Artificial intelligence chips are being upgraded faster than data centers can be built, a reality of the market that exposes AI transactions and businesses to significant risks. oracle’s Expansion fueled by debt.
OpenAI no longer plans to expand its partnership with Oracle in Abilene, Texas, where the Stargate data center is located. Because OpenAI requires a newer generation of clusters. Nvidia A graphics processing unit, according to a person familiar with the matter.
The current Abilene site will use Nvidia’s Blackwell processors, but it is expected to take a year to get power up and running. By then, OpenAI hopes to expand access to Nvidia’s next-generation chips in larger clusters elsewhere, said the person, who requested anonymity to protect confidentiality.
Bloomberg first reported that the company had ended its expansion plans in Abilene. Oracle criticized reports about the activity as “false and inaccurate” in a post on X on Sunday, but the post only said existing projects were progressing well and did not mention any expansion plans.
Oracle secured land, ordered hardware, and spent billions of dollars on construction and staff in hopes of scaling.
An Oracle spokesperson declined to comment.
This is a natural decision for OpenAI, which doesn’t want older chips. Nvidia used to release a new generation of data center processors every two years. CEO Jensen Huang currently ships one unit every year, with each generation dramatically improving in functionality. Vera Rubin, announced at CES in January and already in production, delivers five times the inference performance of Blackwell.
For companies building Frontier models, small improvements in performance can equate to large differences in model benchmarks and rankings, which developers track closely and translate directly into usage, revenue, and ratings.
All of this points to a larger problem. For infrastructure companies, it takes at least 12 to 24 months to secure land, connect power, and start up a facility. But customers want the latest and greatest and track yearly chip upgrades.
An additional challenge for Oracle is that it is the only hyperscaler to primarily fund its expansion with debt. google, Amazon and microsoftIn contrast, they rely on huge cash-generating businesses.
Meanwhile, Oracle partner Blue Owl is refusing to fund additional facilities and plans to cut up to 30,000 jobs.
Oracle will report third-quarter results on Tuesday, but investors will be watching to see how the company handles its $50 billion capital spending plan amid negative free cash flow and whether its funding pipeline holds up.
The company’s stock price has fallen 23% since the beginning of the year and has lost more than half its value since its peak in September.
Beyond Oracle, GPU depreciation is a risk to the broader market and could impact the entire AI landscape. Any infrastructure contract signed today could end up using outdated hardware before power is even connected.
WATCH: Jefferies’ Brent Till speaks to CNBC ahead of Oracle earnings
