Oracle and OpenAI end plans to expand the stargate data center. The original DC will operate. The deal fell through due to financing negotiations. For the economy not a good news, but for hyperscalers stock price it is a good news, rationalizing cap-ex is much needed.
1 Like
I read an article saying that OpenAI is essentially bankrupt and since they would never be able to pay, the deal would bankrupt Oracle. Did Ellison just save Oracle and his fortune?
The Captain
1 Like
I don’t know which article it is, but saying OpenAI is bankrupt is not reflecting their economic value. If tomorrow they come public with an IPO, they may get a valuation around $1T. If you look at how prolific they are in fundraising the bankrupt commentary looks incorrect. However, given how many deals OpenAI has it may need to resize.
The same goes for Oracle. Oracle has been aggressively buying back its shares for the last 10 years, that also coincided with debt increasing from $43 B to $125 B. The debt increase is not entirely due to buyback, of course investments in DC buildout and acquisitions like cerner (all cash deal for $28 B). This severely constraints their ability to keep investing. In fact, $ORCL which had very high gross margin, the margin has come down due to cloud investments from 80% to 68%, will see further erosion of that margin. There is a good possibility they are going to have negative cash flow. The company is actively considering selling cerner because it cannot invest in cerner and in its hyperscale business simultaneously.
Long story short, Ellison is responsible for the pickle Oracle is in. Stopping further expansion of stargate is not the silver bullet that is going to save $ORCL, rather it is just a reflection of where the company’s finances are at present.
The article didn’t mention bankruptcy, that was my deduction based on what it said about Sam Altman. A successful IPO would change my opinion.
The Captain
At this point lenders must be very aware of the risks of too many data centers to be profitable. The first may very well be profitable but the last built could default. Lenders grow more cautious. Will they demand more security? Higher interest rates?
We also note that the big guys are investing their own cashflow. Not relying on lenders. And then we have Nvidia and AMD selling chips for shares in the buyer–taking equity position.
In a world where everything is digitized and data grows by mountains every day its hard to imagine data centers going dark. They will find users even if not for AI. Or are they overbuilding far ahead of the need curve?
US has a history of overbuilding in every sector. So don’t be surprised.
True but what should concern investors is separating the wheat from the chaff. For example, Amazon didn’t build AWS as an investing venture but for internal use. Once built, it sought customers. Elon’s businesses and others are building data centers for internal use to bolster existing business. Not so Sam Altman who does not have a cash producing business.
It will play out like all does, the 80-20 rule.
The Captain
1 Like
Absolutely not true.
OpenAI TTM revenue run rate is $25 B. In fact OpenAI has the distinction of achieving fastest $10B revenue and on track to do $25B.
OpenAI’s is producing revenue, FCF is a different issue because of their investment plans…
1 Like
My understand is that it is true - and Google says so too:
Amazon did not originally build AWS as a planned, highly profitable subsidiary, rather pivoted an internal infrastructure necessity into a business unit that eventually became a massive profit driver. It evolved from a 2000-era effort to streamline internal, underutilized capacity and [service-oriented architecture] into a public-facing [cloud service].
- Accidental Beginning: Initially, Amazon had to build massive infrastructure for Black Friday and holiday traffic, which sat idle most of the year. The idea was to rent out this excess capacity
1 Like
Thanks Goofy! 
The Captain
Incorrect Goofy.
- The 2003 Executive Retreat: During a retreat at Jeff Bezos’ house in 2003, the leadership team realized that in creating these internal tools, they had become experts at managing massive, reliable, and scalable infrastructure. They realized they could offer these “primitive” building blocks (storage, compute, databases) as a service to other developers.
Amazon had build certain tools, processes, skillset, to support their amazon.com business. Then they realized this can be offered as a service to other developers. This, the capabilities developed internally is put together as a service (AWS) to sell to external clients. AWS is not created to meet internal demands. Only in late 2024, early 2025 Amazon IT and AWS were brought together.
I stand corrected, with a small caveat. It appears the “excess capacity” trope is a myth, or at least partly. The actual business, “AWS” was not created because of excess capacity, it was, as you say, created to ease their internal issues with constant changes to their website, adding links to partner pages, and so on. They did, however, has lots of excess capacity by 2000, a result of preparation for the crush of the holiday and sale seasons - so the two factors combined to present an opportunity to create AWS - first for their own internal issues, and later, to offer to outside customers as a “cloud” solution (the term not in use at the time) for database and other data hungry applications.
Without the capacity, AWS would not have been practical. But without the internal need for better data handling it wouldn’t have been developed at all. AWS launched in 2006, FWIW and took off slowly at first, then like a rocket.
Thanks for the correction. Fun learning the real story behind the story.
Let me further clarify. While some data centers hosted both AWS and amazon.com workload, they did not share physical resources, for ex: a single server will not have VM’s for both AWS and Amazon.com; They had NW china wall… The excess capacity used by AWS is a myth.