Description

I’m long APLD.

Applied Digital is a developer of next-generation infrastructure for AI. They have two segments: (i) custom-built, high-performance data centers; and (ii) GPUs as a service. Applied is currently focused on a large 400MW data center project in Ellendale, ND, and is marketing a 1.4GW pipeline of additional capacity. Please see the company’s latest presentation for more details. 

There is a frenzy for compute power by both hyperscalers and Fortune 500 companies, covered extensively everywhere you look. Power and datacenter capacity is the new bottleneck. Nvidia’s CEO, Jensen, just discussed the need to modernize a trillion dollars worth of data centers to tap into AI’s potential. APLD is a first mover aiming to capture this opportunity. 

But let’s get straight to it. The market thinks Applied Digital is a promote. A crypto-to-AI grift. The company has been attacked by reputable short sellers (Wolfpack, Friendly Bear, BearCave) and currently sports a 20+% short interest. Woof. 

The short thesis generally centers around attacking the company’s CEO, origin story, and lack of industry experience. The rest of the points are irrelevant to the story or have already been refuted by company actions. So the remaining short thesis: CEO Wes Cummins is orchestrating a promote and can’t be trusted… Applied is just chasing the market stories to keep the game going…Applied will never sign a real lease with a hyperscaler customer…There is no way Applied can compete with legacy data center players.

I strongly disagree with those assertions. This is why the opportunity exists.

CEO Wes Cummins is a hedge fund guy and a tenured short seller. I’ve known Wes to be smart, hard working, ambitious and entrepreneurial. He started the company in 2021 as a venture to marry distressed Chinese crypto mining assets with low cost stranded power in the US. Like all startups, there are countless zigs and zags. Like most entrepreneurs, Wes is an optimist. Wes saw the wave of compute demand rising, along with a lack of adequate industry capacity, and went all in to be a leader in this new next-gen datacenter industry. APLD is not chasing crypto conversions, they aim to be a leader in purpose-built greenfield next-gen facilities.

“Luck is where preparation meets opportunity.”

APLD’s experience in developing power hungry crypto hosting sites in low-cost power locations was the perfect preparation for today’s AI opportunity. APLD’s foray into North Dakota for cheap electricity to fund crypto mining left the company with relationships and access to massive low cost power, low cost land with space for expansion, room for additional wind/solar farms to be integrated, state tax incentives, and cooler temperatures which are beneficial for the TCO of high density AI data centers. A healthy dose of luck contributed to locking up this opportunity, but industry participants have told me you cannot underscore just how valuable this power and first mover advantage is for prospective customers. If GPUs are the pickaxes behind the AI revolution, power guys are swinging the axes. 

I strongly encourage anyone interested to listen to Wes and APLD’s origin story in a recent interview. Investors can be critical of Wes’s optimistic timelines and guidance, but this is a startup that has gone from chasing rabbits to hunting elephants, all while under public company scrutiny. Wes has never sold a share of APLD and his largest investors, including his previous employer, have only added to their position over time. That is the exact opposite of a pump and dump. Wes has also never fought back against the short sellers and one of his last twitter posts was defending that notion. 

For the record, I’m a huge fan of one of the OG shortsellers, FriendlyBearSA. Just last week he posted “Apparently there were more funds long $apld than short at the recent short seller convention. Sign of the times for where we are in the fraud cycle.” Maybe these short sellers got lazy and want to play hot potato with a promote. Or, maybe, they have talked to the utilities, sat in town hall meetings in North Dakota, talked to potential hyperscaler customers and talked with private next-gen data center players that have validated what Applied Digital is building. 

Applied Digital has put hundreds of millions into the ground already on spec and is full speed ahead. Wes has hired smart people and is working to use his relationships, knowledge, and risk taking acumen to win in the next-gen buildout race. See the latest update video here. Ellendale is a very real project.

Why Now

APLD has been tainted by shortsellers for crypto, B Riley, and other tangential allegations. Applied Digital has missed timelines and was talking about an imminent hyperscaler customer agreement back in April. It is now October. Skeptical investors have concluded that Ellendale is not real and there is no credible hyperscaler customer. APLD even resorted to using Yorkville to raise capital, a funder of last resort and a red flag. 

This narrative did a 180 on September 5th when the company raised $160mm from Nvidia, Related Companies, and other institutional investors. This was a huge vote of confidence from investors and market participants that went across the wall and performed deeper diligence into the potential hyperscaler customer, power contracts, 1.4 GW backlog, etc. 

Based on our conversations with industry participants - power providers, competitors, bankers, investors, hyperscalers - and a review of publicly filed utility information, we believe APLD is the real deal and has a chance to be a leader in the next gen data center buildout. We believe the delays are a function of the complexities of putting together the lego blocks of a very large project with power, financing, customers, and investors. We believe Ellendale is one of the most high-demand projects on the market and that a hyperscaler lease announcement could come any day. APLD’s CRO recently slipped up at an industry conference when being interviewed by enthusiasts and said an announcement was coming in a few weeks. The original video was pulled and the twitter account was blocked for violating terms. 

Applied Digital’s hyperscaler delay could also be a function of grander ambitions by Applied, hyperscaler customer, power providers, and the state of ND. There is industry chatter that much larger projects are being explored in North Dakota for the reasons previously mentioned. 

Unit Economics and Valuation

HPC Data Centers

The rule of thumb for data centers is it costs $8-12 million per MW to build and each MW can generate ~$1mm of NOI. APLD’s 400MW Ellendale project will be a $4 billion capital investment secured by a long term lease with a AAA hyperscaler tenant. The project will be funded with $3+ billion of low cost, project level debt and generate $400 million of NOI - using a 5-6% cap rate (and giving no credit for the 1.4 GW backlog) equates to a $6.7-$8 billion TEV. 

Locking in a long-term lease could be worth $3.7b-$5b of equity, or $17-$23/share, undiscounted. Each 100MW of backlog is theoretically worth $5/share using the same math. If Applied is able to lock in 1.4GW of projects, that’s an additional $70/share of value that could be created. Haircut all of this however you prefer. The numbers can get wild. 

Applied Digital Cloud

This segment has not been a focus for the company or the investment community recently, but is an underappreciated lotto ticket. Think of it as an early-stage CoreWeave. The business model is simple: they purchase GPUs from NVDA, install them in data centers, and lease them to customers for a fixed contract term that is typically 2-3 years. The GPUs typically pay for themselves during the initial lease term, leaving Applied Digital Cloud owning them free and clear at the end of the lease term for several more years of use. Applied takes very little risk here because at the time they purchase the GPUs there is already a contract in place for a full pay back of the equipment.

The NVDA investment is especially important for this segment because this is where NVDA can drive Applied’s future success. The hyperscalers are working on their own chips and are frienemies, so it is in NVDA’s long-term interest to ensure that there is a healthy ecosystem to support the consumption of their GPUs. They want companies like Applied and CoreWeave to prosper. NVDA can assure Applied’s success in a variety of ways by providing: (i) early allocations of the latest products; (ii) the best customer prospects who want to consume NVDA’s products as a service; and (iii) capital.

Consider what NVDA’s support did for CoreWeave. NVDA invested $100MM in CoreWeave in 2023 at a $2BN valuation (press release). As of September 2024 CoreWeave was raising money at a $23BN valuation (Reuters article).

Valuing this segment is tricky because it is just getting started. If we assume it’s worth just 10% of CoreWeave, that’s $2BN or $8/sh. And it goes up from there. In the fullness of time, if Applied Digital Cloud prospers, then the company will spin this out as a separate company.

The High Level Setup

If/when Applied Digital locks in a hyperscaler lease for Ellendale at industry standard economics, APLD will go to hero from zero. Wes will be viewed as an entrepreneur who executes, not a promoter. The backlog will go from being a pipedream to a real opportunity that should be valued. 

APLD is currently uninvestable to most institutional investors. Further industry validation will drive a perception U-turn. Applied Digital will be the only pure play next-gen data center player in the public markets. There is no short thesis when NVDA and Related are anchor investors and a hyperscaler is a long-term partner and customer. APLD will fundamentally be worth a lot more than current prices, but the stock could disconnect from any fundamental grounding as a result of being a public pureplay on the next AI bottleneck. The right tail asymmetry of this investment is under appreciated.  

Risks:

Failure to sign a hyperscaler tenant - for project-specific, general industry, or macro reasons.

Economics of projects worse than expected.

Slower ramp up and project execution risk.

More dilution for future facilities or current burn.

Single asset valuation is lower.

I go 0-3 going long heavily shorted names on VIC. Of note, I have never traded a share of APLD until the 9/5 investment which validated the company, in my opinion. I have not talked to Wes about APLD or its prospects over the past couple years other than the occasional good luck text.

Catalyst

Earnings updates/Project Announcements/etc.