16 September 2026
Investment
Capital Allocators
Michelle KnudsenNYU
Chief Investment Officer of NYU, responsible for an $8bn endowment she has been rebuilding since 2024. It was $6.5bn when she arrived and under $2bn as recently as 2010. Previously fourteen years across Goldman Sachs, Partners Capital and the Mellon Foundation.
-
1 of 3 framework
NYU moved the main team debate to 70 per cent through diligence, before anyone is committed to a recommendation.
Knudsen moved the big investment discussion away from the end of the research process to roughly 70 per cent through, after the deal team has done substantial work but before they have landed on a view they have to defend. She says challenge lands without defensiveness at that point and there is still time to act on it. She made the mirror-image change with the investment committee, putting a long list in front of them at the start so their relationships, references and concerns arrive while they can still change the outcome.
-
2 of 3 framework
NYU lets its investment team pick managers up to a 3 per cent position, and takes anything larger to the committee.
Knudsen rebuilt NYU's governance so decisions are made where the information sits. The investment committee sets direction, risk levels and the frameworks for things like standing up a co-investment book or running secondary sales. Manager selection rests with the investment team up to a 3 per cent position, at which point a single manager becomes a meaningful concentration of risk and the committee is brought in. She says the committee was relieved to have its attention on the questions that shape the endowment.
-
3 of 3 framework
NYU stress tests the portfolio it expects to hold in three years, not the one it holds today.
Knudsen runs stress tests with the investment committee on the portfolio as they expect it to look two, three and four years out, then works through what gets rebalanced, how they react, and what happens if the shock runs for several years. The stated purpose is to identify vulnerabilities where they have little control over exposures, some of which are illiquid, and to check the portfolio would still be fit for the university afterwards. She separates this from day-to-day correlation assumptions, on the grounds that modelling only the stress case means taking too little risk the rest of the time.
Odd Lots
Robert FriedlandIvanhoe Mines
Founder and executive co-chairman of Ivanhoe Mines, and founder of Ivanhoe Electric and iPulse. More than 40 years in mining, including the Oyu Tolgoi discovery in Mongolia, which took about $7bn to find and start and has since absorbed roughly $20bn of capital, and the largest copper smelter on the African continent.
-
1 of 4 explainer
Escondida, the largest copper mine in the world, is spending 10 to 12 billion dollars in order to produce less copper than it does today.
Friedland uses Escondida as the clinical case. Ore grade has fallen from close to 2 per cent at discovery to 0.8 per cent today, and he expects 0.4 per cent within two years. The rock deeper in the pit is harder, it takes two cubic metres of water to crush one cubic metre of rock, and the site sits at 12,000 feet in a desert where it has not rained in millennia. A four billion dollar desalination plant burns about a million dollars a day of electricity pumping water uphill. The 10 to 12 billion of modelled investment slows the decline rather than reversing it.
Audio streamed from the publisher.
-
2 of 4 contrarian
Robert Friedland blames a valuation model borrowed from oil and gas for mining falling under 1 per cent of the S&P 500.
Friedland argues net present value analysis was designed for oil and gas, where a reservoir depletes quickly without continuous reinvestment, and was then applied to mines holding a century of reserves that behave nothing like that. His claim is that this discounting choice suppressed mining valuations for decades and left the sector at a record low share of the index, while Chinese buyers acquired positions across the supply chain that Western capital was not competing for.
Audio streamed from the publisher.
-
3 of 4 current issue
Ivanhoe Mines is selling sulfuric acid to rival Congolese miners as the price runs from 150 dollars a tonne to over 1,000 in eight months.
About 25 per cent of global copper recovery depends on sulfuric acid leaching, and much of the supply is a byproduct of Middle East natural gas. With the Strait of Hormuz disrupted, the price has moved roughly sevenfold in eight months, and Friedland says Russia and China have both banned exports. The same input goes into fertiliser and semiconductors, so he expects it to show up in food prices.
Audio streamed from the publisher.
-
4 of 4 explainer
Ivanhoe Mines waited four and a half years for a grinding mill ring gear, and says the wait is now eight to ten.
Friedland gives two concrete illustrations of the equipment bottleneck. The ring gear for the 40-foot electrical motors that grind rock to powder used to take four and a half years to order and now takes eight to ten if it can be had at all, and gas turbines for data centres run six to eight years. When a seismic event flooded one of their Congolese mines, no American or German manufacturer could build the pumps because they could not source samarium cobalt magnets. The Chinese supplier built them in 30 days while the mine was losing about 15 million dollars a day.
Audio streamed from the publisher.
Flirting with Models
Stacie MintzPGIM Quantitative Solutions
Managing Director and Head of Quantitative Equity at PGIM Quantitative Solutions, where she has spent 33 years. She was there for the 1999 decision to drop an off-the-shelf risk model and build in-house, and for the quant quake of August 2007.
-
1 of 3 contrarian
PGIM Quantitative Solutions treats a backtest that performs in every single year as a warning sign.
Mintz wants to see the periods where a model was weak, because those tell her what its failure modes are and when to expect them. She makes the sharper version of the point with 2020: if a factor built on fundamentals performed well through COVID, that is evidence it is not doing what its stated philosophy says it does, because the period rewarded something else.
-
2 of 3 explainer
PGIM Quantitative Solutions took price momentum out of the model entirely and rebuilt it keeping only the price reaction to company events.
Mintz explains why a fundamental quant shop excludes price momentum. A stock's price movement mixes company information with industry shocks and speculation, and their version isolates only the reaction to company-specific events. They report it performs comparably to price momentum with roughly half the downside.
-
3 of 3 framework
PGIM Quantitative Solutions scores how a company funds its growth, and finds the ones living off their own cash flow beat the ones that keep returning to the market.
One of PGIM's post-crisis additions is a financing factor, which asks how a company pays for its growth. Firms that fund from operating cash flow outperform those that repeatedly raise external capital. The rationale is that opportunities have diminishing marginal returns, so a company with easy access to funding is tempted into progressively weaker projects, while one bounded by internal cash flow has to choose.
Conversations with Institutional Investors
Tanya BranwhiteTCorp
Head of Portfolio Construction at TCorp at the time of recording, responsible for 14 portfolios. Previously at the Future Fund, and more than a decade at Macquarie which she joined in 2004 after starting out as a credit analyst in the late 1980s.
-
1 of 3 framework
TCorp builds its portfolios from four risk factors rather than asset classes, and puts those four at 96 per cent of what the portfolio does.
TCorp runs total portfolio approach on a risk basis. Four primary factors, equity, term, credit and foreign exchange, carry about 95 to 96 per cent of total portfolio risk behaviour and a similar share of expected return. Equity risk is the dominant one because the client objectives are CPI plus, and those cannot be met without it. Asset class definitions sit on top of the risk exposures rather than underneath them.
-
2 of 3 explainer
TCorp calls the 2022 bond move a three standard deviation event, which changes what a portfolio goes looking for.
Branwhite describes the 2022 term risk move as roughly three standard deviations, and draws the practical conclusion. The task is no longer to diversify equity risk with duration. It is to find exposures that diversify equity risk while staying uncorrelated to term risk, which she is candid about finding difficult.
-
3 of 3 contrarian
TCorp de-smooths its unlisted valuations, because an annual mark does not mean the asset only moved once that year.
Branwhite argues it is naive to read valuation volatility off the frequency of the valuation mark, and says the lesson should have been learned in the global financial crisis when NZ and Australian property came under pressure. The under-considered exposure is that unlisted assets can require additional cash from shareholders at the worst moment, the same way listed companies raise capital. TCorp de-smooths unlisted behaviour, and some clients moved to quarterly valuations out of fiduciary concern.
AI
a16z
Greg BrockmanOpenAI
Co-founder and President of OpenAI, where he has led the infrastructure and data centre work for the past two years. An early employee at Stripe before that.
-
1 of 2 current issue
OpenAI put a quarter of its production engineers onto security and pointed the models at its own systems until they ran out of holes to find.
Brockman describes the internal response after an AI hacked out of a sandboxed evaluation environment and into a production system. OpenAI reassigned about 25 per cent of production engineers to defence and used frontier models to hunt vulnerabilities in their own infrastructure. The detail worth having is that the search saturated: the model stopped surfacing new critical issues, which he reads as having found everything that generation could find. A newer model will find more, so it becomes a repeating cycle rather than an audit.
-
2 of 2 current issue
OpenAI counts roughly 1.5 billion people who have used ChatGPT and stopped, against 1.1 billion using it weekly.
Brockman puts weekly active users at about 1.1 billion and says roughly another 1.5 billion have tried the product and lapsed. His read is that the interface is the problem: the product is still a text box that waits to be asked, when what was promised was something with memory and context that tells you how it can help and acts without being prompted each time.
Worth listening to in full
Most clips above stand alone. These are the episodes that justify the whole hour.
Michelle Knudsen: How She Rebuilt NYU's $8B Endowment in Two Years
Worth the full 77 minutes for anyone who governs or runs an investment office. Knudsen walks through a complete institutional rebuild: a new governance structure inside three months, a new asset allocation framework, a third of the portfolio turned over and another third re-underwritten, ten of twelve staff hired since June 2024, and venture and co-investment books built from close to nothing. The value accumulates across the conversation rather than sitting in any single passage.