Miles Dieffenbach: Inside Carnegie Mellon’s $4BN Endowment & The Math Behind DPI, TVPI, Illiquidity
Summary
Most allocators are not being paid for venture’s risk. Ron cites mature-vintage returns of roughly 8% median net IRR, 15% top-quartile IRR, 2.5x top-quartile TVPI and 1.8x DPI; against QQQ, only top-decile managers consistently outperform. His gate for new LPs is therefore stark: unless they can access that tier, “90% of LPs shouldn’t be investing in venture.”
A $7 billion multi-stage fund can require nearly an entire record exit year to achieve one target return. At roughly 5% dollar-weighted entry ownership, the fund deploys into $140 billion of enterprise value; generating CMU’s 4x net target requires at least 6x gross, or approximately $800 billion of exits versus roughly $850 billion across the entire market in 2021. Harry argues outcome sizes could explode; Ron concedes “we could be wrong,” but refuses to underwrite without a margin of safety.
The IPO market is not closed—the price expectations of private holders are misaligned with public alternatives. Ron contrasts a $100 million-ARR SaaS company growing 15% around breakeven with Microsoft growing revenue 14%, earnings 17%, producing GAAP profits and buying back stock. With public capital again rewarding companies such as Circle and CoreWeave, his message is categorical: “Now is the time. Please take your companies public.”
Manager underwriting is ultimately about people, incentives and ownership of past wins, not polished track-record tables. CMU seeks at least 20 references, treats GP-provided references as the least informative and reconstructs partner-level attribution itself. Partnership failures usually reduce to “incentives” and “who’s working the hardest,” while a fund relationship can last 25 years—making patience more valuable than securing a fashionable allocation quickly.
Selling has become venture’s neglected fifth discipline, but premature secondary sales can destroy the right tail. Managers failed to sell sufficiently when software traded at 20x ARR—and 40x for top growers—in 2021, yet CMU’s own 13-year-old fund later gained roughly three additional turns from Circle after the position had fallen below $1 million of reported NAV. The lesson is not never to sell; it is that marks, liquidity needs and tail optionality must be underwritten separately.
Scaled growth funds increasingly charge venture economics for something resembling long-only public investing. Ron estimates one stacked platform with $15 billion across recent funds could collect roughly $300 million annually in fees; at that scale, he thinks growth vehicles should move toward 1-and-10, zero-and-10 or budget-based fees. He does not blame GPs for maximizing a remarkable business model, but asks whether “the magic bond” between GP and LP has broken.
AI can transform the economy and still produce severe losses for today’s capital providers. OpenAI’s improving unit economics do not eliminate its stated $5-$10 billion annual burn or roughly $70-$80 billion financing stack; unlike self-funding SpaceX, it remains dependent on capital markets. Ron also frames a possible Nvidia cycle—not a forecast—in which revenue falls 20%-30%, earnings 40% and the multiple from 38x to 24x, producing a roughly 70% drawdown even if the long-term AI thesis survives.
Deep dive
1. Surviving lymphoma turned adversity into a choice of response
At 26, Ron learned that his lymphoma had already progressed substantially and chemotherapy needed to begin within a week. After spending roughly 12 hours asking “Why me?”, he returned to a football coach’s maxim: success is “10% what happens to you and 90% how you react.”
His response was an extreme workout regimen built around the idea that “cancer can’t kill me if I don’t stop moving”; chemotherapy became his recovery period. Four months later he was cancer-free, receiving the news only after a frightening two-hour wait ended with his doctor entering with open arms.
Ron’s lasting framing is not that adversity is desirable, but that “there’s beauty in the struggle.” Facing mortality gave him a durable perspective: life is a blessing, and relatively few later problems can take him down mentally.
2. CMU’s venture-heavy portfolio still subjects venture to QQQ
CMU manages $4 billion with a top-level allocation of 85% equity and 15% fixed income, managing to that allocation quarterly. Its next layer targets 50% private assets and 50% hedge funds and liquids, including public equities and fixed income.
The private bucket spans venture, buyouts, real estate, natural resources and private credit under a “best athlete portfolio”: capital moves toward the best global risk-adjusted return rather than maintaining rigid subasset-class quotas. Venture represents just under 25% of the entire endowment, five to 10 points above comparable institutions.
CMU’s private-equity book, including venture, funded its own capital calls for the prior three years. Buyouts supplied the most distributions, while venture was the largest detractor as exits slowed and marks fell.
On mature vintages, Ron cites roughly 8% median net IRR, 15% top-quartile IRR, 2.5x top-quartile TVPI and 1.8x DPI. Against QQQ—the public-market equivalent CMU assigns to venture—even top quartile falls short; only top decile consistently clears it, prompting his blunt answer on risk compensation: “Absolutely not.”
3. Early-stage edge requires both non-consensus picking and workable fund math
Ron’s question for any new endowment or family office seeking technology exposure is whether it can access top-decile managers. Public markets offer the alternative immediately; venture only makes sense if the allocator can plausibly outperform that exposure after illiquidity, fees and selection risk.
Harry’s pushback targets the favored $50 million-$100 million seed fund. With average seed rounds around $4 million-$5 million, meaningful ownership can require $3 million-$3.5 million checks; making roughly 30 such investments consumes about $90 million-$100 million, leaving little room for additional diversification, while $1.5 million “tweener checks” rarely win the best competitive rounds.
Ron answers that consensus seed deals have become hostile terrain because multi-stage firms possess cheaper capital and can deploy $5 million-$10 million where seed funds historically wrote $2 million-$3 million. The remaining edge is non-consensus founders and ideas, where rounds may be less competitive and ownership cheaper—though Harry questions whether truly non-consensus pricing still exists.
Asked to weight access against picking, Ron assigns scaled multi-stage firms roughly 70% access and 30% picking; for small, nimble early-stage managers, he flips the balance to 70% picking and 30% access. CMU’s workable fund range starts around $80 million and extends toward $400 million-$1 billion, with its commitment size at the low end around $10 million.
4. Selling is venture’s newest pillar, and marks cannot substitute for exits
Ron defines the five venture capabilities as “sourcing, picking, winning, helping, and selling.” Selling is the newest institutional muscle; he regards Union Square as unusually disciplined, with an explicit process between years eight and 12 that prepares founders for active sales.
CMU prefers cash distributions because a manager can sell an entire position simultaneously for all LPs. Distributed stock creates timing differences among LP sales and potentially a 1%-2% pricing discrepancy.
Managers plainly failed to sell enough in 2021, but Ron preserves the context: median software ARR multiples reached about 20x and top-quartile growers 40x. Public comparables made investors believe holdings could appreciate another two or three times within three years—until the benchmark changed abruptly.
CMU now independently underwrites the top 10 company NAVs for every new manager and re-up. Accel and Sequoia are cited as conservatively carrying securities at 20%-30% discounts; at the other extreme, one 2023 manager still held OpenAI at $13 billion and promised to revise its valuation policy only after CMU challenged the mark.
5. Elite brands source by default, but picking skill appears in rejected ideas
At premier Silicon Valley and London firms, Ron sees no special systematic sourcing engine: powerful partners and enduring brands simply become mandatory calls before a founder signs a term sheet. Automated sourcing may work better in neglected niches such as bootstrapped Australian companies or businesses outside conventional hubs.
Below the flagship tier, his honest assessment is that sourcing contains “a lot of luck.” Managers hustle, accept introductions, take large numbers of meetings and occasionally encounter an exceptional founder; Harry agrees that this favors people willing to keep “pounding the pavements” and show up at 7 a.m.
Harry invokes Mike Maples’s search for businesses “going against the grain of the universe” and names Cyan Banister’s ability to see what others dismissed; he gives his own initial reactions—that Uber and Airbnb sounded absurd—as examples of why genuine category creation can look dysfunctional before it looks inevitable.
Ron says track record alone is insufficient. CMU reconstructs what a partner believed at the moment of investment and asks founders whether nobody else returned their calls before one investor saw “a blink in their eye” and believed early.
6. Manager diligence is partnership underwriting over decades
For a new fund, CMU targets at least 20 reference calls. Only about five come from the GP, and Ron calls those the “worst” because they predictably flatter; CMU seeks outside references to reveal conduct and partnership dynamics.
The diligence focus is less whether outsiders admire the stated strategy than whether a partner has mistreated people and whether the team actually functions. Asked why venture partnerships break, Ron gives two recurring causes: “Incentives” and “who’s working the hardest?”
Partnership turnover over the prior two years exceeded everything Ron had seen in eight years as an LP. Wealthy partners no longer wanted broken cap tables, founder transitions and illiquid portfolios; newer partners saw expected carry evaporate and, in some cases, found themselves earning 70% less than anticipated.
CMU views clean spinouts as rarer than the market implies and rebuilds partner-level attribution rather than accepting reassigned wins after the original deal lead departs. Ron says about half of new funds are not backed immediately: some are watched for six to eight months before a commitment that year, while others are deferred one or two funds. A 15-18-year fund backed across three vintages can become a 25-year relationship.
7. LP construction and endowment arithmetic impose their own constraints
A resilient LP base mixes endowments, foundations, family offices, founders and perhaps venture-fund GPs. No investor above 10% is ideal; Ron can accept a long-term-aligned 10%-30% anchor in a smaller fund, but concentration beyond roughly 30% becomes dangerous.
Endowments typically draw about 5% annually for campus spending and face roughly 3% higher-education inflation, requiring an 8% return merely to preserve purchasing power. The discussed tax of around 8% affecting five institutions could encourage a draw reduction from 5% toward 4.5%; without one, purchasing power comes under pressure.
Managers should spend as little time fundraising as possible, because “you make your money investing.” One close is ideal, but the practical requirement is a crisp schedule, secured LP commitments and legal work completed before each announced close.
Selling part of the management company is a “massive red flag.” The partnership’s magic is carried-interest alignment; transferring that economics to a silent owner means the people taking 100 calls a week must surrender returns to someone who is not doing the work.
8. Megafund math demands exits on the scale of the whole market
Ron models a live $7 billion platform containing roughly a $1 billion early-stage fund, a $2 billion-$3 billion growth fund and a larger opportunity vehicle. Because LPs invest proportionally, their dollar-weighted ownership falls from around 15% early-stage to 6%-7% growth and 2.5%-3% opportunity—about 5% overall.
Dividing $7 billion by 5% produces $140 billion of enterprise value at entry. To achieve CMU’s 4x net target after fees, the portfolio needs at least 6x gross, or approximately $800 billion of exit value; all IPO and M&A exit value in record-setting 2021 totaled roughly $850 billion.
Harry’s pushback—worth keeping—is that future outcomes may be much larger, with OpenAI, Anthropic, SpaceX and Stripe presented as potential $100 billion-scale companies. Ron says “we could be wrong,” then cites only 11 venture-backed IPOs above $50 billion, with Facebook in 2012 and Alibaba in 2014 still the largest examples he identifies.
His simpler sanity check: owning 10% of a generational $20 billion-$25 billion Figma yields about $2 billion before carry, roughly 0.3x of a $7 billion fund before carry and about 0.2x after 20% carry—“you need 15 Figmas.” Index is his scale exception because it preserved performance, held major stakes in Figma, Dream Games and Wiz, and even reduced its latest fund.
9. Fee income can break alignment before investment performance does
Stacking successive $3 billion, $5 billion and $7 billion funds creates roughly $15 billion of fee-paying capital. Ron estimates about $300 million in annual fees, often concentrated among five or six partners, and calls these GP businesses “some of the best high-margin businesses ever created.”
His objection is not that GPs pursue the opportunity; it is that $100 million growth checks enter mature, staffed, “well-oiled” businesses while charging 2-and-20. Harry counters that intrusive early-stage investors often damage companies by forcing enterprise sales, new products, premature scaling and extra fundraising.
Ron preserves premium economics for genuine early-stage work: roughly 2.5-and-20, or 2.5-and-30 and even 3-and-30 for exceptional long-term performance. Scaled growth capital should move toward 1-and-10, zero-and-10 or budget-based fees, although he admits managers able to raise without concessions may simply say, “Thanks for the advice.”
CMU does not re-up from loyalty alone, and Ron attributes brand-driven allocations to career incentives: “No one gets fired buying IBM.” It also underwrites deployment promises literally; returning after two years is acceptable if two years was promised, while slower pacing can be wise—Mark Suster’s 2021 strip sales left his LPs with realized DPI.
10. Venture’s liquidity drought is fundamentally a pricing problem
Ron calls the fundraising market brutal: US venture fundraising was on track to be the lowest since 2017 and Europe’s since roughly 2016. The primary cause is three years without enough distributions.
From 2002 through 2004, IPOs raised more public-market dollars than during 2022-2024 despite venture later becoming roughly 10 times larger. Even after the dot-com collapse, which required 13 years for QQQ to recover to its peak, the three years afterward produced more IPOs.
“I am not a believer ever that the IPO markets are closed. It’s purely a function of price.” Ron contrasts private investors demanding eight or 10 times ARR for a $100 million-ARR SaaS company growing 15% near breakeven with Microsoft’s 14% revenue growth, 17% earnings growth, GAAP profits, buybacks and deep moat.
Harry doubts private equity will rescue mediocre companies, many of which grow 10% and remain unprofitable. CMU therefore requests trending revenue, gross profit and free cash flow for each fund’s top 10 NAVs, accepting that an unforeseeable right-tail outcome may still overwhelm any present valuation.
11. Secondary sales can dispose of the very tails venture exists to own
Ron interprets Harvard’s reported $1 billion sale in the context of a roughly $50 billion endowment: meaningful, but potentially a portfolio refresh rather than capitulation. The harder question is whether selling mature funds eliminates unexpected late optionality.
CMU committed to one fund in 2012 that, by year 13, held a single remaining asset worth under $1 million to the endowment and absent from its main monitoring system. That asset was Circle, marked around a 30% discount to a roughly $5 billion last round before reaching about $50 billion publicly—adding roughly three turns to the old fund.
CalPERS reportedly bought about $500 million of Yale’s portfolio, including a General Catalyst position whose largest holding was Circle; within roughly two months, Circle alone created a $100 million write-up. The reported 10% discount looked exceptional to Ron, who expected a larger discount.
His concession is candid: CMU might also have sold Circle after concluding little juice remained. But if many endowments simultaneously offer 10% of their venture books, limited secondary capital turns the market into straightforward supply and demand, creating pricing pressure.
12. A 2026 exit wave can reopen liquidity without repairing three lost years
CMU’s venture portfolio became self-funding this year for the first time since 2021, amid liquidity involving Dream Games, Figma, Revolut, Circle, CoreWeave, Hinge Health and Chime. Much announced liquidity remains pending: Ron expects Wiz after regulatory approval in Q1 2026, while Figma and Dream Games still faced their respective processes.
Private capital was paradoxically cheaper than public capital for years, allowing top companies to avoid earnings calls and listing requirements. With public investors now assigning strong multiples to Circle, Nebius, CoreWeave, Palantir and Cloudflare, Ron tells venture managers: “Now is the time. Please take your companies public.”
Even a strong 2026 would not fully normalize fundraising. “One year is not going to solve the industry’s problem”; multiple years of distributions are necessary, although the first wave would unquestionably help LPs resume commitments.
Google, Microsoft, Amazon and Meta collectively generate about $600 billion of annual operating cash flow, giving them more incentive to make strategic acquisitions than repurchase another 50 basis points of stock. Approval of Wiz could provide a green light, but 12-month reviews are perilous in AI: products can become obsolete while buyers remain exposed to multibillion-dollar breakup fees.
13. China’s historic upside now comes with a structural alignment problem
CMU’s best-ever fund came from China and returned more than 20x net. Ron still praises the intelligence and work ethic of long-standing partners there, but says the bar for new investment has become extremely high.
A US executive order prevents US dollars from entering specified Chinese AI, semiconductor and defense companies. Meanwhile, managers traditionally raised parallel USD and RMB funds—often investing pari passu—but RMB vehicles backed by local governments can now access deals prohibited to USD LPs, creating a direct exposure and alignment problem.
Many strong Chinese founders are choosing the US, Singapore or London, reinforcing Ron’s description of China as “a hard market today.” CMU remains agnostic between sector specialists and generalists; despite openness to both, it added only one new sector-focused fund over roughly three and a half years.
14. AI’s economic promise does not remove financing and cycle risk
Ron expects AI to resemble railroads, automobiles, electricity and the internet: a transformative technology accompanied by a bubble that eventually pops. OpenAI’s unit economics are improving, but a stated $5 billion-$10 billion annual burn, two enormous rounds within 12 months and a roughly $70 billion-$80 billion financing stack leave it dependent on capital: “The music will stop eventually.”
Google and Meta entered public markets with roughly 30%-40% GAAP operating margins, while SpaceX and Starlink have reached escape velocity. “SpaceX cannot be killed”; by contrast, Ron will not call OpenAI or Anthropic independent trillion-dollar companies five years out because losing access to new equity could remove control of their destiny.
A major GDP effect looks more plausible over 10 years than three or five. Hyperscalers are expected to deploy about $1 trillion of capex from 2024-2027, alongside a US venture run rate near $100 billion with perhaps 80% touching AI; if productivity takes a decade to arrive, “there will be a lot, a lot of pain.”
Harry argues that belief in AI implies owning Nvidia. Ron does not call the current business overpriced, but stresses its historical cyclicality: a possible 20%-30% revenue decline, 40% earnings decline and compression from roughly 38x to a 24x trough multiple could produce a 70% drawdown. It is a scenario, not his one-to-three-year prediction.
15. Hard coaching and founder references matter more than founder-friendly branding
Ron compares useful investors with demanding football coaches: criticism feels bad but can improve performance when it comes from aligned people. Founders inevitably have weaknesses, so “tough conversations aren’t bad things”; CMU does not screen for whoever can claim to be the most founder-friendly.
The fundraising lie he hears in nearly every other introductory meeting is: “This is the perfect fund size for us…we’re never going to raise a bigger fund.” The supposedly permanent $300 million-$400 million cap is, in Harry’s phrasing, “99.9%” fiction.
GP commitment is one of CMU’s two strongest quantitative forward indicators, but the nominal amount matters less than what it represents to the individual. Ron names Kevin Hartz and A* Capital as an underrated combination of founder scar tissue, operating experience, premier access and a right-sized fund—then jokes that Hartz should not use the endorsement to raise $1 billion.
Ron has changed from over-indexing on historical data to treating venture as a people-driven business. CMU reconstructs attribution and asks founders, “Why did you choose that partner, why did that partner choose you?”; the fund he most wishes CMU owned is Union Square.