Recent years have been characterized by an unusual spike in the number of companies listed on the public stock market that ended up filing for Chapter 11 shortly after. Notable cases include WeWork, Lordstown Motors, ELMS, Bird Global, Proterra, and Nikola. While some companies were simply bad businesses, others like WeWork or Nikola were outright frauds. Did this trend ring any alarm bells at the SEC offices? No, it didn’t, which is why investment banks kept trying to profit from dumping toxic companies into the public market to pocket fat fees and benefit their prized big institutional clients. However, after being burned repeatedly, unlike regulators, investors started to back off from “fomoing” on these offerings, and as a result, the market flow has significantly collapsed in recent times. The implosion of CoreWeave’s attempt to IPO might mark the end of this trend and potentially trigger significant problems for private equity funds that bet billions of USD on questionable (to say the least) companies.
9 months ago in “IS MAGNETAR CAPITAL BLOWING AIR INTO THE NVIDIA (PONZI) SCHEME SO THEY CAN BET ON ITS IMPLOSION?” I wrote about how a little-known hedge fund to the general public that during the GFC structured billions of toxic CDOs (sold by banks to customers so Magnetar could bet on their implosion and pocket billions in profits) was potentially trying to replicate the same trade through CoreWeave. Now that the S-1 has been filed, we can tell for a fact that CoreWeave, like the CDOs structured decades ago, is an outright fraud.
Let’s start with the list of related party transactions with Magnetar Capital.
1 – AI Computing Service Reserved Capacity and Prepayment Agreement
In August 2024, CoreWeave and a Magnetar fund entered into a “MagAI capacity agreement” where the fund agreed to purchase computing capacity for its portfolio companies from CoreWeave, which is also a portfolio company. According to the agreement, Magnetar paid a 230 million USD refundable deposit to CoreWeave to purchase future capacity at predetermined rates. Considering that CoreWeave disclosed it receives ~20% deposits from customers to lock in future computing capacity, with this agreement, Magnetar effectively created 1.15 billion USD of potential future revenues for CoreWeave, since as of December 31, 2024, CoreWeave had provided no service.
As if what I just described wasn’t already troubling, what comes next is even juicier
2 – In June 2024, CoreWeave invested 50 million USD in the Magnetar AI Venture fund (MAIV), which two months later was exactly the same fund that purchased future computing capacity from CoreWeave itself!
3 – Magnetar helped arrange two loan facilities in 2023 and 2024 for CoreWeave for 2.3 billion and 7.6 billion USD respectively to finance the purchase of GPUs that would later be pledged as collateral for the very same loans. Magnetar acted as arranger of both facilities, meaning it pocketed millions in fees upfront while also underwriting part of the loans itself that are costing the company an average interest rate of 10%+. Hold on a second, how is it possible that a collateralized loan with such prized collateral like Nvidia GPUs has been granted at a spread equivalent to that paid by a Single B-rated company? Kind of weird, isn’t it? Clearly “the math isn’t mathing” here.
Now it’s time to put Nvidia and Microsoft in this picture, the two companies that have been perfecting the most sophisticated revenue round-tripping scheme of all time in recent years (HOW TO FABRICATE REVENUES FOR DUMMIES).
After I wrote thousands of words about it for more than a year, this past week The Information “shocked” the investors’ public with this article: “Project Osprey:’ How Nvidia Seeded CoreWeave’s Rise”.
Here is what Nvidia did:
- In 2023 Nvidia invested 100 million USD in CoreWeave
- In 2023 alone CoreWeave spent 380 million USD to purchase hardware from Nvidia, with the amount likely in the billions in 2024.
- In 2023 Nvidia committed to rent CoreWeave chips and servers for a total of 1.3 billion USD until August 2027
So basically, Nvidia gave money to CoreWeave so they could have the funds to place an order for Nvidia GPUs, while at the same time, Magnetar was arranging a very expensive loan facility to allow CoreWeave to settle the remaining payment with Nvidia, which committed to renting its own chips from CoreWeave so the company could have revenues to pay the interest on this loan facility. If this isn’t fraudulent revenue fabrication, I don’t know what else it could be, my dear reader.
What about Microsoft? The biggest customer of Nvidia for the number of GPUs purchased in 2023 and 2024 (by far) entered into an agreement with CoreWeave in 2023 to rent the company’s servers, ending up accounting for 35% and 62% of CoreWeave’s total revenues for 2023 and 2024 respectively. Hold on a second, isn’t Microsoft Azure a direct competitor of CoreWeave? Of course, it is, and here is how the company describes it in the S-1:
“Our primary competitors are larger, global enterprises that offer general purpose cloud computing as part of a broader, diversified product portfolio. Key companies in this category are Amazon (AWS), Google (Google Cloud Platform), IBM, Microsoft (Azure), and Oracle. While these businesses have greater resources than us across sales and marketing and research and development, and benefit from broad brand awareness, they are not purpose-built for the AI and accelerated compute use cases that we serve. As a result, some of these very competitors have become customers of, and partners to, CoreWeave in a number of cases, demonstrating our competitive differentiation and ability to deliver highly performant, purpose-built infrastructure that outperforms existing general purpose cloud solutions today”
Now please explain to me how Microsoft, which bought the exact same GPUs and server infrastructure as CoreWeave, does not have data centers “purpose-built for AI and accelerated compute use cases”. This is, of course, a lie. What’s the truth then? One potential explanation is that Microsoft is coordinating with Nvidia to fabricate high demand and scarcity for GPUs in the market to claim to their investors that there was huge potential for their own respective services and products for many years ahead. Unfortunately for them, the illusion is becoming more and more expensive to sustain considering that, incredibly, it has already surfaced that Microsoft has recently been cutting orders for future renting of CoreWeave computing power (Microsoft drops some CoreWeave services ahead of $35bn IPO). This news was, of course, denied by CoreWeave since this bombshell from the Financial Times significantly undermined the success of its IPO.
At this point, there should not be any more doubt about the fraudulent nature of CoreWeave, which represents one of many examples of startups used by hyperscalers to greatly inflate their revenues and valuations through the years. This scheme isn’t indeed an isolated case; another prominent one is OpenAI, as I documented in: THE SMOKING GUN THAT PROVES HOW OPENAI IS MICROSOFT’S REVENUES LAUNDROMAT.
Where is the regulator in all of this? Personally, I have no idea, but I can guarantee you it won’t end well, and perhaps after DeepSeek, CoreWeave’s Initial Fraud Offering implosion will be another step toward the overall collapse of the whole fraudulent AI Narrative sold to investors for many years.
JustDario on X | JustDario on Instagram | JustDario on YouTube