AI cloud provider Corvex has raised 33 million dollars to grow from 1.5MW to 8MW by the end of 2026, and says the scarce input is energised power inside a finished building rather than chips.
What was announced
- Corvex secured 33 million dollars from a private placement of common stock and a Series D round to fund expansion from its current 1.5MW footprint to 8MW, targeted by the end of 2026.
- The plan doubles capacity at its existing Mid-Atlantic data centre and adds a Midwest site, bringing 2,000 latest-generation GPUs online in the Midwest and 1,000 more in the Mid-Atlantic.
- Corvex's platform lists NVIDIA B300, GB200, B200 and H200 GPUs, and the company expects revenue from the new deployments from Q1 2027.
- It holds a right of first refusal on a further 12.5MW at the Midwest site, ready for service in Q3 2027, which would take its footprint past 20MW.
- Co-CEO Jay Crystal said the scarce input in AI infrastructure is energised power inside a finished building.
The ETON view
The number that matters here is not 33 million dollars, it is 1.5MW growing to 8MW. A provider with capital, GPU allocation and customer demand still measures itself in megawatts of energised, commissioned space, and it is booking revenue from that capacity four to six months after the funding lands. That is the real clock on AI infrastructure in 2026, and it is the same clock UK hosting providers and enterprises are running against.
If power and shell are the constraint, then every kilowatt has to earn its place. That pushes buying decisions towards performance per rack and per watt, towards higher density per U rather than more racks, and towards kit that can be delivered into a slot you already have energised instead of kit you wait two or three quarters for. A mixed estate of H200 and B200 class systems bought on availability will usually beat a single-generation order that lands after your power window has been reallocated.
ETON's role is to make that timing work. We source across NVIDIA, HPE, Dell, ASUS, GIGABYTE and Supermicro rather than a single vendor's allocation queue, we quote against what can actually ship, and we deal in both new and certified refurbished so a rack can be filled now instead of held open. We will also tell you when the honest answer is that the generation you want cannot be sourced in your window, and what the sensible interim build looks like.
Related infrastructure
Category: AI & GPU · Vendor: NVIDIA · Technology: B300, GB200, B200, H200 · Last verified: 08 Sep 2026 · ~2 min read
Sourcing this kind of infrastructure? Talk to ETON about availability, lead time and pricing.
