Investors are likely aware of the meteoric rise in the stock price of Nvidia: a whopping 2,500% over the last 5 years, achieving a $2.9 trillon (!) valuation as we go to press. Its sheer size guarantees that nearly any active equity investor (indeed, any investor) underperformed the major US indices in the last few years.

Since its humble roots supplying the video game industry, Nvidia has become a juggernaut. Soaring revenue, profits, and investor enthusiasm reflect the role of their graphical processing units (GPUs) as the go-to processor for AI and other cloud computing applications.

Investors licking their wounds for not owning (or owning more of) the stock have an opportunity to step back and take stock before jumping on the Nvidia bandwagon.

Consider:

• Nvidia’s price-to-earnings ratio is about 55, nearly twice that of the broad market.
• Outsized profits invite competition, and both competitor chip makers and Nvidia’s major customers are scrambling to get a piece of the action, as well as manage their own costs.
• A keen observer and friend of Magnolia has dubbed Taiwan (where the chips are manufactured) the new Arrakis. A reference to the coveted spice in Frank Herbert’s Dune, the analogy suggests eventual military conflict with China and likely disruption of supply to the West.

Potential investment opportunities around AI include:

• Nvidia itself: aside from the rich valuation, competition, and geopolitical risk, Nvidia's large customers appear to be stockpiling processors ahead of their immediate needs, pulling forward demand and causing overestimates of future revenue.
• Other chipmakers: starting at a disadvantage, their new chips will face stiff competition (e.g., from Nvidia) and margin pressure.‍
• Nvidia’s large customers: the Googles and Microsofts have the resources to compete in the arena of applications of AI, but to the extent their vast computing power and prompt training is spent making (free) images like the ones you see here, its enormous cost will lead to investor disappointment. Another question is whether any success they have can move the needle (given the size of their existing businesses), or whether it may simply replace existing revenue streams.‍
• Nvidia’s smaller customers: for venture capitalists, there is and will be a bevy of startups: many of them moonshots but few of which will actually land on the moon. Anyone considering a sideline in venture investing is advised to take a deep breath and read Sebastian Mallaby’s excellent history of the industry first.‍
• Electric utilities: they have indeed been able to raise prices in view of growing demand, but regulation will ultimately limit further increases.‍
• Power generation: this is hard sector to invest in, and carries idiosyncratic risks associated with the diverse sources of power and challenges with building new capacity.

This brings us to Magnolia’s perspective on where to find the greatest visibility into future returns and the most forbidding moat. AI and many other applications require massive data centers to house the servers that run these expensive chips and to meet the needs of whichever customers require their processing power. For the builders and operators of the leading edge data centers, here’s the moat:

• The hyperscale data centers required to run advanced applications, including but not limited to AI training, now cost as much as $5 billion to build.
• These data centers require a hefty supply of reliable electricity, which is already limiting new construction due to the strain on utilities.
• The race to add capacity is putting strain on upstream elements of the supply chain, like electrical equipment and construction expertise.
• Computing power isn’t the only source of growing demand for megawatts: the electrification of the automotive fleet, home heating, and other applications will compete for reliable power from the grid.
• Two of society's greatest desires are now in direct conflict: to harness the power of AI for human progress, and to reduce and eliminate our dependence on the major sources of global energy at the same time. Mandated limits on future AI capacity are a distinct possibility.

‍

For these reasons, the companies that are able to develop and operate the coveted hyperscalers (now) stand to command premium property values and rents. There will be winners and losers: the required access to capital and operational capability, like the future, will be unevenly distributed. We would be delighted to discuss this unique opportunity with you.

Further reading:

https://www.principalam.com/us/insights/real-estate/data-center-investment-opportunity
https://events.greenstreet.com/green-street/Industry-Leaders-Data-Centers-A-Global-Perspective
https://www.cbreim.com/insights/articles/the-infrastructure-of-ai
https://www.mckinsey.com/industries/private-capital/our-insights/how-data-centers-and-the-energy-sector-can-sate-ais-hunger-for-power#/
https://www.brookfieldoaktree.com/insight/infrastructure-ai-driving-demand-electricity-production

Images: DALL-E 3