Summary I support the CEO's decision to avoid independent AI/HPC hosting, reducing risks tied to low margins and rapid tech obsolescence once hyperscalers fully deploy their infrastructure in 2026. The company’s energy strategy, converting flared gas to electricity, significantly reduces power costs to 1 cent per kWh, aiming for over 1 GW capacity of low-cost energy. Recent acquisitions and facility upgrades are expected to increase computing power by 70% by late 2025, with potential operational cost reductions using its 2PIC tanks. Falling Bitcoin prices, ongoing share issuance, and negative sentiment toward non-mining expansions are some big risks to consider. Following the 30% decline in the share price, MARA's low valuation ratios and strong growth prospects make this stock a speculative strong buy. MARA Holdings, Inc. ( MARA ) has seen its share price decline by over 30% following the release of Q3 earnings. Seeking Alpha With the company planning to release more details on potential partnerships with hyperscalers in the upcoming quarters, I considered reviewing in this article some of the arguments made by the CEO, Fred Thiel, against expanding into AI/HPC hosting independently , opting instead to pursue partnerships with big data center and cloud computing players. I will advance in this introduction that I fully agree with his arguments, including the challenge of competing in price with hyperscalers like AWS, Microsoft Azure, and Google Cloud once they have fully deployed their infrastructure over the next two years. I will also discuss the company's expansion plans, including its technology advancements and energy strategy. Overall, considering these factors and the company's attractive valuation ratios, I rate this stock as a strong buy. As a matter of fact, I have started a long position in this company, taking advantage of the pullback following Q3 earnings. As always, I will include a risk section where I will discuss some factors that could potentially wipe out my investment (i.e. highly speculative bet) in this company. CEO Skeptical About Entering AI/HPC Hosting Independently Fred Thiel emphasized during the last earnings call that the AI/HPC hosting market will soon face a "race to the bottom" in terms of costs and profits, as hyperscalers dominate. I think anybody else trying to go into this business today is looking at a race to the bottom from a cost perspective and a very challenging capital market, once people wake up to the fact of how difficult it really is to run an AI HPC hosting business when you're not a hyperscaler yourself. One of the key points he mentioned was the decline in GPU rentals. He used the H100 rental prices as an example. The rental prices for this high-end GPU model from NVIDIA (widely used for AI workloads) have plummeted by 75% in six months. He also mentioned another reason for his apparent hesitation about getting into AI/HPC hosting independently: the rapid obsolescence of GPU hosting for AI tasks. If you think that the technology obsolescence of Bitcoin mining is aggressive, welcome to the world of GPUs. It is even more aggressive. Anybody who comes from the Bitcoin mining side , who is excited about going into the hyperscaler or large-scale data center for AI HPC hosting business is in for what I believe will be an abrupt surprise come two or three years when the hyperscalers deployments of the data centers they have in production today start coming online Instead, I believe Mara will focus on partnerships with large-scale AI hyperscalers. As the old saying goes, if you can't beat them, join them. These partnerships with hyperscalers would allow MARA to leverage its technology, such as 2PIC immersion cooling, and its expertise in low-cost energy generation to support the needs of hyperscalers. Although no specific details have been announced about these partnerships, the company plans to announce more specific details in the upcoming quarters. In my opinion, an official announcement of a partnership with a major hyperscaler could be a big catalyst for the company, especially if the share price keeps its current declining path until this event happens. Energy Strategy And 2PIC Technological Advancements In October, MARA launched a 25 MW microdata center operation with NGON across wellheads in Texas and North Dakota. These data centers convert excess flared gas, (i.e. a waste by-product in oilfield operations) into electricity to power co-located data centers. The benefit of using excess flared gas to generate electricity is a significant reduction in cost per kWh, from the typical market price paid by miners using grid energy of 4c/kWh, down to 1c/kWh. Aside from cost benefits, this method is highly beneficial for the environment, with a 99% methane mitigation efficiency This initiative is part of Mara's broader strategy to identify and develop sites with high potential for on-site power generation. Additionally, in the last shareholder letter , the company mentioned its long-term goal of expanding its gas-to-power capacity to more than a gigawatt in the next few years. We believe MARA's ability to pair onsite demand to onsite power (fueled by low-cost natural gas) will unlock hundreds of MW-if not more than a GW-of low-cost and long-duration gas-to-power opportunities. Moving on with other initiatives, by the end of this year, Mara will have installed 40 2PIC tanks across its facilities, including the new one in Ohio. What I find encouraging about this technology is the high-power density of 1MW that can be achieved per each 2PIC tank. As a matter of fact, 2PIC tanks can achieve up to four times the power density compared to traditional cooling methods. As a side note, the 2PIC technology offers superior thermal management by submerging equipment in a dielectric fluid. This results in up to a 60% reduction in cooling energy costs. Additionally, in the past quarter, Mara has secured its first commercial orders for its 2PIC tanks from third-party miners and data center operators. In my view, this is a strong indication of the benefits of this proprietary technology. Expect a 70% Increase in Computation Capacity By the End of 2025 Since the beginning of 2024, Mara has secured about 1 GW of nameplate capacity through acquisitions and new developments. The company's operational footprint now spans over 15 data centers across four continents, with plans to expand further in the Middle East and Africa. As a matter of fact, the company's total owned and operated computing capacity is expected to increase by over 70% by the end of next year. This includes two recent acquisitions in Ohio, along with the development of a 150 MW greenfield operation that will provide 372 MW of capacity by the end of 2025. Salman Khan, the CFO of the company, mentioned during the last earnings conference a significant reduction in operating costs for one of the newly acquired centers. Owning the sites will provide us with greater operational control and could further reduce our operating costs at Hopedale data center, which is part of this acquisition, up to 50%. I believe they will mass deploy the 2PIC tanks at this facility to achieve this cost savings. Aside from these acquisitions, the company has upgraded and expanded the capacity of five of its existing facilities: Ellendale and Jamestown, Garden City, the Wolf Hollow facility in Texas, and the Kearney facility in Nebraska. Overall, the company expects to achieve a fleet efficiency of 19.5 J/TH by the end of 2024. To put things in perspective, this is a 14% improvement from last quarter's efficiency of 22.7 J/TH. Valuation Considering the 30% decline in the share price following the Q3 earnings release, most valuation ratios have cooled off to attractive levels when compared to some of its direct competitors. An example is the price to sales ratio, which trades now below RIOT's value (see table below). Seeking Alpha Another example is the EV/EBITDA ratio (both FWD and TTM), with Mara's value being the lowest among its competitors. The historical chart below shows that most of the company's valuation ratios are trading close to the 5-year minimum. Stock Rover Therefore, I believe there is solid evidence to suggest that Mara is undervalued at best, or fairly priced at worst. Risks I believe one of the biggest risks to my bull thesis, is the current negative sentiment surrounding the expansion into AI/HPC hosting. As I already mentioned, I don't believe Mara will go into AI/HPC hosting independently. It all points out that the company will partner with hyperscalers. However, considering that most of the shareholders support the company's Bitcoin mining operations, any news related to the expansion of their operations away from mining could drive a potential selloff. Other risks include the high volatility of Bitcoin prices. At the end of last quarter, the company had over 27,000 BTC on its balance sheet. Therefore, a pullback in Bitcoin's price could erode the value of the company's assets. Finally, I am not particularly encouraged by the recent dilution of shareholder value. In the past year, the company has raised a significant amount of capital ($1.5 billion TTM) by issuing shares. TradingView The recent $850 million offering of zero-coupon convertible senior notes (due 2030) allows noteholders to convert their holdings into MARA's common stock under certain conditions. I have included here more details about this offering, although I believe this offering increases the risk of further dilution in the long term. Conclusion I rate this stock as a strong buy for highly speculative investors, like myself. As we all know, Mara is the largest publicly traded miner in terms of market capitalization and hash rate. Therefore, I am highly encouraged by the company's decision to avoid jumping into the AI/HPC hosting business independently (emphasis on this last word), given the high obsolescence risks and low margins in this space once the hyperscalers fully deploy its infrastructure in the next 2 years. Instead, Mara plans to partner with these hyperscalers (think of AWS, Microsoft, or Google), leveraging its advanced 2PIC cooling tanks and cheap energy solutions to support their infrastructure expansion. Speaking about the company's energy strategy, its gas-to-power projects could cut power costs to just 1 cent per kWh. The company is scaling these projects, with a goal of exceeding 1 GW of low-cost energy capacity in the next years. From an operational perspective, the recent acquisitions, combined with upgrades to existing facilities, are expected to increase the company's computing power by 70% by the end of 2025. Even more encouraging is the potential for some of these facilities to cut in half their operating costs thanks to the company's 2PIC tanks. Risks to my bull thesis include the negative shareholder sentiment to expand into AI/HPC hosting through partnerships with hyperscalers. I think an official partnership announcement could upset investors focused solely on its mining operations. On top of that, I believe that the current dilution could continue into 2025, which could weigh on the share price. However, I believe the company's low valuation ratios compared to its peers, combined with its growth rate and technological advantages, make it a strong buy. Nonetheless, this is a speculative play, which could take several years to potentially yield positive results.
