Oracle will have to borrow at least $25B a year to fund AI fantasy, says analyst As part of its $300 billion cloud compute contract with OpenAI, Oracle may need to borrow roughly $100 billion over the next four years to build the datacenters required, according to KeyBanc's projections. KeyBanc Capital Markets reportedly estimates that Big Red may need to raise about $25 billion a year in debt over the …
He notes that the assumption by the media and investors is that AI would always advance and improve. If you look at individual benchmarks, this may seem to be true, but if you do a broader analysis of the changes you quickly see that they're not advancing much at all, it's a tug-o-war. They're improving specific areas and degrading others.
This is immediately apparent in image generation models where the issues are visually clearer, models will improve problematic areas like text clarity or hand anatomy correctness, but also lose the ability to distinguish person from floor, or bring back previously "resolved" issues like the dreaded "same face".
not just that but the diminishing returns are staggering. Ten times the compute resources to generate 10-20% improvement is not exactly sustainable (the actual numbers are hard to derive as there are so many variables but all of the models are suffering this) and the magical land of actual intelligence is still a very, very long way away. But the VCs have sunk so much and some companies have also that the cannot possibly, ever, stop now just in case a competitor makes it to the magical land of actual intelligence, which they won't, but the money has to be piled into the bottomless pit just in case.
Image generation is one of the areas where it's easy to demonstrate the cluelessness of the approach. These models have no domain of knowledge whatsoever, therefore as one to produce a image of a dog and all it can do is mash together previously scraped images of dogs and then layer upon layer of post processing is applied to try and make it less likely that a dystopian nightmare is output. These systems don't know what a dog is, they don't know how they move, the number of limbs and where they should be, or anything like that, it's just copy-and-paste and mash the images together.
This doesn't mean that the tools can't be useful to some extent, but there is no intelligence and domains of knowledge and there never will be through this approach.
Of course it is - for him.
He'll be squirreling away a million here, a million there, stashing the dough in bank accounts he controls.
When the bubble bursts, he won't be left holding the purse in any case. He'll just trot off to whatever mansion he has and open a bottle of champaign for what he will believe is a well-deserved rest.
Meanwhile, Oracle will hit the wall with tens of billions in debt it cannot repay.
Will it fold ? Or will it hike up the licence prices by 1000% ? Who knows ?
It's an interesting comparison between Google and Oracle. Both were unable to keep pace with the huge infrastructure spend or AWS and Azure.
Google specialised to try and make their spend sustainable. While they have challenges, they also have some very sizeable revenue streams even as search falters.
Oracle fell behind for years, depending on renting space rather than building like the others.
Then OpenAI comes along and signs a deal that promises expansion at rates that AWS or Azure would be laughed at for suggesting. The lead times required for power and planning new builds make it time consuming and competition for existing space drives up pricing. Sure, you can get lucky a few times, but Oracle has to get lucky for years and years...
While this is true, the outflows for AI capital spending are more, leaving Oracle cash flow negative to the tune of about $6B last year. Two more quarters of that would complete deplete its cash reserves.
And it's going to get even worse as it starts building datacenters that OpenAI has not - and might not - pay for.
OpenAI has enough cash on hand to carry ORA thru the first couple years of this. Not to pay for the data centers of course (ha ha, are you kidding?) but to cover that $6bn gap until enough capacity is built and they can continue working regardless of what happens to Larry. If he blows up they can move to whoever else is afloat that has over-built and wants their business. Sam Altman is a skeezy monorail salesman here to pit Shelbyville against Springfield.
No, that's not what those numbers mean. If OpenAI tried to host that many GPUs in house, they would end up having to build all the same stuff that Oracle is going to build, and it would probably cost them more because they couldn't easily sell unused capacity to someone else.
Oracle has some money they can spend on building these, especially if you include payments that OpenAI is to give them during the project. That assumes, of course, that OpenAI is able to come up with those payments which they probably aren't, but that's a completely separate problem. That money is included, and since some of it is coming from OpenAI, OpenAI could not count it as savings.
This post has been deleted by its author
He has been demanding money with menaces from his 'allies'. Proper money in return for reduced taxiffs. $550 billion from Japan. In some cases, investment, in others underwriting borrowing. So when the AI bubble bursts, the damage could be offshored. The taxpayers of the UK, EU, Japan, South Korea etc, would be America's financial air bags for the AI crash.
And politicians (can't wire a plug/consider AI to be magic etc) would happily invest in US tech.
Oracle does not have the hundreds of billions required to build all those data centers.
OpenAI does not have $300B to give Oracle. It doesn't even have $30B. Unless two miracles happen, it won't even have the $20B SoftBank has promised it (and which SoftBank doesn't have either.)
And no, NVidia is not going to give OpenAI $100B to build datacenters. It's going to give it $10B at a time, each time it brings a datacenter online. And those datacenters will cost $30B to $50B each.
All of these numbers are completely imaginary. None of this money exists. None of these "commitments" can possibly be fulfilled.
Larry Ellison might as well be sitting on his next quarterly call with his pinkie finger to his mouth demanding "One trillion dollars!".
Lately the AI folks are tossing borrowing numbers around like the US government does. 100B here, 100B there and pretty soon you're talking about real money. And I doubt the AI co's are getting US interest rates. Maybe 7-ish % or 7B per year per 100B borrowed instead of the 4-5% TBills trade around. Staggering numbers. This is going to be a Hindenburg level event when the pimple pops.
Looking into these amounts of money to be "invested" - in the end the investments have to pay off, give some return, right?
How on earth is society capable in paying up all that money (forget "profit" at the moment) for something that has so little contributed to its wellbeing?
In other words: why would I (we, you,...) pay for this "service" ??
All these companies suffer from FOMO - and society gets the bill in the end.
I used to think a $1000 graphics card used purely to make sure every blade of grass is waving around while people play FarCry was a total waste of computing resources and the planets sort-of-finite energy supply .
.
Now we have datacentres with millions of GPUs being used to draw inane doodles the general publics imagination has spunked up , somehow due to the ecosystem of internet economics , at no cost to them , as yet.
