AI infrastructure, built energy-first. We develop, deploy & operate power-ready data centers for the AI era. Energy + Compute, Unified. We're here & we respond
How do you get a GPU running AI inference to respond to the grid?
One integrated loop: ERCOT conditions β Luxor's QSE signal β @ZianiSystems orchestration β GPU curtails to ~25%. Under half a second, no lost work.
The full picture π
Together with @BentausOfficial's Zianiβ’ software, Luxor has successfully demonstrated AI Inference load flexibility for the first time.
A NVIDIA B200 GPU responded to Luxor Energy's 4CP curtailment signals in ERCOT, curtailing power consumption down to 25% in under half a second, with no disruption to its AI inference workloads.
A live AI GPU responded to @ERCOT_ISO 4CP curtailment signals for the 1st time. Signal to verified curtailment in under 0.5 second. Power down to 25%. Inference workload uninterrupted. @Luxor's Ethan Vera walks through how it went. Watch the video β¬οΈ
https://t.co/tuOFnV3Oou
First response of an AI inference GPU to a live @ERCOT_ISO 4CP signal.
@luxor Energy's QSE generated the curtailment signal. @ZianiSystems by @BentausOfficial took the GPU from full draw to roughly 25% in under 500 milliseconds, closed loop and verified, with inference workloads checkpointed, paused, and resumed. Nothing dropped.
Four fifteen minute peaks set a large share of a Texas data center's annual transmission bill. Now AI load can respond to them. ERS enrollment next.
https://t.co/l67tLbBXWr
Luxor Energy, the energy division of @luxor and @BentausOfficial report that their Ziani platform cut a live GPU's power draw to 25% of normal in under 500 milliseconds, while inference workloads kept running. It's the first demonstrated AI GPU response to an ERCOT 4CP signal.
Why it matters: ERCOT's Four Coincident Peak program rewards large loads that pull back during the grid's highest demand intervals each summer (June through September). Until now, AI data centers have largely sat outside that conversation as fixed, inflexible load. This result suggests that doesn't have to be the case. GPU fleets could become active participants in market programs, not just consumers of power, with real potential to lower transmission costs.
Congrats to the Luxor Energy and Bentaus teams on a genuinely useful proof point for grid-aware AI infrastructure.
https://t.co/3dwPdUrg8d
Two weeks out from AI Infra Summit. Our CEO Bob Davidoff and COO Kevin Matcham will be in Santa Clara September 15 to 17, and the reason to go is the same reason it is always worth going. Hammerspace, DriveNets, Supermicro, AMD, and the rest of the partners are all in one place, and the conversations that move things forward happen between sessions rather than during them.
If you are building AI capacity against a power constraint, that is the conversation we want to have. Come find us.
Bentaus is heading to #YOTTA2026, the event bringing together the entire digital infrastructure ecosystem to solve one problem: how to scale AI compute sustainably as demand explodes. Three days. The people building the future of digital infrastructure, in one room.
Our team, comprised of Robert Davidoff Kevin Matcham Erika Leal Nelson Bruton and Davy Wittock, will be on the ground connecting, comparing notes, and finding the next partnership.
If you're building AI infrastructure and thinking about power as the real bottleneck, come find us.
#AI #NeoCloud #DigitalInfrastructure #DataCenters
The @AMD Advancing AI 2026. San Francisco last week. Good to be part of the room where this next generation gets built.
The best part of Advancing AI last week was not on stage, it was meeting the people and the conversations in between.
Back to work.
The @AMD Advancing AI 2026. San Francisco last week. Good to be part of the room where this next generation gets built.
The best part of Advancing AI last week was not on stage, it was meeting the people and the conversations in between.
Back to work.
βοΈ AI demand is not the hard part. Reliable infrastructure is. It's not who has GPUs, it's who can deploy them well.
βοΈ We are deploying NVIDIA B300 that gives teams more headroom for larger models, longer context windows, and heavier inference workloads.
βοΈ The value proposition is straightforward: more memory, more bandwidth and more room to run demanding workloads without constant compromise.
π Get NVIDIA B300 https://t.co/4KJWGMuTVX
1/2 While the market talks about early access to B200s, we are now deploying B300s -> Execution beats roadmaps!
Our Texas and Colorado facilities light up 60 nodes this September.
β 480 Blackwell Ultra GPUs entering production.
β Tuned for maximum token input and output.
β Fully managed NeoCloud infrastructure.
Compute is only useful when it's live. >>
Day 1 in Hamburg is officially under way.
Today, @Supermicro opened BoothB10 at ISC High Performance, and the newest version of SuperCloud Composer is live with our @ZianiSystems Systems' GPU Flex built in for the first time.
This is what months of engineering, testing, and partnership with Supermicro look like when it reaches the floor. We are sharing a few photos of Galyn Black and Jeremy Swyers from the event today.
If you are at ISC 2026, you know where to find them.
If you are not, follow along here as we share what happens next.
Hamburg. #ISC2026
We are thrilled to see our CEO, @bobdavidoff, take the stage at Innovate APAC in Taipei alongside Kevin Culp, Trent Prestegar, and our partners at Supermicro! π
As the energy demands of artificial intelligence rapidly scale, the need for flexible, grid-responsive infrastructure has never been more critical. At Bentaus, we are proud to collaborate with Supermicro to bridge the gap between #AIInfrastructure and energy orchestrationβturning high-density AI factories into Virtual Power Plants.
By bringing together real-time power control and grid-aware computing, we are paving the way for the next generation of resilient AI infrastructure.
If you are attending Innovate APAC or COMPUTEX TAIPEI, be sure to connect with Robert to discuss BYOP (Bring Your Own Power) and the future of energy-smart data centers.