Miles Dieffenbach on Endowment Investing, Venture Returns and Liquidity

Harry Stebbings interviews Miles Dieffenbach, a managing director of investments at Carnegie Mellon University, about endowment allocation, venture fund selection and liquidity. His figures describe his own analysis and the market as discussed at the time; they are not a complete audited dataset or individual investment advice.

Dieffenbach recounts receiving a lymphoma diagnosis at 26, undergoing chemotherapy and later being told he was cancer-free. He describes exercise and a determined attitude as part of his personal experience. His story does not establish that exercise treats cancer or guarantees an outcome.

He says the experience changed his perspective on adversity and made him appreciate life. Stebbings asks whether it changed his benchmark for what constitutes a difficult day; Dieffenbach says the perspective has helped him through subsequent challenges.

At the time of the interview, he says Carnegie Mellon manages about $4 billion for the university, with an 85% equity and 15% fixed-income allocation. Across another classification, roughly 50% is targeted to private assets, with the rest in hedge funds and liquid assets. These are overlapping views, not percentages to add together.

He says private-portfolio distributions covered its capital calls over the prior three years, with buyouts contributing most and venture being the largest detractor. That describes the private book as a whole, not proof that every venture fund was self-funding during those three years.

Venture represented slightly less than 25% of the total endowment, he says, about five to ten percentage points above comparable endowments. He describes a flexible “best athlete” approach across private assets, seeking return relative to risk rather than fixed sub-allocations.

Dieffenbach compares relatively stable real-estate cash flows with an early venture fund backing unprofitable companies and new ideas. In his view investors should demand additional return for illiquidity and risk. The example is a framework; specific property and venture investments have different risks.

He argues many LPs have not been adequately compensated for venture risk. He cites roughly 8% median net IRR and 15% top-quartile net IRR for his analysis of mature vintages, with around 2.5 times a top-quartile value multiple. These are his historical dataset and selection, not universal or guaranteed results.

For 15-year mature vintages from 1998 through 2015, he cites roughly 1.8 times top-quartile DPI. DPI measures distributions relative to paid-in capital; TVPI includes distributed value plus remaining value, so the two are not interchangeable. Their different measurements and vintage windows must be kept in mind.

He uses a public-market equivalent, including Nasdaq-100 exposure through QQQ for venture, to compare private results with investing cash flows in public markets. The choice of benchmark, timing and methodology materially affect that comparison; the interview does not reproduce the calculations.

Asked who should invest in venture, he says prospective allocators should consider whether they can access top-decile managers and outperform their public-market alternative. Stebbings notes that past returns are backward-looking and do not directly evaluate first-time funds. Dieffenbach acknowledges emerging funds can be attractive but calls the market competitive and risky.

Stebbings challenges the economics of $50 million to $100 million seed funds, given larger rounds, ownership goals and diversification. Dieffenbach says consensus deals face intense competition from large multistage firms, while less obvious founders and ideas may offer better entry conditions. Neither approach guarantees access or returns.

Asked whether venture depends on access or picking, he gives different illustrative weightings for multistage and small early-stage firms. He sees access as especially important at scale and picking as more important for smaller funds. The percentages are his heuristic, not empirical attribution.

For Carnegie Mellon’s commitment size, he says funds from roughly $80 million to $1 billion can fit, depending on team, skill and track record. His range reflects this allocator’s needs, not a recommendation that fund size alone determines performance.

He lists five activities in venture: sourcing, picking, winning, helping and selling. He sees selling as a developing skill for the industry and says some managers have historically handled it better than others. He names Union Square as an admired example while disclosing that Carnegie Mellon is not an investor there.

Dieffenbach prefers cash distributions to stock where possible because different LPs may sell distributed shares at different times and prices. Stock distribution and cash proceeds are distinct: an LP receiving shares still bears price risk until it sells them.

On private portfolio marks, he says quality varies and describes underwriting the ten largest company values when evaluating or renewing a manager. He mentions conservative discounts used by some firms but does not establish that every book is accurately marked.

Stebbings worries some paper multiples may prove too optimistic. Dieffenbach returns to his approximate 2.5 times top-quartile TVPI and 1.8 times DPI figures as a reminder that unrealized value and distributions differ. These are historical summary figures from his analysis, not predictions for a specific current fund.

He argues established brands often see deals through relationships and reputation, while specialized or geographically remote strategies may use different sourcing tools. He also believes luck plays a role. Stebbings emphasizes the time and effort needed to build networks.

To assess picking, Dieffenbach wants to understand the original reasoning behind a decision and speak with founders as well as managers. He favors investors able to back non-consensus ideas, but acknowledges that judging skill before outcomes are mature is difficult.

He says his team examines partnership dynamics and incentives through references, rather than accepting only a polished account. Differing effort and compensation can strain partnerships, in his view. These are diligence concerns rather than findings about a particular firm.

On spinouts, he says Carnegie Mellon has historically backed few, and discusses exceptions. The firm may invest promptly in some managers and spend years building a relationship with others. He notes a venture fund can take 15 years or more to wind down, making a manager relationship unusually long and illiquid.

For fund managers facing LP churn, he favors a diversified investor base when available, recognizing that not every manager can choose. He says more than 30% from one LP can create concentration, while 10% or less would be preferable in an ideal case. These are guidelines, not guarantees of stable capital.

Discussing endowment spending, he uses an illustrative 5% annual draw plus 3% inflation to show why roughly 8% nominal return may be needed to preserve purchasing power. Actual spending, inflation, taxes and fund restrictions differ by institution; the arithmetic is a simplified example.

He sees selling part of a management company as a potential alignment concern because economics may go to a passive owner. Stebbings agrees that founders of investment firms should understand the long-term implications. The effect depends on terms and incentives, not simply the existence of an outside owner.

Dieffenbach worries that larger multistage funds may not reproduce the returns achieved when their funds were much smaller. Stebbings asks whether lower returns could still satisfy a different set of investors. Their disagreement is about fund economics and desired returns, not evidence of a particular fund’s future result.

He offers a simplified example: a $7 billion combined fundraising program with $1 billion in early-stage funds, $2 billion to $3 billion in growth and the rest in opportunity vehicles. For an LP investing proportionately across them, he estimates a roughly 5% dollar-weighted entry ownership using assumed stage-level stakes.

Dividing $7 billion by 5% gives about $140 billion in implied entry company value. He then assumes roughly 6 times gross is needed to reach 4 times net after fees and carry, giving a rough $800 billion-plus exit-value requirement. This is a simplified portfolio illustration, not an exact forecast: dilution, timing, fees, recycling, reserves, exits and actual ownership can change the calculation.

He compares that rough aggregate with his estimate of approximately $850 billion in venture-backed IPO and acquisition value in 2021, while noting exits would occur over multiple years. The comparison illustrates scale, not a requirement that all exits occur in one year or that every firm’s exit value is available to one fund.

Stebbings counters that future company outcomes could be much larger than historical exits. Dieffenbach agrees he could be wrong and that some private companies might reach trillion-dollar valuations, but wants a margin of safety rather than assuming many unprecedented exits. Both sides are discussing uncertain scenarios.

He praises Index Ventures’ portfolio and discipline on fund size. This is his assessment of a manager and several named holdings; the interview does not provide a complete audited fund-return series or prove the future performance of that portfolio.

He argues a very large fund can require many exceptional outcomes to return a meaningful multiple. The host tests whether the issue is size or the quality of an individual manager. Dieffenbach says he leans on the arithmetic and then examines the team and incentives.

He says management fees on successive large funds can create a lucrative business for the manager and may weaken GP–LP alignment. His illustrative fee calculation assumes a level of assets and fees; actual fee bases, step-downs, expenses and carried interest depend on fund documents.

He argues mature growth funds should charge less than smaller early-stage funds because their work resembles long-only investing more closely. Stebbings counters that active interference by early-stage investors can itself harm founders. These are competing views about services and fees, not a finding that any manager is passive.

Dieffenbach suggests lower management fees and carry for large growth vehicles, while accepting higher fees for exceptional early-stage managers. He recognizes that managers with strong demand may have the bargaining power to maintain current terms. His proposed structures are preferences, not current industry requirements.

He says Carnegie Mellon does not automatically renew a manager solely out of loyalty after a successful fund; the new strategy and size matter. He warns that LPs may sometimes favor recognizable brands for career safety, while describing investment committee review of returns, fees and risks. This is his account of CMU’s process.

He stresses time diversification and whether a GP follows the investment period promised to LPs. Faster deployment can be acceptable if disclosed and underwritten in advance; a change from a three- or four-year plan to two years merits questions. Diversifying vintage exposure does not ensure a profit.

He attributes fundraising difficulty partly to weak distributions. He argues public markets are open at a price, and that a slower-growing software company competes for capital with profitable public alternatives. Stebbings is also skeptical that private equity will buy every struggling company.

Carnegie Mellon asks for trends in revenue, gross profit and free cash flow for top underlying positions, he says. The aim is to assess whether assets are durable rather than rely only on marks. He acknowledges that exceptional right-tail outcomes can still be impossible to predict in real time.

Discussing reported Harvard and Yale secondary sales, Dieffenbach says endowment needs and portfolio composition matter. He uses his own experience with a 2012 venture fund holding Circle to illustrate how a small marked position can later rise sharply. His estimates of valuation and fund impact are time-specific and not independently audited here.

He says a secondary seller may give up an unexpected late winner, even after careful analysis, while a buyer can benefit. He speculates about the decisions behind a reported Yale transaction but does not claim to know the complete portfolio or decision process. A mark-up is not the same as cash distributed.

He says Carnegie Mellon’s venture book has become self-funding again that year for the first time since 2021, distinct from the broader private book’s three-year record. He notes that announced transactions may still require approvals and closing before cash reaches investors.

Dieffenbach expects more liquidity in a subsequent year and urges venture-backed companies to consider public markets. He also says one strong year alone would not repair several years of weak distributions. Both the timing and the volume of exits are forecasts.

He argues high private valuations and capital needs can leave model companies vulnerable if financing conditions tighten. He uses OpenAI as an example and discusses cash burn and preference-stack figures, but those numbers are his estimates rather than verified financial disclosures. The possibility of failure is a risk scenario, not a prediction of a particular outcome.

On Nvidia, he describes a hypothetical downturn in which revenue, earnings and valuation multiples all fall, producing a large share-price drawdown. He explicitly does not guarantee the timing or occurrence. The scenario depends on multiple assumptions and is not a price target or investment advice.

Discussing “founder-friendly” investing, he says constructive, difficult feedback can help founders and that his team does not select GPs solely for that label. Stebbings agrees. Their preference for candid coaching does not justify abusive management or establish how any particular founder is treated.

In quick-fire questions he says GP commitments matter relative to each manager’s resources, not just as an absolute amount. He also favors checking who truly sourced and led prior investments, since attribution can change when partners leave. These are diligence heuristics, not proof of future fund returns.

He says he has shifted from relying heavily on quantitative data toward combining it with references and qualitative assessment of people. He names an admired fund in which Carnegie Mellon is not invested and cautions a manager he likes against raising too large a fund. These are allocator opinions with their own incentives.

He recalls questioning a manager’s 2023 private mark for OpenSea, after which the manager said it would review its valuation policy. He does not identify the manager or provide complete underlying data. The story illustrates his concern about marks, not an independent valuation of the asset.