NVDA Chips and Bad Banks

NVDA Chips

We don’t typically zoom in this far on a particular company, but I felt compelled to zoom in on Nvidia this week for a variety of reasons. Quick disclaimer: we may or may not hold shares in NVidia at any time. We may change our viewpoints at any given time. This is not advice. Nvidia has just 26,000 employees, making it the smallest team of the MAG 7 by a wide margin. Apple employs 161k and Google 182k. If you’ve been reading for a while, you’ve seen us write about Nvidia in the past because they’ve been the tip of the spear of the advanced semiconductor world for many many years.

Nvidia’s valuation today is stratospheric, but so is its growth rate. Pre-Covid, their sales were stagnating at around $10B per year. The Covid acceleration brought them up to $25B per year. The GPT acceleration has put them on track for $45B, and analyst estimates are up to $125B for 2027 (don’t put any weight on analyst estimates!) Their top AI chip, H100 chip, sells for roughly $30,000 per unit and costs approximately $3,000 to produce. Imagine being in charge of a company on track to top $100B in sales with an extremely high-cost, high-demand and high-margin product.

Last quarter, Nvidia generated $18B in sales and nearly $10B in profit, an incredible feat. They are currently trading at a valuation of $1.83T as of this week, but don’t count on that number if you’re reading this even three days from publication. The price has gone parabolic and continues to grow 2-5% daily, even on bad market days. That’s why I wanted to write about this. This has been a great ride for all Nvidia holders, but the valuation today is higher than the peak of the Covid bubble (32x EV/SALES v 40x EV/SALES today). The next earnings announcement is next week on Wednesday.

Given the momentum, the stock could double from here today. For long-term holders on this bus for a while, consider that there are other opportunities where your gains can be redeployed. At this price, buying into NVDA with fresh capital is purely a trader’s game and certainly not a capital preservation game.

CEO Jensen Huang has repeatedly said he runs the company as if it will run out of money within 30 days. That tension makes a perfect high-risk and high-return combination. Earnings must be consistent and high quality going forward to justify sky-high valuations. This is where the problem lies.

Sam Altman announced last week that he’s seeking to raise $7T to produce more AI chips. There is a talent shortage at the top of the chip world. To add some math to the size and scope of GPT, OpenAI has a first-mover advantage but not necessarily a moat around their GPT product. I use Google’s Gemini and ChatGPT interchangeably, and I’m happy with the results of both products.

It cost OpenAI around $100m, 50 gigawatt hours and six months of compute to create GPT-4. Loose estimates say each GPT iteration requires 30x more capital, power and compute. The exponential math of machine learning is how Sam arrived at $7T, and, of course, the marketing reasons. GPT-6 could cost $75B and require more power than the largest existing power plant. However, if GPT-5 is not materially better than GPT-4, the interest in creating the next model will dwindle significantly.

Back to Nvidia, this vertical rise in valuation happens when the most significant customers actively accelerate their spending from the hundreds of millions to the hundreds of billions. How does $7T even get spent on Nvidia or adjacent products? I suppose Sam could buy Nvidia at a mere $1.8T or TSMC at $540B, where all of these chips are manufactured.

My problem with this math is that it’s priced way too far in the future. It hasn’t happened yet, not even remotely. Nvidia’s extraordinarily high margins are up for grabs within the industry. At some point, their products can be replaced by ASICs, application-specific integrated circuits. This happened when Bitcoin and Ethereum miners switched from expensive Nvidia GPU’s to less expensive single-purpose chips created solely for Bitcoin mining. Years ago, Apple, Google and others also shifted their design in-house which was a big blow to Intel. In other words, it’s happened before and can happen again.

To sum up, massive value creation is happening in generative AI. I don’t think that additional new dollars deployed into Nvidia at this valuation is at all a reasonable way to capture that potential value. There are and will be many other beneficiaries. That doesn’t mean we’re in an AI bubble, but in some places, it does. No other Mag7 companies are near Nvidia’s valuation levels. The AI trend will continue regardless.
(Again, this is not advice. Consult with an expert.)

Bad bank behaviors

Ares is one of the oldest and largest private credit companies in the US with roughly $380 billion in assets under management across their strategies. While researching their latest private credit offering, I came across a shockingly bad pricing sheet between Merrill’s wealth management group and Ares. This is consistent across the major wealth management groups at big banks like UBS, Goldman and JPMorgan. The latest Ares private credit fund is named ASIF.

If you invested in ASIF directly (Class I shares) you received a 12.15% return in 2023. If you invested through Merril, you received just 4.55% (Class D shares). If you invested in the highest fee category (Class S shares), you received just 0.64% after fees.

Banks that offer off-the-shelf private investments to their clients often receive extra compensation (aka kickbacks) for allocating into certain funds. That’s not necessarily a problem as long as it’s reasonable. I wasn’t aware that those kickbacks are as high as 7-12%, or 50% to 100% of the total returns. You can see the numbers on the Merril Connect website here. These banks often preach access to proprietary private market deals, but this type of fee gouging is egregious. We have access to many if not all of these same funds at the Class I level. Additionally, we only allocate to strategies because we like the returns and risk. We never receive kickbacks. Please reach out if you would like more information.