- News at OCP's Global Summit this year spans data center pods, power and cooling, all in service of AI
- Flex highlighted rising demand for modular data center solutions that can be plopped onsite
- Meanwhile, Accelsius forged ahead with two-phase cooling gear and 800V DC power architecture got some major backing
The Open Compute Project's (OCP) Global Summit in San Jose is expected to be bigger and better than ever this year, with attendance forecasted to jump from 7,400 last year to more than 11,000 in 2025. Why? Because compute power – and all of its supporting technologies – has never been more important.
Naturally, there’s a flood of news coming out of the show, but we thought these three announcements should definitely NOT fly under the radar.
Flex
What it is: The company’s new AI infrastructure platform is a modular stack that wraps the company’s power, cooling, compute and services solutions up with a pretty bow to help data center operators and enterprises alike deploy AI infrastructure faster.
The package includes rack designs to 1 MW (we told you those were coming) with 400 volt (V) and 800V direct current (DC) iterations; an energy storage system to help avoid AI workload disruption; and a rack-level coolant distribution unit (CDU). The platform can be delivered in prefabricated skids or pods that can be delivered to a data center site.
Why it matters: Flex claims the new platform can help slash the traditional 18-month time to market for AI infrastructure by 30%. But it’s choice to deliver a modular, prefabricated stack is reflective of a growing trend in the market.
“It seems like every single customer is switching to more of a prefabricated solution,” Flex President Chris Butler said on a call with journalists. “I won’t tell you who exactly, but it’s the usual suspects and a combination of both hyperscalers and co-lo providers around the world. Super strong growth in that area.”
Butler added that in response to this rising demand, Flex doubled its manufacturing capacity for power pods over the last nine months. Now, the demand has expanded to include not just prefabricated power solutions, but also compute and cooling as well.
Flex isn’t the only company seeing this, either. Vertiv CEO Giordano Albertazzi highlighted increased demand for its modular data center solutions on the company’s Q2 2025 earnings call.
Accelsius
What it is: The company announced the general availability of its NeuCool MR250 system, an in-row two-phase coolant distribution unit (CDU) designed to support up to 250 kW of capacity.
The MR250 was first mentioned in May when the company announced its plans to participate in the U.S. Department of Energy’s ARPA-E COOLERCHIPS Project, but wasn’t broadly available.
Why it matters: The MR250 is the first in a series of beefy CDUs Accelsius is planning to offer to support higher rack densities in the AI era. Additional releases are set for next year.
As you can tell from Flex’s 1 MW rack announcement above, there’s certainly a need for more cooling power. Indeed, Dell’Oro Group recently noted that direct liquid cooling revenue jumped 156% year on year in Q2 2025 and stated the technology has become “the de facto standard” for AI clusters.
But among direct-to-chip cooling vendors, Accelsius is one of few focused on two-phase technology, with Zutacore among its peers (others like LiquidStack are focused on two-phase immersion cooling, which is very different).
All of this matters because there have been concerns that single-phase direct-to-chip cooling could be nearing its limits even as chip power levels continue to rise. Two-phase cooling, which involves converting the cooling liquid to a gas and back in a closed loop, can help increase the level of cooling offered.
Some, like JetCool, insist single-phase technology has a long runway, but it’s interesting to see Accelsius’ two-phase system take off. The company said a handful of MR250 deployments began late in Q3 and teased “broader rollouts continuing through the fourth quarter and into 2026.”
Nvidia, Heron Power and others
What it is: Ok, we’ll admit, this one is less of a single announcement and more an observation of a trend found across several at OCP: the rise of 400V and 800V DC power architectures.
Why it matters: In short, this marks a shift both away from alternating current (AC) configurations as well as toward drastically higher voltage. And it matters because 800V DC in particular now has backing from Nvidia as well as a slew of other ecosystem partners including Oracle, Vertiv, Foxconn, CoreWeave, Lambda, Nebius, Flex, Hitachi, Schneider Electric and Siemens. There are even startups like Heron Power targeting this space and working with the likes of Nvidia and Crusoe.
This all marks “a major step in rethinking how energy flows through the data center,” Dell’Oro Group Research Director Alex Cordovil noted on LinkedIn. “This shift toward high-voltage DC distribution will ripple across the supply chain — from converters and busbars to breakers and energy-storage systems — driving new innovation across the entire physical infrastructure stack.”