NVDA and Private Credit Part 2

NVidia + AI Growth

Jumping right in this week. I’ve clearly been far too risk-averse when it comes to Nvidia. I’ve been concerned that their margins will be eroded by competitive forces like in-house design at Meta, Google, Amazon or Microsoft. While that may happen eventually, that’s not something that this market is concerned with at the moment.

I’ve started to wonder if NVidia’s actual advantage is less their technology and more that they have gobbled up all of the available physical resources needed to support the current configuration of top-tier AI. It turns out that building new power plants or substations is much harder than building new AI data centers, for example. Competitors can’t compete if they can’t access the physical resources needed to participate in the AI race.

There is a point when a company like NVidia becomes priced to perfection. I won’t prognosticate about where and when that happens. Besides the fact that Nvidia is in first place in the AI race, the links between NVidia’s sustained profitability and the overall AI economic opportunity are not clear-cut and dry to me. Leopold Aschenbrenner’s recent interview with Dwaresh Patel presented some interesting numbers to support the overall AI opportunity.

Let’s break down how AI tools can generate a staggering $100B in incremental new revenue. Consider the 300 million Microsoft Office subscribers, which includes CoPilot, their AI model. If 1/3rd of these users opt for a next-gen AI product at $100/mo, that’s a whopping $100B in incremental revenue. This justifies investments in AI superclusters and the necessary power infrastructure to support it.

If 1/3rd of subscribers paying $100/mo sounds like a lot, consider how advanced these tools can become. ChatGPT costs $20/mo because it’s still relatively primitive compared to the opportunity. Leopold claims that GPT-4 with 100 tokens per minute is roughly equivalent to him thinking for three minutes. When AI models can process millions of tokens in a single problem-solving session, theoretically, that could translate to an output equivalent to months of work, not just minutes. AI models will eventually incorporate planning, drafting, critiquing and error-correcting, producing dramatically better outcomes. In that context, GPT-4 seems primitive. An additional $100/mo is a bargain if it enables an $8k/mo employee to produce months of incremental work.

I don’t know where NVidia goes from here. The economic impact and growth potential of AI are undeniably significant tailwinds. I’ll also add that the AI growth theme is too substantial to care about 5% per year base interest rates, so rate cuts, rate hikes or higher for longer really have no impact, but those rates will impact other assets and sectors.

Private Credit – Part 2

I wrote last week about the overall context of private credit. I like the framing that private credit is a portfolio of loans funded by private investors. Private credit risk boils down to whether the managers act as responsible stewards of capital. That said, some issues are starting to emerge. Here are some risks that will eventually sour the overall reputation of private credit.

• PIK stands for payment-in-kind or synthetic payment-in-kind. It occurs when the borrowers cannot afford their interest payments, so lenders accept additional debt or equity added to their balance in lieu of payment. Intuitively, this should seem problematic. It’s happening now, and it’s not transparently reported to investors. These would be the first questions I’d ask about. Each fund has different concentrations of PIK.
• Flawed valuations. Are these loans worth what their owners say they are? Some funds may be justifying higher valuations with questionable methods to prop up their values and keep investors happy. There are incentives for managers to misrepresent the value of their loans.
• Leverage. What’s worse than a bad loan portfolio? Answer: a bad loan portfolio with leverage. Many private credit funds are levered 100%. Not all private credit funds use leverage.
• Historical Justifications. This is the blind leading the blind. Private credit producing X% over various historical periods tells us nothing about the quality of the loans or the economic environment supporting those loans. You have to look forward even though the future is uncertain and complicated. There’s no way around that exercise. Sales tactics that use historical performance justifications to sell these funds should make you uneasy.
• Liquidity and Duration Mismatch. Many private credit funds have loans with 5-7 year durations but offer liquidity quarterly. The problem with bank loan risk is that they fund 5-7 year loans with daily liquid deposits, which enables bank run risk. Shifting the liquidity from daily to quarterly does not solve the problem. Realistically, investor durations need to better match loan durations.
• Defaults within Private Equity. A majority of private credit loans, by volume, are to private equity borrowers. Many funds say they only lend to private equity-backed companies for financial safety. It may seem like private equity is immune to losses given their incredible run over the past ten years. On Friday last week, Vista Equity Partners wrote down the value of PluralSight, their portfolio company, from $3.5B to $0. Many large private credit managers own PluralSight loans with leverage. What will the value of those loans be post-restructuring, and how many other private equity portfolio companies are hanging on by a thread? Owning exclusively senior debt does not solve the bankruptcy risk.

Fortunately, some private credit funds address these issues better than others. No fund is perfect, and every investment comes with unique and complicated risks. I feel that niche managers can do a better job avoiding these issues.

(Additional disclaimer: NVDA and private credit comments should not be interpreted as financial advice – talk to a professional directly)