.png)
3 weeks ago
2
Everything is bigger in Texas and that includes the GPU bit barns at the heart of the AI boom.
The Lone Star State is quickly emerging as a datacenter mecca, thanks in part to an abundant supply of natural gas necessary to power the multi-gigawatt-scale facilities.
This week, Meta announced its 29th datacenter project, which will be built in El Paso, Texas, and eventually scale to a gigawatt of compute capacity by 2028. Power for the facility will be supplied by El Paso Electric, which sources power predominantly from natural gas, nuclear, and solar. But don't worry, Meta says that all the energy consumed at the site will somehow be matched with renewable energy.
The initial investment - like many large-scale datacenter build outs, Meta's latest site will be built in phases - is expected to cost somewhere in the neighborhood of $1.5 billion.
At the same time, the Wall Street Journal reports that Nvidia-backed AI dev Poolside is working with rent-a-GPU outfit CoreWeave to deploy a two-gigawatt bit barn on West Texas oil fields.
To say AI is the new oil isn't just a metaphor. In this case, the two companies plan to power the facility using onsite generators fueled by natural gas pulled straight from the source.
They argue that, by building the facility in close proximity to these resources, they can cut costs and speed up development. Much like Meta's El Paso datacenter, Poolside's will be built in phases, with CoreWeave set to be the anchor customer for the first 250 megawatts of capacity beginning in 2027.
While they wait, Poolside plans to rent a 40,000 GPU cluster of Nvidia GB300 rack systems from CoreWeave.
In total, the facility is expected to set the AI dev turned infrastructure provider back about $16 billion, not including compute hardware, though Poolside execs hope to cut costs by utilizing a modular design and construction philosophy.
While Poolside and CoreWeave plan to run the site on fossilized dinosaur remains, ECL is working on a one-gigawatt datacenter on the other side of the state in Houston that'll supposedly be powered by hydrogen fuel cells.
Lambda is set to be ECL's first tenant. However, despite plans to bring the first 50 megawatts of capacity online this summer, we're told that, as of late September, it isn't quite ready yet.
- Feeling lonely? Microsoft Copilot can now listen to your every word, watch your screen
- Salesforce's Benioff warns of AI 'false prophets' while promising true profits by 2030
- Oracle goes all-in on AI, customers still figuring out how they'll use it
- US hyperscalers to guzzle 22% more grid juice by end of 2025
Then, of course, there's OpenAI's flagship Stargate campus in Abilene, where partners Crusoe and Oracle recently brought the first 200 MW of what is expected to scale to 1.2 GW by the end of next year.
This week, Oracle founder and CTO Larry Ellison announced that, when complete, the site would house 450,000 Nvidia GPUs, the equivalent of 6,250 Nvidia GB200 NVL72 rack systems.
None of this, we should note, is going to be cheap. A single NVL72 rack is estimated to cost $3.5 million, though at the volumes Oracle, CoreWeave, and others are deploying these systems, they're likely receiving some kind of volume discount from Nvidia. ®
