AI Risk Is Everywhere and It's Making Billion Dollar Funds Nervous
I've been pushing this point for a while, but investors and fund managers simply don't have the appetite for more AI investments. Anyone who has invested already is already in, and more exposure doesn't make sense. Either you are allowed to go all-in and you already have, or you have a responsibility to diversify:
> But for Monte Tarbox, the pension funds’ chief investment officer, ... his job is to diversify, and right now he sees AI everywhere he looks. > “I said to the manager, ‘We have to pass on this precisely because we don’t know when enough is enough or when it’s too much,’” he said, referring to the current AI cycle. “When that day comes, we don’t want to be hung out to dry and somebody says, ‘Well, why did you say yes to everything?’”
> Tarbox’s predicament is representative of almost every major institutional investor on Wall Street and beyond just now. Managing vast piles of cash first and foremost comes down to spreading it out across assets to avoid concentrated risks. But the depth and breadth of the AI revolution has turned the usual diversification they rely upon into an illusion.
For the people that have already invested, there's no strong reason to pivot to frontier labs as they make the lowest profit margins while companies up the supply chain make more, e.g. data centers/inference providers, chip manufacturers, and memory fabs.
People who need to avoid have a hard time avoiding, which means the demand for AI debt and investments is necessarily limited:
> About 40% of the S&P 500’s total market value is comprised of AI infrastructure companies including chipmakers and hyperscalers, according to an estimate by Goldman Sachs Group Inc. Shares of just three chipmakers make up more than a quarter of the benchmark for emerging markets. AI accounts for nearly half of all investment-grade bond issuance this year and 87% of venture capital funding, data compiled by Apollo Global Management show.
But this is rather interesting evidence pointing to the fact that the broad market is not blindly invested to AI.
> “At some point you’re going to have enough infrastructure. And we’ve seen this a hundred times over history. Whether it’s railroads, whether it’s the internet. It’s going to happen, it’s just a question of when.”
However, as we see with the Nscale S-1, it's abundantly clear that infrastructure business is doing quite terribly. People can argue that infrastructure companies have larger operating losses in the beginning, and this is true. However, certainly within reason of no more than 100%! Nscale has a spectacular 2308% net loss in 2025, showing the complete unprofitability of renting GPUs for AI inference and training.
> CIO Jonathan Grabel subsequently led a review of LACERA’s [Los Angeles County Employees Retirement Association] investments... they estimated that between 8% and 19% of the pension’s holdings were exposed to AI.
This is certainly significant for pension funds, but this also means plenty of capital is not available for more investment, meaning appetite for hyperscalar bond issuance and etc. has clear limits. It is still my stance that neither OpenAI or Anthropic will be able to IPO for this very reason - insufficient demand from the broader market. Those who can be exposed are already exposed to better investments and those cannot be exposed more are locked.