honored to be included in the @BessemerVP ai data center map. reliability engine is building the liquid cooling infrastructure that helps the whole ecosystem scale
Our newest roadmap on the AI data center stack is live!
Here @BessemerVP, we often have 5-10 main themes (aka roadmaps) running through the fund. They almost always stem from a founder's unique insight accompanied by an indisputable "why now".
We would be hard pressed to find a "why now" more obvious than the current infrastructure buildout needed to support token demand.
As we dove into this new world, Brie, Josh, and I wanted to be humble and honest about what we didn't know - which was a lot! We're now excited to open source our research detailing the six main areas of opportunity we see in the space:
1/ Permitting and site selection
2/ Power generation
3/ Transmission, power conversion, and the middle mile of power
4/ Software and orchestration
5/ Construction, maintenance, and labor
6/ Cooling
Please reach out to us [[email protected]] if you're building in the space! We also can't wait to announce some of our early investments on this roadmap soon.
Thanks so much to @davidcowan, Olivia Wang (Research @ Sightline Climate), Karly Wentz (Partner, Energy Tech @ B Capital), and Megan Cain (Westly Group) for your thought partnership and feedback!
Stable pH does not tell the whole story.
A coolant buffer can hide acid buildup while the dashboard looks normal. When the buffer runs out, the fluid can turn corrosive and damage cold plates.
Part 1: https://t.co/2z543bx3bT
Your GPUs aren't failing first. Your coolant is.
Conductivity drift. pH slide. Rising particles.
Liquid cooling loops warn you for days before racks overheat. AI turns those signals into a forecast, not a fire drill. 👇
https://t.co/QmPYegHVzP
#DataCenter#LiquidCooling#AI
Liquid cooling isn’t a plug-and-play upgrade.
Connect loops in series instead of parallel? You overpressure your infrastructure. Miss a 20-micron filter swap? You clog a cold plate.
The step-by-step guide to D2C commissioning: https://t.co/oEn8Cu6QpW
#LiquidCooling#DataCenter
everyone talks about ai in tokens and models. the physical reality is steel. a 1mw liquid cooling loop doesn’t appear from a figma board. someone has to machine the parts, weld the pipe, and move the liquid. that’s what scaling a supercluster actually looks like
In direct-to-chip cooling, failure doesn't announce itself. It accumulates. Coolant degrades, sensors drift, microchannels clog; all while your dashboard stays green.
5 maintenance disciplines to stop the silent decline: https://t.co/ClFXWBXhqf
#DataCenter#LiquidCooling#AI
Your liquid cooling pipes won't blow up. 72 GPUs in parallel = ~3 PSI drop. The real drain is from UQDs, hoses, and bends. Bad plumbing can force the pump to eat 15-20% more power. That's the Token Tax.
Stop fighting the compute. Master the friction:🔗 https://t.co/mDwF9eFsCu
We experienced an outage at Coinbase last night, which is never acceptable. The root cause was a room overheating in an AWS datacenter when multiple chillers failed. We design our services to be redundant to downtime in any one AWS Availability Zone (AZ), and most of our systems worked this way last night, but not all.
Our centralized exchange did not. Exchanges have unique architectures that optimize for latency and co-location of clients. It is possible to make exchanges resistant to AZ failures, but this can introduce latency delays that are not desirable along with breaking customer co-location. Given this incident, we'll revisit these tradeoffs to ensure we're giving you the best possible venue to trade. At a minimum, the duration of an outage should be able to be reduced considerably when an AZ move is needed.
Thank you to the AWS and Coinbase teams for working through the night to mitigate the issue. We’ll share the detailed technical summary once it's ready.
Retrofitting data centers for immersion cooling is like turning an office into an aquarium. Raised floors can't hold heavy fluid vats.
Direct-to-Chip (D2C) cools silicon without structural overhauls.
The breakdown: https://t.co/PcaDwOYWop
#DataCenter#LiquidCooling
@chigrl we’re splitting atoms to boil water so we can cool servers with more water. the ai nuclear campus is two industrial fluid systems arguing with each other
AI is driving a major surge in energy demand.
That’s why I’m proud to advance the bipartisan Liquid Cooling for AI Act in my Energy Subcmte. this week.
My bill w/ @ChrisCoons will help strengthen energy management, computing infrastructure, and U.S. leadership in AI ⚡️🇺🇸
@mvcinvesting the dutch infra engineers are all six foot five and can violently rack a two hundred pound liquid cooled server without a step stool. biology is a capex advantage
@PeterDiamandis we wont have to think until we hit a rate limit. the moment the utility stops working we will regress into feral apes because we completely forgot how to think
@jukan05 loaning your boyfriend a million dollars to buy a house and recording your own rent payments to him as passive interest income. we are in the schizo accounting era
@firstadopter while we wait for more GPUs we can enhance the existing cluster and improve reliability by reducing thermal throttling because an invisible layer of biofilm choked the cold plates. we haven’t fully used the tools already available to us.
@wallstengine 50 million buys exactly two racks of gb200s and the physical plumbing required to run them. infra capex is going to violently vaporize the shoe guys