💥University Ranking by Exit Values, How...

1. Key Themes

AI Is Radically Bifurcating the Seed Market — Both in Valuations and Revenue Expectations

The seed stage is fracturing into two incompatible realities. At the top end, AI-driven hype is producing unprecedented valuations, while the broad market remains disciplined.

"Top 5% seed valuations reached $115M in Q4 2025, a level that is historically unprecedented and clearly decoupled from the rest of the market."

"The data shows a clear bifurcation between the top 5% (AI-driven outliers) and the remaining 95% of seed rounds."

On the revenue side, the same bifurcation holds — companies are either pre-revenue or already at serious traction, with the middle ground disappearing:

"The real signal is variance: the best seed companies today are either pre-revenue with extraordinary teams or already at $2M+ ARR. The middle is getting squeezed."

The AI Value Chain Is Structurally Inverted — Semiconductors Capture the Economics, Not Applications

The profit structure of AI looks nothing like the cloud stack it is supposedly disrupting, and the timeline to equilibrium is much longer than the market assumes.

"The semiconductor layer still captures ~70% of all AI revenue ($300B) and ~79% of all gross profit, with NVIDIA alone representing ~80% of the semi layer at ~$250B annualized."

"The AI value chain is almost exactly the mirror image of the cloud stack, where apps capture 70% and semis capture 6%."

"At the current pace of profit share shift (~4 points per two years), it would take well over a decade for the app layer to reach cloud-like economics."

Compounding AI Automation Is Becoming a Durable Operating Advantage

Two separate pieces in this issue converge on the same insight: AI is most powerful when it is self-reinforcing — building knowledge or automation loops that compound rather than deliver one-time productivity gains.

On knowledge management (Karpathy):

"The system creates a compounding knowledge loop: queries and explorations get 'filed' back into the wiki, so personal research always accumulates rather than evaporating in chat windows."

On automation discovery (Lieberman):

"The system compounds because each new skill creates more data for the meta-skill to analyze."

University Talent Sourcing Is Broader Than the Elite Network Narrative Suggests

Exit value data challenges the VC assumption that founder talent concentrates in a small set of elite private universities.

"Public universities hold their own across the top 25: Berkeley (#4), UCLA (#5), Michigan (#7), Illinois (#11), UC Santa Barbara (#16), Maryland (#20), ASU (#22), and UW (#24) all make the list."

"The data challenges the assumption that venture-backed startup success is concentrated exclusively in elite private institutions."

Hyperscaler Capex Is Accelerating Into a Custom Silicon Arms Race — With Direct Implications for NVIDIA's Moat

Infrastructure spending is growing at a pace that is reshaping competitive dynamics at the model layer.

"Top 5 hyperscalers spent $443B in capex in 2025 (up 73% YoY) with $600B+ projected for 2026, roughly 75% directed at AI infrastructure."

"Every major hyperscaler is hedging with custom silicon. Google TPUs are now merchant hardware, forcing NVIDIA to cut pricing ~30% for some customers. Amazon's Trainium crossed $10B annual run rate growing triple digits."

2. Contrarian Perspectives

AI Can Be Both Genuinely Transformational and a Frenzied Hype Cycle Simultaneously — and Most Investments Won't Return Capital

The prevailing narrative treats AI's transformational potential as validation for current investment prices. The article draws a sharp distinction between the two.

"Walker makes an important distinction: AI can absolutely transform the world and still be a frenzied hype cycle where most investments do not result in returns. Both things can be true simultaneously."

The evidence: the top 5% of seed rounds are at $115M valuations — historically unprecedented — while the other 95% remain disciplined. The hype is real and measurable, even as the technology's long-term value may also be real.

The "Average" Seed Valuation Is a Meaningless Number That Actively Misleads Investors

Most market participants benchmark against aggregate figures. The data shows this is analytically dangerous.

"Investors looking at 'average' seed valuations are seeing a blended number that masks two completely different realities. The top end is moving faster than ever while the broad market remains disciplined."

"The VC world is heading somewhere many experienced investors cannot relate to. Massive funds are playing at the earliest stages, wild valuations are being signed in hours rather than weeks, and secondaries are flying off the shelves." — Bryce Roberts, Indie VC

Elite University Networks Are an Overused and Systematically Biased Sourcing Filter

VCs who over-index on Harvard, Stanford, and MIT are systematically missing high-value founder pipelines. The exit value data makes this concrete:

"Rensselaer Polytechnic Institute (#14, $66B) sits ahead of Yale (#12, $73B). University of Illinois Urbana-Champaign ($74B) outperforms both Columbia and Yale. Rice University ($65B) is in the same tier as several Ivy League schools."

"Firms over-indexing on a handful of elite networks are systematically missing founder talent. The best origination strategies go where the data points, not where the brand signals."

3. Companies Identified

NVIDIA

Description: Dominant AI semiconductor company
Why mentioned: Captures ~~80% of the semiconductor layer (~~$250B annualized), which itself captures 79% of all AI gross profit; facing pricing pressure from hyperscaler custom silicon
Quote: "NVIDIA alone representing ~80% of the semi layer at ~$250B annualized... forcing NVIDIA to cut pricing ~30% for some customers."

OpenAI

Description: Leading AI application/model company
Why mentioned: Represents ~~75% of the application layer alongside Anthropic (~~$40-50B combined); signed a multiyear deal with Broadcom for 10GW of custom accelerators
Quote: "OpenAI and Anthropic together represent ~$40-50B and ~75% of the layer."

Anthropic

Description: AI safety-focused model company
Why mentioned: Co-dominates the application layer with OpenAI; mentioned as a key beneficiary of current app-layer economics
Quote: "OpenAI and Anthropic together represent ~$40-50B and ~75% of the layer."

Cursor

Description: AI coding assistant
Why mentioned: Named as a "distant third" in the application layer, leading the coding AI cohort
Quote: "A distant third tier includes coding AI players like Cursor."

ElevenLabs, Glean, Sierra, Perplexity, Replit, Lovable, Harvey, Abridge

Description: Fast-growing AI agent/application companies
Why mentioned: Cited as the next tier of application-layer growth, representing the emerging agent wave
Quote: "Fast-growing agent companies (ElevenLabs, Glean, Sierra, Perplexity, Replit, Lovable, Harvey, Abridge)."

Amazon / AWS (Trainium)

Description: Hyperscaler with custom AI silicon
Why mentioned: Trainium custom chip crossed $10B annual run rate growing triple digits — a direct competitive threat to NVIDIA
Quote: "Amazon's Trainium crossed $10B annual run rate growing triple digits."

Google (TPUs)

Description: Hyperscaler with proprietary tensor processing units
Why mentioned: Now selling TPUs as merchant hardware, directly pressuring NVIDIA's pricing power
Quote: "Google TPUs are now merchant hardware, forcing NVIDIA to cut pricing ~30% for some customers."

Broadcom

Description: Semiconductor and custom silicon designer
Why mentioned: Signed a multiyear deal with OpenAI for 10GW of custom accelerators — a landmark custom silicon contract
Quote: "OpenAI signed a multiyear deal with Broadcom for 10GW of custom accelerators."

Affinity

Description: AI-powered CRM for investment firms
Why mentioned: Sponsor; launching hosted MCP server and AI chat beta for investment teams
Quote: "Affinity walks through its hosted MCP server and AI chat beta — giving investment teams conversational CRM access, automatic meeting briefs, and a self-serve data layer."

Obsidian

Description: Personal knowledge management tool
Why mentioned: Named as the IDE in Karpathy's LLM-maintained personal knowledge base workflow
Quote: "Karpathy uses Obsidian as the IDE, Cursor as the LLM interface."

4. People Identified

Peter Walker

Description: Head of Insights at Carta
Why mentioned: Shared two datasets: (1) fundraising revenue benchmarks by stage from SVB's H1 2026 State of Markets Report; (2) Carta data showing top 5% seed valuations at $115M
Quote: "Walker makes an important distinction: AI can absolutely transform the world and still be a frenzied hype cycle where most investments do not result in returns."

Apoorv Agrawal

Description: Analyst/investor at Altimeter Capital
Why mentioned: Published updated analysis of the Gen AI value chain showing the ecosystem has grown 5x to ~$435B in annualized revenue
Quote: "The ecosystem has grown 5x to ~$435B in annualized revenue but the economic structure remains deeply inverted."

Andrej Karpathy

Description: AI researcher, former OpenAI/Tesla, founder of Eureka Labs
Why mentioned: Shared a detailed LLM-based personal knowledge base workflow that received 18.2M views; framed the shift from "manipulating code" to "manipulating knowledge"
Quote: "The majority of his token throughput now goes into manipulating knowledge rather than code."

Alex Lieberman

Description: Co-founder of Morning Brew
Why mentioned: Shared a Claude-powered "meta-skill" workflow that auto-discovers what to automate across his tools; post received 207K views and 2.6K bookmarks
Quote: "Every Monday at 9am, Cowork scans Linear, Notion, Slack, Gmail, and prior Cowork sessions to identify repeatable processes that should become new skills."

Bryce Roberts

Description: Founder of Indie VC
Why mentioned: Provided color commentary on the unprecedented nature of current seed valuations
Quote: "The VC world is heading somewhere many experienced investors cannot relate to. Massive funds are playing at the earliest stages, wild valuations are being signed in hours rather than weeks, and secondaries are flying off the shelves."

5. Operating Insights

Build a Self-Maintaining LLM Knowledge Base — Not a Static Document System

Karpathy's workflow is directly applicable to any knowledge-intensive firm (VC, consulting, legal). The key architecture: raw source documents feed an LLM that maintains a structured wiki of .md files with summaries, backlinks, and cross-references — with the LLM doing the maintenance, not humans.

"You rarely edit manually; the LLM reads, writes, and maintains the entire knowledge structure... The wiki is treated as a living, self-improving artifact rather than a static document."

"No fancy RAG needed at small scale. The LLM auto-maintains index files and brief summaries, reading all important related data fairly easily when the wiki stays under a certain size."

Deploy a "Meta-Skill" That Auto-Discovers What to Automate — Before Manually Auditing Workflows

Rather than spending cycles deciding what to automate, Lieberman's approach lets an AI agent observe workflow patterns across tools and surface automation candidates automatically.

"Rather than manually deciding what to automate, the system observes workflow patterns across tools and proactively recommends automation candidates. The meta-skill (a skill that creates skills) removes the biggest bottleneck in AI adoption: knowing what to automate in the first place."

"The approach allows anyone to 'proactively become more AI-native without making it a full-time job.'"

Recalibrate Stage-Based Revenue Benchmarks by Sector, Not Just Round

Investors and founders using aggregate stage benchmarks risk drawing wrong conclusions about fundraising readiness. The SVB/Carta data includes pre-revenue biotech and other non-SaaS sectors, pulling medians down artificially.

"Walker notes this caveat is important for interpreting the data. Investors should be careful benchmarking their own portfolio against aggregate figures without adjusting for sector mix."

6. Overlooked Insights

The Engagement Pattern on Lieberman's Post Is Itself a Signal About Market Demand

The bookmark-to-view ratio on Lieberman's automation post is unusually high — 2,600 bookmarks on 207,000 views — suggesting practitioners are saving it for implementation rather than passive reading. This is a meaningful leading indicator that operational AI workflow adoption is shifting from curiosity to active deployment.

"The engagement ratio (bookmarks to likes) suggests people are saving this for implementation, not just passive consumption. The pattern is tool-agnostic and applies to any knowledge work environment."

Karpathy Hints at a Future Where the Knowledge Base Is Fine-Tuned Into Model Weights

Buried at the end of the knowledge base section is a forward-looking prediction that upgrades the significance of the entire workflow — from a productivity tool to a potential moat-building mechanism.

"He predicts the approach extends to fine-tuning, where the LLM would 'know' the data in its weights rather than just the context window."

This means firms that begin building structured, LLM-maintained knowledge bases today are also accumulating a fine-tuning dataset for future proprietary models — a compounding advantage with a long lead time.