The Rising Role of AI Infrastructure in Retirement Portfolios

Nowadays it is hard to have a conversation in the retirement space without talking about the two largest trends reshaping our industry: the convergence of public and private markets and artificial intelligence, or AI. While we have typically viewed these trends in isolation, it is worth exploring how they intertwine.

A few weeks ago, Nvidia announced a partnership with multiple private asset managers, including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, to establish AI compute infrastructure financing platforms to mobilize more than $500 billion in capital.

This news is a big deal, but possibly not surprising. It represents a continuation of Nvidia’s business strategy to finance an ecosystem of neoclouds and open-source models that increases its chip and infrastructure dominance, while reducing the risk of only one or a handful of buyers purchasing their products.

The most striking aspect of these agreements is the attempt to make graphics processing units, or GPUs, and AI computing infrastructure a financeable asset class.

Traditionally, microchips have been bad collateral for long-dated debt financing. Moore’s law observes that the number of transistors in a chip doubles every two years, implying that old chips would need to double their efficiency or halve their price to compete. Thus, the value of an old chip declines precipitously, making its useful life quite short. Furthermore, new technologies might change chip requirements, necessitating full redesigns that further shorten their lifetimes.

How, then, is Nvidia proposing that private asset managers finance this as infrastructure?

For starters, Nvidia isn’t focusing only on financing chips, but data centers—or as they call them, data factories. Nvidia not only manufactures chips but also designs systems and networking solutions. In other words, Nvidia’s relevance in the AI buildout is so disproportionate that its products account for more than half of a data center's bill of materials.

Second, data centers have been shown to pay off their construction costs extremely quickly. For example, SpaceX’s data center costs are being recovered in a single year by leasing computing capacity to Anthropic. The expected lifetimes of these data centers and their servers are materially longer.

In fact, hyperscalers—Big Tech companies—have increased the depreciation cycles of their servers and GPUs to between five and six years. Some will argue that they’re trying to reduce their depreciation expenses by spreading the cost over more years; however, this increase in lifespan seems to reflect current market utilization trends.

A final element that private asset managers have most likely been paying attention to is the secondary markets for servers and chips. Surprisingly, contrary to history and market predictions, compute rental prices continue to rise due to a lack of supply and strong demand.

This phenomenon makes GPUs and servers an excellent collateral for debt.

As we have seen, Nvidia has a clear interest in lowering the costs of financing the AI buildout and making its products more accessible to its clients. Nvidia also has a compelling argument to make this buildout financeable with debt.

But what’s in it for the asset managers?

Since the Department of Labor’s recent guidance on alternative assets in retirement plans, private asset managers have been expanding their offerings into the defined-contribution space.

As they grow assets in the retirement arena, they will need new sources of deal flow to put this money to work. Building these new financing platforms enables them to originate more deals in the artificial intelligence infrastructure space.

To get a sense of the magnitude, I did some back-of-the-envelope calculations to gauge whether these AI opportunities could represent a significant source of new deals. The combined size of these private managers’ credit platforms is approximately $1.8 trillion. Introducing $500 billion in incremental new deals would represent roughly 21.5% of their portfolios, a sizable amount.

Of course, this is a crude measure: assets may continue to rise, managers could syndicate parts of these deals, asset deployment will happen over time, and some of these investments could be structured with equity.

Even so, these calculations indicate that AI financing can be a significant source of new private credit originations.

Private managers are also incentivized to hedge their portfolios. Earlier this year, we saw significant declines in the stocks of private asset managers and publicly traded business development companies due to concerns about their exposure to software. Investors in interval funds also rushed for the exits, forcing managers to impose gates on quarterly redemptions.

Given investors’ concerns about software, surely, they should welcome more AI exposure in their portfolios to counterbalance. Yet, the market remains uncertain about AI’s prospects as the unwind of the AI trade over the summer seems to show.

Nevertheless, for these private asset managers, it is economically rational to hedge their exposures to software by also investing in the other side of the technology disruption.

It should be noted that these managers are no strangers to investments in AI. Multiple deals have occurred over the past few years in which private credit has financed parts of this buildup.

To name a few examples, Blackstone lent to CoreWeave in 2023, Blue Owl and Meta created a joint venture to build the Hyperion data center in 2025, and Apollo provided credit to lease chips to xAI earlier this year.

Taken together, I expect these private funds to hold more AI-related investments.

Will we see them as infrastructure deals expanding beyond the traditional energy investments we’ve mostly seen over the past decade?

Will they show up as asset-backed loans or in private credit deals?

I don’t know, but it seems clear that AI exposure, as a percentage of the total pie of assets in these funds, will be higher going forward.

In an interesting twist of fate, if we truly believe Dario Amodei’s and other AI pundits’ admonitions that AI will take away all jobs, then a natural hedge against this risk would be for workers to invest in the data centers that power AI.

Sounds like the perfect marriage between private assets and AI.

Just to be clear, I don’t think there won’t be any jobs left to do because of AI, but what do I know? I am just human.