First off

☁️ $BITF and Cloud Monetization at Lake Moses - The Trojan Horse into Large Scale GPUaaS???. ☁️🐴 On November 13th $BITF held their earnings call, and it was their paradigm shift moment, going from BTC miner to AI DC infrastructure player. IMO, $BITF had a fantastic call, and offered clear and obtainable guidance. They have even mapped out their site development with the NVDA hardware cycle in anticipation of being Vera Rubin ready for 2027. As of now, I don't know of any other publicly listed DC infrastructure provider that has guided as such. It just came at a bad time when there are fears in the market about the CDS of some AI Cloud/Infrastructure players to get the positive attention it deserved. investor.bitfarms.com/static… One of the biggest pieces of guidance was provided in a separate PR, BITF has confirmed the offering of GPUaaS at their Lake Moses site in Washington. This is something that I have pressed CEO @captainkeel about in both of our prior spaces meetings. teapaw.org/TheKamaHsutra/status/1… For my AI investment thesis, I would much prefer the ability to monetize up the AI vertical as much as possible. If you can do IaaS/GPUaaS, then you should do it. Colocation isn't really that much safer IMO, you are placing the same bet on AI succeeding. If AI succeeds, you will monetize your MWs much at a higher $/MW value. If AI fails then everything will fail, and colocation isn't going to be much safer. However, the biggest caveat is that you do have to own that entire vertical you are trying to monetize. If colocation fees for a cloud providers is going to eat up 20% of a deal then that is a lot of margin compression, something that asset lite providers will eventually have a reckoning with. $BITF is planning to offer liquid-cooled GB300s in their GPUaaS. Luckily, I have already started my own framework to look at such potential deals due to my work based on $IREN's GPUaaS. One of them also happens to be based on liquid-cooled GB300s, which makes it a very quick plug and play for analysis. My framework is very straightforward, it's simply the sum of all revenues minus the sum of all costs. The framework also breaks down the inputs and outputs on a per MW basis. On the revenue side there is only one input, and that is the dollar value per MW in a given deal. For GPUaaS deals, we can expect the deal to be between $10mil to $12mil per MW per year. To fit this into my framework, we have to break this number down to $/GPU hour. The biggest unknown for this calculation is the number of GPUs/MW, but luckily IREN has already provided this number for us with their MSFT deal for GB300s. They are providing about 76k GPUs for 200 MW of critical IT load. The number of GPUs for critical IT load is independent of PUE, and for IREN the number is 380 GPU/MW. Revenues: Based on 100% utilization, the GPU hour price would be calculated as: Revenue/MW/year / GPU hour/year $10mil/MW/year / (365 days x 24 hours x 380 GPUs) = $3.004 / GPU hour. $11mil/MW/year / (365 days x 24 hours x 380 GPUs) = $3.305 / GPU hour. $12mil/MW/year / (365 days x 24 hours x 380 GPUs) = $3.605 / GPU hour. Costs: Cost per GPU IREN has also provide these numbers to us. They appeared to have gotten a volume discount for the 200 MW of GB300, which works out to be just over $76.3k per GPU. However, I we'll use the higher number they provided for their smaller order sizes at $80k per GPU (this includes all ancillary equipment). I have also assigned a high single digit 9% interest on the cost of the GPUs through a leasing model amortized over 5 years. Ben Gagnon has stated that they would lease the GPUs as well. teapaw.org/i/grok?conversation=19… Electrical costs has been calculated for 1.2 PUE per MW based on $0.03 kWh pricing. Note that 18 MW gross load at 1.2 PUE = 15 MW critical IT load (18/1.2). Data Center Costs "Bitfarms has signed a fully funded, binding agreement of US $128 million with a large publicly traded American multinational provider of critical infrastructure and services for data centers. Under the terms of the agreement, the partner will supply all critical IT equipment and building materials for 18 MW of gross capacity with an anticipated industry leading PUE, between 1.2 and 1.3. The site is targeted for completion in December 2026." investor.bitfarms.com/news-r… $128mil/18MW = $7.1mil MW for all materials only. There should be around another $3mil/MW for the rest of the buildout cost to bring it more in line with industry expected cost of $10 to 12mil/MW. Can using these modular type builds bring BITF closer to the $11mil/MW number? If so, this would help tremendously. Since this has been fully funded, there will be no cost of capital further assigned to this section. From my numbers, if $BITF is financing the entire amount of the GPUs, and whether the DC build cost is at $10mil/MW or $11mil/MW, they would need a deal that is around $10.5mil/MW to breakeven for all GPU and DC cost in 5 years (slightly lower if at $10mil/MW build cost). If BITF is able to get a deal in the $11mil/MW range, then that would be ideal. Aside from DC build cost, the other big factor that BITF will have control over is the amount of interest paid for the GPU leases. If going by the 380 GPU/MW number, 15 MW critical IT load, and $80k GPU + ancillary total cost, then the total cost of GPUs before interest is $456mil. BITF is sitting on a good amount of cash, $588mil from the converts, before accounting for the credit facility from Macquarie. If they use some of that cash and in conjunction with some possible prepayments to pay for half of the GPUs up front, thus cutting the interest payments, it would greatly reduce their breakeven number. Prepayments has been a fantastic move on the part of $IREN, and it will get copied in other deals going forward. At $11mil/MW, it makes everything much easier. But will $BITF be able to get an $11mil/MW deal given its their first deal in AI, and GPUaaS no less? If you listen to Ben Gagnon and factor in the time to power in the Pacific Northwest, someone might just be willing to do it. In addition, we have seen enterprise customers willing to pay more than hyperscalers do, but they do not offer the same cachet for the same level of creditworthiness. For 18 MWs, this deal would be tailor made for an enterprise customer. I have always been a proponent of moving up the stack if you are capable, it simply provide better returns if the AI demand is real. Quick napkin math for the payback of just the DCs in a colocation type deal at 80% net margins and assuming deal for $2mil/MW: For $12mil/MW build cost = $12mil/MW / ($2mil/MW/year x 80%) = 7.5 years. For $11mil/MW build cost = $11mil/MW / ($2mil/MW/year x 80%) = 6.875 years. For a 10 year colocation deal, you're spending at least 70% of your time paying back the cost of the DC, Where GPUaaS can pay for the DC and the GPUs in 5 or less. The math is so much more compelling for GPUaaS, with the caveats that have to be able to do it and the demand is sustainable long term. However, if colocation deals are being handed out for 10 to 15 years, then there shouldn't be any reasons there isn't demand for GPUaaS after the initial 5 years. As seen from the deals given to $IREN $NBIS and Nscale, GPUaaS deals are in 5 year terms. This is likely due to the obsolescence of GPUs being incredibly hard to account for beyond 5 years. 5 year segments are likely how deals will be structured. Profit margins are dopamine for companies, they can't get enough of it, and once you show them there is a better way to get the fix, they cannot go back. I know Ben Gagnon has been adamant that they just want to focus on being the electrical infrastructure provider, and not getting into another "hamster wheel" like situation that BTC ASICs are. However, we now know that the useful life of GPUs are longer than that of BTC Asics. On top of that, they pay a lot more during their useful life. I have a suspicion that Moses Lake is a live pilot for BITF's ability to provide GPUaaS. If successful they will offer it to willing takers, partially or fully, at their other larger sites in Pennsylvania. Their sites in Quebec and Sharon are already primed for these type of deals. For a 15 MW IT load, 5 year, $11mil/MW, that is a $825mil deal. Pretty astonishing given the size of BITF's market cap. Once your bottom line is in the black you cannot go back.