I started covering Nvidia’s accounting shenanigans a long time ago, but being honest, I could not have expected things to reach the absurd level we are witnessing lately.
The first partner in crime of Nvidia to pull off its epic revenue round-tripping scheme that began in the aftermath of the Covid crisis was Microsoft, which found in OpenAI the perfect vehicle to inflate its languishing Azure business revenues: “WHAT IF NVIDIA SIMPLY TAGGED ALONG WITH A MICROSOFT AZURE SCAM PLAYBOOK?“. The key that allowed Nvidia to support the quick build-up of Azure infrastructure, so Microsoft could justify all the “computing credits” given to OpenAI for a chunk of its equity, was a little-known networking company: Mellanox. Hilariously, prior to Nvidia’s very aggressive bid to buy Mellanox away from IBM, Mellanox was already a company with an established record of cooking its books (“MELLANOX, THE CORNERSTONE OF NVIDIA-MICROSOFT REVENUES ROUND-TRIPPING SCHEME“). Mellanox-Nvidia was such a match made in heaven, wasn’t it?
Why was Mellanox such a key move? Keeping things simple, it allowed Nvidia to start selling clusters of GPUs rather than single ones. This was a very fundamental element of the whole AI narrative that, let’s not forget, Nvidia tried to pursue in the past before jumping on the crypto mining bandwagon, but it failed (“NVIDIA, ITS RECYCLED “AI REVOLUTION” AND THE DARK SIDE OF IT KEPT AWAY FROM THE PUBLIC“). In 2019, Nvidia had a very big problem dealing with a lot of GPU inventory that crypto miners no longer wanted, while the secondary market was flooded with GPUs from crypto mining companies in liquidation. All of a sudden, Microsoft became the taker of all that oversupply, and the great revenue round-tripping scheme of the century began. Because Microsoft’s demand was starting to plateau in 2022 and OpenAI had not launched ChatGPT yet, the Nvidia stock price started crashing all the way down to $11 (split adjusted) in October 2022. Then, on November 30th, 2022, ChatGPT launched, and Nvidia did not waste any time throwing gasoline on the AI FOMO fire.
All other Big Tech companies, from Google to Amazon to Alibaba, started to panic and place a massive number of orders for the now-prized Nvidia GPUs, believed to be a must-have to develop AI fast and at scale. However, Nvidia knew very well that its pool of customers was limited, so the first strategy was to start releasing new GPU models very quickly rather than after multiple years, and feed the FOMO to have its latest and most powerful version. This strategy worked very well until late 2024, when the Blackwell platform started to encounter significant development issues to the point that Nvidia started to deal with order cancellations or orders shifted to its future Rubin platform. The final blow to the Nvidia narrative arrived in January this year when DeepSeek showed the world that cutting-edge AI could be developed with a small number of GPUs and for a fraction of the cost (“THE REAL ERA OF AI BEGINS, THE ONE OF THE AI CHARLATANS ENDS“). Understandably, Nvidia’s share price started to tank again, hitting a ~$90 bottom in April.
Nevertheless, Nvidia kept claiming it had “overwhelming demand” for its GPUs, and investors did not argue that, considering Big Tech companies continued to pledge absurdly high CAPEX spending to pursue AGI. However, it isn’t enough for Nvidia to sell a ton of GPUs; the key to continuing to push its stock price higher and higher is to sell more and more of them. How could the company achieve this when its big customers started to cut their orders, favoring their own internally developed GPUs? With even Microsoft stopping the buildup of its own infrastructure to support the nonsensical computing requests from OpenAI, Nvidia, now on its own, had to quickly find a solution, and this came in the form of datacenter companies like Lambda, Nebius, and, of course, CoreWeave.
I already exposed how Nvidia explicitly revealed that most of its GPU demand came from OpenAI in “CHEF’S TABLE: NVIDIA“, and now that OpenAI is running around signing bloated agreements it cannot afford, like in the recent one with Oracle (“OPENAI: LITTLE CASH, BUT PLENTY OF PROMISES, IN AN EFFORT TO AVOID BANKRUPTCY“), it is clear that most of the “overwhelming demand” Nvidia continues to claim exists for its GPUs is unrealistic. But hey! Till the music plays, you must dance, so what can you do to convince people to buy a product that the market is already saturated with? You tell these clients that you will rent back from them all the computing capacity they won’t be able to sell. In other words, Nvidia is now PAYING more and more to sell its own GPUs, and as a matter of fact, it did not even start doing this recently. As you can observe in this chart, Nvidia started to commit to purchasing computing capacity years ago. Clearly, in 2024, it accelerated doing so to make up for the falling demand for additional GPUs from its core customers like Microsoft.
The smoking gun proving all I am saying here can be found in the latest 8-K filing from CoreWeave that just revealed how this agreement with Nvidia has existed since 2023.
Hold on a second, shouldn’t CoreWeave have disclosed the existence of this agreement in the S-1 filing ahead of its IPO? Yes, they should have, but who cares about enforcing securities fraud legislation these days, right? It is also now understandable why Nvidia had a lot at stake with the CoreWeave IPO and could not let it fail, which is why it backstopped it with an anchor order at $40 per share and even continued buying shares in the open market afterward to inflate CoreWeave’s price to absurd levels.
So what’s going on here? It is fair to assume that OpenAI and Nvidia are cooperating to maintain the impression that demand is still huge for Nvidia GPUs. However, how can companies afford to pay for the infrastructure to serve OpenAI when this company has no money to pay any cash advance or provide guarantees? They take the agreement with OpenAI and go to lenders to borrow money to finance their infrastructure build-up (“CoreWeave closes US$2.6 billion loan for OpenAI commitment“) or they raise capital from the open market, like Nebius did on the back of a contract Microsoft obviously signed on behalf of OpenAI’s demand (“Why Nebius Is Raising $3.7 Billion Just Weeks After Microsoft Deal“). Where will all this money land? Of course, most of it will be used to purchase Nvidia GPUs. And what happens if OpenAI ultimately does not buy the computing capacity it committed to? Nvidia will buy the excess capacity from these data center companies, ultimately reassuring lenders that there will be a stream of revenue from those GPUs they agreed to lend money against.
Additional proof that there is already an oversupply of GPUs in the market also comes from the fact that Nvidia a few months ago agreed to buy a GPU reselling company: “Nvidia Nears Deal to Buy GPU Reseller for Several Hundred Million Dollars“. Please tell me, how is it possible for a company to operate a GPU reselling business when, in theory, there is so much demand for GPUs as Nvidia claims, to the point that it cannot promptly fulfill the requests of its clients? Of course, there is a stark contradiction here with Nvidia’s claims.
To conclude, after years of digging and putting together pieces of a very complicated puzzle, it is now very clear the extent to which Nvidia lied to the market and has been inflating its revenues, thereby extracting capital from companies and private investors in what is, in fact, a Ponzi scheme. But like all pyramid schemes, sooner or later the fraudster runs out of gullible investors they can get money from, and as soon as their own cash is depleted, the whole house of cards crumbles. This is exactly what’s going to happen to Nvidia; it is only a matter of when, not if.
JustDario on X | JustDario on Instagram | JustDario on YouTube
SYNNAX SUMMER PROMO
Get exclusive intelligence on stocks you can’t find anywhere else – insights that could give you a serious edge. Claim your free Synnax trial month now: https://synnax.app/PROMO/JUSTDARIO
If you love what we built at Synnax (and I think you will), this advanced market intelligence will cost you less than a cup of coffee a month for the next 12 months. After the 30th of September, the annual subscription price will be 149$ – redeem your free access now and give it a try.