Your ULTIMATE guide to AI’s least understood business: the “Neoclouds”. Here are additional points: 1. The advance rate, meaning how much neoclouds can borrow relative to the amount of money they need to spend on Capex (to buy chips and equipment) can get QUITE high, in some cases 90-95%+ depending on the credit of the “off-taker”, so from Meta for example. 2. Asset backed financing and project financing (one of the groups I worked in at Morgan Stanley) isn’t new. What is unique about these GPU backed financings are the much shorter contract length; in the power world you often see power purchase agreements of 15-20 years and assets that have a much longer useful life (think a gas turbine). In a power deal, the debt is shorter than the PPA, and the PPA is shorter than the asset’s life, which leaves a long tail. Here the loan, the contract, and the asset useful life are much closer in length, so there is a shorter tail behind the debt. While Coreweave does have other types of debt, this structure accounts for over half and funded much of its massive growth over the past few years. They also have a decent amount of convertible debt, which is utilized by higher risk companies; I also worked in convertible debt origination at Morgan Stanley as well so let us know if you want a video on how that works. :) 3. And finally what suprised me was the fact that on the other side, so if you look at an Anthropic’s balance sheet, “take or pay” agreements to buy the compute do not show up as a liability on the balance sheet, rather just shown as a footnote!! While this is consistent with accounting rules, it’s surprising. With that said, there is a huge need for compute and these companies are major players in providing the needed capacity. Many are starting to own more of the infrastructure (Coreweave even tried to buy Core Scientific). #ai #Tech #aiinfrastructure #TikTokTechContest

When I heard Anthropic has agreements to spend $518 billion on compute, I had to understand what companies benefit from that half a trillion dollars of spend. And wine video is back stopping, much of the financing for many of those companies. As when we was researching, and now I'm going to share everything I learned with you. Let's begin with where we are within the AI ecosystem, we're in layer 3, the infrastructure, which at its core can be thought of as the company's who sell compute. Now, computers to raw processing power needed to train AI models and Wall Street split those companies into two categories. First are the hyperscalers, household names like Amazon's AWS or Microsoft's Azure, and second is the Neoclouds, where the poster children are core weave and nebius. To understand the Neocloud business, let's run through it in video CEO, Jens and Wang calls an AI factory. First you need energy, second you need chips, and third you need the building or power shell. Many of these Neoclouds started as crypto miners, but pivoted their business when they realized AI was a better opportunity. The output of the AI factory is compute, but companies make money in different ways. Applied digital and core scientific or landlords, they own the power shells and make money renting the real estate without taking on the risk of buying the chips. Of the companies that sell compute, vertically integrated players like I run, own their own powered shells. Crusoe got there from the other direction, it started with power generation and then built from there. But on the other hand, core weave, nebius and lambda historically didn't own the power or the buildings, and unlike the hyperscalers, they don't have any wildly profitable business throwing off cash flows either, which brings us to a key difference between hyperscalers and Neoclouds. Hyperscalers have an actual business beyond compute. Microsoft's Office, Amazon is the everything store and Google and Metasel ads. These earnings then translate into real cash flows that at least at first was able to fund AI catbacks. Recently though, catbacks has gotten so high that the hyperscalers have started raising debt. But when banks land at the corporate level, they look at leverage, or debt to EBITDA, and against their huge earnings, the borrowing is still modest. Neoclouds by contrast need money to buy the chips, but they have two problems. First, they don't have the cash flows from another business to pay for them outright. And second, their existing businesses often can't support the amount of debt they need to fund the chips. What they do have, however, are valuable contracts with big players like Meta, and their chip back financing, borrow against the contract and the chips themselves. Put the contract and chips into an SPV and raise that off the cash flows. For example, say where we've signed a five-year contract with Meta for 100 million a year. Lendership out the expenses like rent and energy, and size the debt. What's amazing for core weave is they can borrow at much lower rates because Lenders look at Meta's credit not theirs. Core weave gets the cash, buys the chips, and voila, they sell compute. While this might seem great, the depreciation and interest eat away at much of the profit. But that's just one contract and one SPV. In reality, there are many. Lenders like OpenAI and Anthropic need their businesses to keep growing to make payments. And the contracts are short. Unlike a 15-year agreement in the power industry, some run just three years. Not to mention, the assets backing the debt, the chips, have relatively short useful lives. That's where Nvidia comes in. Core weave is a major customer for Nvidia's AI chips, so Nvidia has a huge incentive for it to succeed. Not to mention it's 11% ownership stake. On top of that, Nvidia has agreed to buy any compute capacity Core weave cancel, which helps lower borrowing costs. It's one of the many reasons the economist called Nvidia the central bank of AI. Stay tuned for part two, where we'll do a full deep dive.