of relationship building over a surprise verbal offer."

3. Companies Identified

Meta (Muse) — AI consumer app. Mentioned as outpacing ChatGPT's early adoption curve while facing platform friction.

"Muse reached an estimated 642,000 US daily mobile users by day 12, roughly three times ChatGPT's pace, alongside 2.8 million downloads... faced an Amazon shopping block over security concerns."

OpenAI — AI lab. Mentioned for productizing its internal agent tooling for external developers.

"The Agents API gives developers access to the same core harness used by Codex, without a platform fee beyond token costs... evaluation scores rising from 0.71 to 0.85."

Carta — Cap table/data platform. Cited as the source for VC fund performance data illustrating the IRR/DPI liquidity gap.

"Megafunds now capture 64% of capital raised, while DPI remains just 9% for those vintages even nine years in."

Dealroom.co — Market intelligence firm. Cited for global startup ecosystem rankings showing geographic diversification.

"Dealroom assessed 325 ecosystems across 77 countries across investment, innovation, talent, and outcomes."

TEKEVER, Island, Snorkel AI, Enveda, Precision Neuroscience, Hubble Network, Basecamp Research, Numeral, Go.AI, Ultraviolette Automotive, Ema, BigHat Biosciences, Anaconda Biomed, Confido, Ande — Notable recent large raises spanning defense-tech, AI infrastructure, biotech, and enterprise AI, signaling where big late-stage capital is flowing (autonomous systems, enterprise browsers/agent control, AI drug discovery, BCI, satellite connectivity).
Dvash Group & Hebrew University, Connect Ventures, NewSchool VC, Project Ventures, Myriad Defence & Final Frontier, Mirae Asset & SparkLabs, Pulse Fund, Misumi Ventures — Newly launched funds, notable for concentration in deeptech, defense-tech, climate, and frontier hardware — an emerging thematic pattern in fund formation.

4. People Identified

Aaron Harris — Referenced for insight on how funding rounds actually close. Why mentioned: distinguishes real "pre-empts" from founder mythology.

"A real pre-empt starts with a written term sheet and usually follows months of relationship building over a surprise verbal offer."

Tomasz Tunguz — VC commentator. Why mentioned: highlighted the shift toward cheap, specialized AI classifiers over expensive generative models.

"roughly doubled accuracy while cutting inference costs by as much as 100x."

Auren Hoffman — Referenced regarding the erosion of software moats. Why mentioned: connects declining retention metrics to the need for faster product shipping as the new competitive edge.

"product velocity is becoming a more important defense."

Jane Frankland — Cybersecurity leader, author, advisor (sponsor content). Why mentioned: featured in a Vanta-sponsored session on AI governance and regulatory readiness (EU AI Act, ISO 42001, NIST AI RMF).

5. Operating Insights

Use specialized, narrow AI models for classification tasks instead of full generative calls — this cuts inference costs by up to 100x and improves accuracy, a high-leverage tactic for founders building AI-heavy products.

"cutting inference costs by as much as 100x"

Don't rely on switching costs or contract lock-in as a durable moat — with GRR falling and AI easing migrations, founders should prioritize continuous shipping velocity as the primary defensibility strategy.

"product velocity is becoming a more important defense"

Real pre-empts require sustained relationship-building, not cold outreach — founders should treat investor relations as a long-cycle activity months ahead of a raise, not a reactive scramble.

"usually follows months of relationship building over a surprise verbal offer"

6. Overlooked Insights

New fund formation is skewing heavily toward deeptech, defense, and frontier hardware rather than generic SaaS/AI, suggesting LPs and GPs see differentiated opportunity in harder-to-replicate categories (e.g., Dvash Group's "deeptech, quantum technologies, dual-use technologies," Myriad Defence's "defence-tech VC fund targeting early-stage companies across the New Nordics and Ukraine," Misumi Ventures' "frontier hardware, robotics, aerospace, medtech, clean energy, eVTOL").
Agent infrastructure is being commoditized/productized by foundation model labs themselves — OpenAI packaging its internal Codex harness into a public API (with no platform markup beyond token costs) suggests infra-layer startups building "agent harnesses" may face direct competition from the model providers they depend on.

"gives developers access to the same core harness used by Codex, without a platform fee beyond token costs"