M3 hasn't been published in the US since 2006 and M4 has never been published publicly in the US. Aggregates don't turn into one another like they're moving on a conveyer belt. They're just classification buckets. If you have a dollar in a checking account, it's in M1, M2, M3 and M4 at the same time.
A purchase commitment or a capacity guarantee on Nvidia's balance sheet is just a contract. It isn't in any aggregate, it isn't traded in a money market and nobody accepts it as payment.
> ...via a similar mechanism to the Fed buying Treasuries, thereby increasing deposits at the Federal Reserve (MB) which in turn prompts banks to increase M1
Sorry but this is just wrong. Only the Fed creates reserves. Nvidia signing a contract doesn't do anything at the Fed.
Banks create deposits when they lend and they're constrained by capital and loan demand, not a reserve ratio. Since 2008, reserves have gone up 10x and M1 hasn't. See "Money creation in the modern economy". The multiplier theory as an explanation of how money gets created has been dead for years.
> The $500 billion isn't net-sixty trade credit
I wasn't even talking about this. My comment addressed the OP's argument that Nvidia is holding customer debt as a bank-like asset. That debt is receivables and it's 53 days on average, not years. The $500 billion in the article is mostly bank and private credit cash being loaned to Nvidia's customers and Nvidia providing backstops and guarantees. So if money is being created here, it's the lenders who are creating it, not Nvidia. In this case, Nvidia isn't a bank, it's a credit enhancer.
The legitimate concern is that Nvidia's guarantees are encouraging lenders to lend money to neoclouds on better terms than they otherwise would get (or should get if you want to make that argument). But that's not money creation and trying to pretend that it is only distracts from the real issues.