Miles Dieffenbach says 60-70% of the S&P is tied to AI capex, and even value names like Nvidia and Micron are AI trades. At CMU, lower-middle-market private equity, industrials, bonds and TIPS are cleaner diversifiers than public equities. He sees almost nowhere to run within a broad equity book.
Miles Dieffenbach says multi-stage firms cannot use secondaries at scale. Selling half a stake might return only a tenth of the fund while damaging the relationship with a founder who is raising again, so they must chase hundred-billion-dollar IPOs for the math to work. That is why secondary liquidity is awkward for them.
Miles Dieffenbach says true early-stage venture backs ideas that look strange today, like Airbnb, before consensus capital will touch them. That creates a protective moat and fat right-tail returns, though he says only a few managers can execute it. CMU still wants exposure because nowhere else offers that optionality.
Miles Dieffenbach says public markets spent 12-18 months pricing AI risk, citing Nvidia's forward multiple compression and Broadcom's negative reaction to Anthropic ASIC exposure. Venture naturally avoids marking risk down because GPs are forced to keep the fundraising hamster wheel rolling. That delays distress but does not eliminate it.
Miles Dieffenbach says Carnegie Mellon wants roughly a 4X net return, or 6-7X gross, from venture funds. For a $6B-plus multi-stage fund, even Anthropic's $18B round returning 20X may only return about 1X net; it cannot multiple the fund enough. He questions whether LPs can pencil clean risk-adjusted returns.
Miles Dieffenbach asks whether a GP puts own cash into the fund and would feel real pain if it underperforms. He contrasts that with managers collecting $30M in fees while holding a free call option on LP money: heads they win, tails LPs pay. It is his cleanest alignment test.
Miles Dieffenbach argues top private AI companies route cost of goods and capex to hyperscalers and chipmakers that dominate QQQ, with overlapping large shareholders. He says it is hard to imagine public tech imploding while Anthropic or OpenAI continue ripping privately. That weakens the case for private growth as a diversifier.
Miles Dieffenbach says QQQ compounded about 6X over ten years, roughly a 19-20% IRR. Applying a 2-and-20 fee load would leave the same return as the S&P 500, stripping out the alpha. He calls modern mega venture large-scale growth equity. Its risk-adjusted fee structure is hard to defend.
Beezer Clarkson says demand for top-decile access creates scarcity, so allocators convince themselves they can find a wedge into venture. She agrees the go-forward average return may be worse and says average venture returns are not good; venture is not an asset class where you want to be average.
Miles Dieffenbach says CMU will not back a manager for only a fund or two if the job is a 15-year commitment. He checks whether they love the work, speaks with founders and past coworkers, and looks for enough GP commitment that leaving would hurt.
Miles Dieffenbach says a key investor departure gets deep scrutiny in CMU's re-up decisions, especially if that person drove performance. He also flags funds whose shape, form, or strategy has drifted from what was underwritten five to seven years earlier, making the old return math unworkable.
Miles Dieffenbach says early-stage funds can sell secondaries, but an opportunity fund complicates it. Taking money out of the early fund while putting new capital into the opportunity fund is a huge conflict; Nick asks how investing net new while selling an existing fund is not a conflict.
Miles Dieffenbach says multi-stage firms will keep capturing most big outcomes, and large sovereign wealth funds will accept fees if a $400M commitment also brings co-investment into the next Anthropic or OpenAI. Those LPs sign up all day for that access. It explains why mega-fund fee pushback does not stop every allocator.
