ThePeachPit Model Portfolio Vol. V
This episode lays out my broader AI thesis with firm examples of how to invest in the space. Trades and the model portfolio will be available to premium subscribers.
ThePeachPit model portfolio has returned 4% since tracking returns beginning on February 28, 2024. Some notable returns include a short position in US Steel, C3.ai, Unity Software, and my EV charging names EVGO & CHPT. The selected AI cohort, IOCs, and energy services names have performed exceptionally well within the portfolio. Cybersecurity has posed to be a major laggard and should experience an upswing if/when the Fed pivots. Be sure to upgrade to the premium subscription for the full list of equity positions within the portfolio.
This week on the Growth & Gains Podcast, Josh and I discuss some of the key themes in the AI space. For those that regularly follow my research, you may already be building a conclusion on the direction I’m looking at in terms of winners and losers, software vs. hardware, and consumer vs. corporate. Based on my research, it is clear that there is no cookie-cutter answer for which firms will succeed solely based on these factors.
I will attempt to better articulate my thesis below.
Corporate vs. Consumer
Corporate:
This one should be relatively obvious; however, there will be some minor overlap for devices. When considering the two sides, I believe that corporate will be the clear winner when it comes to investible AI-related companies. These are firms that design or manufacture the hardware & software that will run the end-customers’ large language models. Companies that I expect to succeed in the near-term are those that are designing components for the server infrastructure. This includes semiconductor designers like Nvidia, Marvell, Micron, and Broadcom. I anticipate these firms to experience significant tailwinds as demand for their infrastructure continues to ramp while the supply chain is relatively tight. Much of these constraints fall on the back of Nvidia as their GPUs are essentially the backbone for accelerated computing. Though I do anticipate Intel to catch up, I believe the firm may experience some headwinds coming from the device side, which I will dig into shortly.
Another firm that will benefit from the AI renaissance is Oracle. Oracle is to data centers as Nvidia is to GPUs. They cannot keep up with demand. Oracle is scaling to be the regional hyperscaler of choice in terms of cloud computing. As discussed in previous reports, data privacy laws like GDPR and localized versions of the guidelines can be quite strenuous in how and where data is stored. Oracle is taking the bull by the horns on this one and quickly, or as quickly as possible, building out large, modular data centers for customers to run both their enterprise-class software package or building an Excel spreadsheet as the firm is partnering with Microsoft. This is what I would define as the company’s second wind. Tying in the other listed firms, vast amounts of servers & storage capacity will be required in building out these data centers, leading me to believe that these server-oriented chip designers/manufacturers will realize strength in the coming years.
For enterprise hardware, I believe we have a good pairs trade to consider, one that was discussed on the latest episode of G&G. HPE & DELL. HPE being the loser & DELL being the winner. I anticipate HPE to be running into similar challenges as Intel as the firm’s device segments will have a stranglehold on growth. Why is that? A common theme I have regularly referenced involves corporate IT spend and inflationary pressures on the consumer. As companies tiptoe into enterprise AI, I believe it will be in phases, resulting in somewhat of a cascading effect that may accelerate in CY25/26.
Phase 1: Servers (2024-2027)
Phase 2: Networking equipment (2h25-2028)
Phase 3: Devices (2h25-2026)
Despite anticipated growth across corporate IT departments, investments are not all equal. CIOs surveyed by Gartner suggested that 2024 will be a very budget-conscious year in which new projects will be focused on cost-cutting and short in duration. My takeaway from this is that department heads will be investing in critical infrastructure, cybersecurity, and software. I believe much of the focus this year will be in securing the next-generation AI infrastructure and laying out the groundwork for rolling out GenAI capabilities.
The next phase will be networking equipment. Networking equipment is forecast to be relatively sluggish for CY24 as we’re caught up in between refresh cycles. Despite major headwinds for HPE in the networking arena, Arista Networks has experienced a significant upshift in growth as their cloud-based networking software finds itself in high demand. From what I gather, companies will be slowly moving towards implementing 400g routers in the coming years, likely in 2h25-2026. I anticipate this to pick up as more enterprises explore implementing private GenAI applications across their network, which will require greater networking speeds. My thought process behind this will be once the AI servers are in place, department heads will quickly realize networking speeds will be necessary to reduce latency as these massive LLMs are built upon a lot of data and will require lower latency to function properly.
The last phase will be on the device side. I cannot fathom a purpose for an enterprise investing in AI workstations without the foundational AI servers and networking equipment in place. Otherwise, the company will have just invested in a Ferrari with a Honda Civic engine. Another theory is that if the firm were to invest in devices too soon without the backing infrastructure, the device’s tech may become obsolete by the time the firm secures those much-needed GPUs to run these applications. As fast as semiconductors are evolving for AI/ML, it wouldn’t be too far out there to make this presumption, especially as Nvidia just released their latest and greatest chips, the Nvidia Blackwell Platform that is said to save up to 25x in operating costs and energy when compared to their predecessor.
Consumer
On the consumer device side, I believe that AI PCs and laptops will lag behind the market by a year or two before seeing significant upside potential. My thesis is based around consumers facing persistently high inflation that has resulted in tighter budgets just to cover the day-to-day expenses. If a consumer can push out their PC refresh cycle, they will. This also applies to the latest Samsung Galaxy S24, the latest AI-enabled smartphone to hit the market. I anticipate the same headwinds for handhelds as I do for PCs in which if a consumer can push out their next device, they will. This mobile device is pushing $1,119, which is actually in line with previous generations of the smartphone. Despite the flat pricing, I anticipate demand to be relatively limited to those technocrats that absolutely need the latest and greatest device, not necessarily the general public.
Software
My software thesis is relatively in line with my hardware thesis. Enterprise CIOs will invest on an as-needed basis and will be focused on cost reduction projects. I have been adamant about my investment in Palantir as the firm does just that. One of the factors that I absolutely adore about the firm is their sales and marketing tact. The firm created a masterclass that brings together small groups of department heads and addresses their challenges with real examples. Palantir made its claim to fame post-9/11 in counterterrorism efforts. The firm has the best in class AI software on the market that is quickly becoming a tool found in enterprise IT departments. The firm’s government/corporate split for FY23 was 55/45%, suggesting that more enterprises are grasping onto their ability to improve operations with the use of their AI/ML software platform. THIS is driving the hardware thesis stated above.
On the other side of the fence, companies like Adobe are struggling in making strides into the next generation of generative AI. I believe that the firm will be hit on multiple fronts, being late to the party, having clunky integration, and OpenAI Sora. Sora is an advanced AI video generator that has been proven to create a near-real life video, paces ahead of previous iterations of the software. Given how quickly the software was developed, it is only a matter of time before the next generation is released, making the need for Adobe’s video editor almost nul. I’m not saying that video editing software will be phased out. What I’m saying is that for a short marketing piece, the cost difference between hiring actors, props, directors, film crews, and video editors will be overwhelmingly expensive when compared to an AI-generated video of the same quality. This was a major piece to my core thesis on ADBE shares, which I have been actively managing short-term, short positions in the name.
Another headwind is that if companies are experiencing the level of inflationary pressures as we are led onto, Adobe may be in more trouble than initially thought. Marketing budgets are oftentimes the first component to kick the bucket during a downturn. What happens when we go through the next upcycle? Are companies going to hire Photoshop experts or GenAI experts?
Power Demand
Electricity consumption is another component worth considering when looking at the broader AI picture. This plays into a lot of the verbiage semiconductor designers are voicing when releasing their latest and greatest chips. What percentage of power consumption is reduced with the next iteration of chips? As Oracle scales their data centers, more localized power generation will be necessary to cater to these sites. Data center power demand is expected to balloon from 17GW in 2022 to 35GW by 2030 as a result. I can almost guarantee that these data centers will not be powered by wind or solar power. Maybe in part, but not relied upon. My vision involves modular nuclear power plants to be strategically placed to cater to the high demand for electricity. Though these facilities are not expected to be built out domestically until the end of the decade, at the earliest, I anticipate small modular reactors, or SMRs, to provide a significant portion of the base load capacity. This may be paired with natural gas-fired power plants; however, given the climate-conscious bills being set in place, SMRs may be the winner of the next generation of power capacity.
Before we hop into trades for the week, I want to update my thoughts on Nvidia. I have seen headlines, journalists, and advisors calling the peak in price performance for NVDA shares. This is a touchy subject as the firm’s valuation is quite bloated at 38x trailing sales and 66x EV/EBITDA. A prudent investor should always be cognizant of the fact that companies with higher valuations fall the hardest; however, given the level of demand outlined above, I find it hard to build a case against the company. My favorite portfolio manager, Cathy Wood, is constantly beating up the name, suggesting it is set up for a pullback. I’m not going argue for or against her because I cannot predict the direction of share prices; however, it should be noted exactly where Cathy Wood sold her lot.
So, next time you hear her talking down about the name, keep this chart in mind.
Trades For The Week
Keep reading with a 7-day free trial
Subscribe to ThePeachPit to keep reading this post and get 7 days of free access to the full post archives.