Oracle’s rock-and-hard place in Abilene TX: building out a data center with Nvidia chips that are already obsolete–and the financing it takes (updated)

March 10, 2026 | by

Another mystery solved, not looking good for Oracle. In yesterday’s update to pending record layoffs at Oracle, affecting Oracle Health, new reporting mentioned the breakdown of Oracle’s expansion agreement with OpenAI in building out a leased data center for OpenAI in Texas. Additional details have now come to light in reporting by Bloomberg (hat tip to Brody Ford and team), with separate reporting by CNBC.

The overall impression is not a good look for Oracle. This analysis combines both articles from their sources, plus additional background.

  • The location is part of a 1,000 acre site in Abilene. It is not just any datacenter build site. It is the first part of the $500 million Stargate Project, announced last year by the White House–a multi-year, public-private initiative with the objective of creating leading AI infrastructure within the US. The initial equity funders are OpenAI, Oracle, MGX, and SoftBank. OpenAI release 21 Jan 2025
  • Several parts are already built by the developer, Crusoe, and are up and running. But power for much of it won’t be on for another year.
  • Oracle has already filled the Abilene site with servers which are used by OpenAI for training and deployment. These facilities are on track.
  • Oracle, Crusoe, and OpenAI had been discussing since midpoint 2025 about expanding the facility by almost double, from 1.2 to 2 gigawatts of power demand. (See below for explanation of how big a gigawatt is.)

The differences between Oracle and OpenAI, two of the Stargate equity investors, apparently center on timing for opening the expansion site, which won’t be till next year; the use of a now older generation of Nvidia chips; and Oracle’s financing for the site. There are also differences with Crusoe, the developer.

  • Oracle had already committed to the site expansion, ordered the hardware, and already spent billions on construction and staff. The processing power would be based on Nvidia’s Blackwell chips.
  • The problem: Nvidia is now bringing out chips every year. It is already producing its Vera Rubin* chip, unveiled at January’s CES. Vera Rubin delivers five times the inference performance of Blackwell. Inference is critical to AI actually doing a job based on real world data.
  • OpenAI was not happy about being tied to Oracle’s commitment to Blackwell, already a less powerful and capable chip. There was also Oracle’s unhappiness over OpenAI’s often-changing demand forecasting.
  • Crusoe was also unhappy about the current Oracle facility’s downtime for days during the winter, attributed to weather affecting some of the liquid cooling machinery. This points to Oracle’s planning and building leading to reliability problems.

Three part unhappiness=plan cancellation. Meta is considering leasing the expansion site, brought to Crusoe by Nvidia, according to Bloomberg’s sources.

The ‘hard place’ that Oracle is in is this: the construction of data centers and their power sources is a 12 to 24 month taffy pull. At minimum. Things change in that time, like chips. Meanwhile, one ‘rock’ is that computing power for AI, whether Nvidia or AMD, is growing every year. First line AI companies like OpenAI (or Anthropic for that matter) want the latest, because that is critical to their business. Another ‘rock’ that Oracle has is that the datacenters are being financed via at least $100 billion in debt. Google, Amazon, Meta, and Microsoft are able to finance datacenter builds via their cash-generating businesses, even if this hyperscaling means that the cash cows become somewhat starved for feed. Oracle has to advance money in construction and equipment it must raise in debt markets for a return that may come in a year, two, or even more. [TTA 5 Feb] In other words, Oracle is in a tight spot compared to competition. The likely solution? Further downsize its businesses and employees to afford the Ellisonian Transformation as noted yesterday. 

The ‘rub’ of course is that what the OpenAIs and Anthropics want–the latest and greatest chip in their datacenters–isn’t possible in the brick-and-mortar world. Not even if you have tremendous cash flow out of your faucets and Blackbeard’s Chest in your bedroom.

Update 10 March: Oracle denies all of the above reporting. Here is Oracle’s reply on X.

Recent media activity about the Abilene site are false and incorrect. First, Crusoe and Oracle are operating in lockstep to deliver one of the world’s largest AI Data centers in Abilene at record-breaking pace. Two buildings are completely operational and the rest of the campus is on track. Second, Oracle has completed leasing for the additional 4.5GW to deliver on our commitments to OpenAI.

Unpacking this, Abilene is already a center for Oracle and OpenAI, as noted above. The expansion was only supposed to be 2 gigawatts, not 4.5, but Oracle and OpenAI have other sites that this could be referring to.

In effect, Larry Ellison, a 40% shareholder of Oracle, should bet the farms, the boats, and Lana’i to make these AI datacenters happen–if he really believes this is the future. Will he? And will they be revenue positive–quickly–to pay off the bet? Mr. Market’s stock price is still stuck on skepticism.

A corollary issue: there is major pushback against datacenters, rising up like daffodils in the early spring, but far less beautiful:

  • Extreme power consumption of datacenters causes rising rates for commercial and residential users. The Crusoe Oracle facility, according to the Bloomberg article, uses about 1.2 gigawatts and was seeking to expand to 2 gigawatts. One gigawatt is equivalent to one nuclear reactor and power to 750,000 homes.
  • Their massive, brutalist landscape footprint. If you like warehouses, you’ll love datacenters filling what used to be fields.
  • Their low employment after they are built, the scale of tax incentives that are being dished out, versus the capital investment required. This article from Futurism reports that one heavily subsidized datacenter facility run by Ark Data Centers in Ohio will employ upon completion exactly 10 people. Yet it is being heavily subsidized by Ohio through a 50%, ten-year sales exemption covering mainly new equipment–estimated to total $4.5 million. That is $450,000 per person over the 10 year span, or $45,000 per year subsidizing generally lower wage jobs in IT and security. One cited analysis, which could be exaggerated, indicated that in Virginia, one datacenter job took 100 times the capital investment for similar jobs in other industries. (Editor’s note: yes, these are single analyses and could be biased, nor factor in cash flow.)

Returning to Oracle, we await another shoe drop today on their earnings, projected layoffs, and their impact on Oracle Health.

*Named after the pioneering American astronomer, known for her work on galaxy rotational rates and the discovery of ‘dark matter’.

Breaking–Oracle to lay off thousands due to AI data center cash crunch, possibly as early as next week. What’s next? (Updated)

March 6, 2026 | by

We now know another piece of the puzzle on why so many Oracle Health top executives have departed. Bloomberg’s Brody Ford has followed up his earlier report on five departures  of key executives at Oracle Health [TTA 3 Mar] with the not-unsurprising news that there will be thousands of layoffs at Oracle, starting as early as this month. The reason why is Oracle’s aggressive expansion into data centers and the shortage of free or loaned cash available for that expansion, necessary to remain competitive in cloud computing with Amazon and Microsoft. (That situation, and the speculation around it, is explored in more detail in our article here.)

The pennydrop was as early as last September in a filing, according to Mr. Ford. It was estimated in the filing that $1.6 billion in restructuring costs will hit this FY, which ends in May. Oracle as of last May had 162,000 employees worldwide.

According to Mr. Ford’s sources, the layoffs will not be the ‘usual’ rolling layoffs, but wider reaching. He cites an internal announcement that “it would be reviewing many of the open job listings in its cloud division.” Some of the cuts will be targeting jobs Oracle needs fewer of because of AI. He cites the reception of Microsoft’s AI-related layoffs and Block, Inc, founded by CEO Jack Dorsey, laying off nearly half of its staff due to supposed leaps in AI (but more likely due to ballooning hiring not compatible with cash flow).

The scuttlebutt on Reddit indicate the cuts could be as high as 20% with the US operation hard hit, and strike as early as next week. Since Oracle has not been shy about cutting jobs over the years (see Mr. Ford’s article), this high number is a surprise. Another bit of information gleaned off Reddit is that the OHAI reporting line has changed from TK Anand to “Clay”–possibly co-CEO Clay Magouyrk, versus Mike Sicilia who testified before Congress two years ago when the VA implementation cratered?

Editor’s analysis and opinion: With five major executives leaving OHAI (Oracle Health and AI), she continues to believe that many of the cuts will hit the health area. Yet OHAI is the area that has taken tons of flak from current customers, from Congress on Veterans Health, from the VA, and from health systems. 

  • Oracle has major Federal contracts. The prominence of the VA contract and rollout timing makes cuts in this area problematic. Just because EHR problems have supposedly been fixed and that both the VA and Oracle are set to roll it out, VISN by VISN, does not mean that AI can do it. It is a long and customized implementation due to the sheer number of VA locations and diversity of functions [TTA 8 Feb]. And for that, you need people with deep experience and buckets of patience who know the system and can get along with their Federal counterparts. VistA in over two decades of implementation was so highly customized for both patient care and additional areas such as research that Oracle, in replacing it to VA satisfaction and to be better than VistA, has to accommodate a lot of, shall we say, discovery along the way.
  • In health systems, the discontent with Oracle was about declining vendor partnership and communication. This points to problems with people and continuity. This was highly apparent in the KLAS survey from October 2025 cited here. When half of the interviewees tell KLAS that they would not buy the system again, that is disastrous.

Apparently missing in action is Seema Verma, the general manager of OHAI.

When your current customers providing your business and cash flow are restive, yet what’s coming out of Oracle has been about 1) refocusing on cloud computing and AI datacenter contracts, not health, 2) massive job cuts to pay for them disproportionately affecting Oracle Health, 3) rumors about a sale of Oracle Health to pay for the datacenters, and 4) still paying a $1.4 billion dividend to shareholders that largely benefits Larry Ellison, holder of 40% of stock–what are the next pennies to drop? Stay tuned!

Sources for this article: Bloomberg, Investor.com

Updated 9 March. SimplyWallSt pegged the layoffs at 18%. One of Oracle’s data center contracts is with OpenAI, but they canceled a large planned AI data center expansion in Texas. Other potential tenants, including Meta, reportedly are interested in the site. Their analysis depicts Oracle as “trying to reconcile very large capital commitments to AI data centers, negative cash flow pressure, and debt and equity raises, with the operational reality of supporting customers such as OpenAI, xAI and Meta.” Yet they are aggressively pushing AI through promotion in healthcare, F1 racing, and construction. Their rock-and-hard place is making commitments versus not having the cash to quickly fulfill them. This returns to our 5 February report. Tuesday is the day that Oracle reports results.

Oracle’s Ellison set last quarter the company’s transformation as three steps: From the Fortune article:

  • Oracle making its database available inside its competitors’ clouds, including Amazon’s AWS, Alphabet’s Google, and Microsoft’s Azure.
  • “Vectorizing” the data to make it readable by AI models, which makes the data customers have in Oracle’s systems more valuable.
  • Building an “AI Lakehouse,” which vectorizes all a company’s data and not just what’s in Oracle databases or applications.

But what if you don’t want your data ‘vectorized’ to be read by AI models? Something called proprietary information and data comes to mind, like business and marketing plans. What about PHI and PII? Those could be the danger points to consider in this ‘transformation’. (Forgive me for being oh-so-tired of ‘transformation’–the last time Mr. Ellison trumpeted this was for…Oracle Health, which may be hollowed out to finance this.)