The Central Bank of Nvidia, Part #2

In a Philochat (now only for paid subscribers) from this week I wrote that Nvidia may be the Central Bank — using its cash and strong balance sheet to fund the AI cycle — but Open AI is now the subprime lender.

In Part #1 we dissected the first 3 layers that explain how Nvidia catalyses and backstops the boom in AI investment.

In Parts #2 and #3 we will show how 1) compute and GPU dealmaking has become a tool of financial engineering 2) present extra layers (very important) that go deeper into the process 3) integrate the deals that Open AI has recently unleashed into the equation and 4) touch on the Nvidia Put and its effects in today’s AI-drive cycle.

Finally, we will explain how these monstrous deals make sense from Nvidia’s point of view — but spell disaster for the counterparties.

*Note: The extra layers also connect with Nvidia’s relationship with CoreWeave and Nebius.

Layer 4: The CSPs

While Hyperscalers can be understood as “Big Cloud” — we can understand AI infrastructure companies like CoreWeave (CRWV) and Nebius (NBIS) as “Small Cloud” — or the alternative Cloud Service Providers.

Nvidia has been running a very interesting playbook with Small Cloud.

While CRWV and NBIS have different business models, with the former running a debt-heavy balance sheet and the latter relying on cash-financed growth — they have been both on the receiving end of Nvidia’s alchemy.

Trojan Horse Leverage

• NVDA’s playbook with the CSPs and what it tries to achieve 👇
• Gain leverage against Big Cloud by creating alternatives to them
• Create CSPs that are big enough to make a difference to NVDA’s business..
• This is in case Big Cloud decided to bring the claws out and tries to take down NVDA, probably already in pay —> Custom Silicon is already gaining traction, and NVDA hates anything that threatens its dominance.
• By seeding Small Cloud with equity/debt or hard-to-get AI infrastructure, or even all of the above — NVDA creates stock market value for a CSP.
• The CSP can use that “stock market value” to raise equity/debt from non-NVDA funders.
• That capital is then used to buy more infrastructure from NVDA.
• This helps keep the business flowing while sucking the air out of from competitors like AMD. We will get to this in more detail.
• Holding all the cards in the equity/debt stacks of these companies means it will be tougher for them to pivot away from NVDA — if that decision ever came up.
• And so, we can think of these funding deals as Trojan Horses where a CSP that deals with NVDA becomes “captive” to them.

—> But what happens if the ever-expanding AI capacity of these CSPs at some point finds no demand to match it?

—> Doesn’t this eventuality risk bringing the house of cards to an abrupt end?

Enter the off-take agreement…

Layer 5: Onassis and Charter-Backed Finance

Legendary ship owner Aristotle Onassis used one major innovation to propel himself to the shipping halls of fame. Instead of using cash capital as the equity part of the stack — he used charter agreements. That financial innovation is called charter-backed finance.

To put it simply — instead of having cash collateral used to raise debt from banks, he went and signed charters with companies that needed to charter his vessels.

He then took those charters to the banks and raised debt with those charters as collateral.

I was reminded of Ari as I read the AI-related news flow recently — it’s not exactly charter-backed finance but similar.

And I will tell you why…

Ari could not get the financing to buy those vessels without those charter agreements — he was enabled by his charterers.

In the case of CoreWeave today, the company isn’t even profitable.

So, how does Mama NVDA and Captive CoreWeave manage to remove market fears that CoreWeave’s ever-expanding compute capacity would be left without demand — and end up bankrupting them.

—> So they signed a deal that would guarantee NVDA off-taking all excess supply from CoreWeave. Simple right?!

Layer 5 ends up looking the same as Layer 4. Not only NVDA backstopped CRWV to get them started and growing — but also guaranteed that CRWV would never have excess unsold capacity (as long as the agreement remains, current one is up to 2032).

Natural market response…..

“We see this as a positive for CoreWeave given concerns from investors around the company’s ability to fill data center capacity beyond its two largest customers (Microsoft and OpenAI).” —Barclays Analyst

So, I guess nothing to worry about?!

This keeps the cycle moving 🔄 —> because if the share price crashed — they couldn’t continue funding themselves.

Remember, everything is in the valuations. This is pure reflexivity. You don’t understand what’s happening in the AI cycle unless you understand this.

Layer 6 — The Enablers 🔄

Small Cloud’s position is now reinforced and strengthened.

The reinforcement stage is where NVDA confirms their support for CSPs (or “Neoclouds”).

—> In the case of CRWV, with the offtake agreement.

—> In the case of Nebius there is no offtake agreement as of yet, but Nebius is a strategic NVDA partner. This has enabled them to close the major deal with Microsoft.

In this layer, NVDA enables Neoclouds by extending prime access to AI infrastructure — which Hyperscalers require for their business. This allows Neoclouds to close major compute deals with Big Cloud — causing their valuations to skyrocket in the process.

This makes them confident enough to allocate capital beyond AI infrastructure, and to AI-related startups! A baffling situation, but we aren’t done yet.

Closing Thoughts

In Part 3, we will continue on the same thematic offering more proof that markets are operating with the NVDA Put.

We will see how NVDA funds both supply and demand in AI and how Open AI ended up becoming the “subprime lender” of the process.

Philo 🦉