Nautilus Data Technologies

Greenfield AI facilities are now designed for 200+ kW per rack. Retrofit facilities top out near 80 to 120 kW. That gap defines the competitive landscape for the next decade of colocation. The CDU architecture an operator selects today determines which side of it they compete on. How colos are deploying CDUs in 2026: https://lnkd.in/gqP2nnce #DataCenters #AIInfrastructure

NVIDIA's Rubin platform is being read as a chip announcement. That is the wrong frame. Rubin is a statement about how AI infrastructure gets built from here. Compute, networking, memory, power, and cooling now function as a single system. They are not specified separately, and they cannot be cooled separately. The consequential detail: Rubin and its successors are designed to operate with warmer coolant temperatures than legacy data center practice assumed. Three things change when supply water gets warmer. First, mechanical chillers become optional in more climates. Heat rejection efficiency improves across the entire energy chain, not only inside the facility boundary. Second, heat reuse becomes economically viable. Coolant exiting at 45 to 50C and above can feed district heating, industrial preheating, and water treatment. The data center stops being an isolated energy sink and becomes an infrastructure asset. Third, water consumption becomes a design choice rather than a silicon constraint. Closed loops, dry coolers, and economizers become viable in arid climates that previously required evaporative cooling. The industry spent two decades optimizing for the coldest available supply water. The next decade will reward operators who design for thermal headroom and flexibility instead. Those already building for warm water hold a structural advantage. Those still designing around 15C supply have a retrofit conversation ahead of them. Our full analysis of what Rubin means for the cooling industry: https://lnkd.in/g8NfGbRH #NVIDIA #Rubin #LiquidCooling #AIInfrastructure #DataCenters

Nautilus has operated a fully liquid-cooled AI data center in production for more than 650,000 unit hours. Not a pilot. A revenue-generating facility running customer workloads. The distinction matters. Every cooling vendor has a reference architecture. Few have failure data from the third year of continuous operation. Few can describe what happens to a CDU when a tenant doubles GPU density mid-contract, because few have been there. Operational experience compounds into design decisions. Our approach to water source flexibility, modular hydraulics, and field serviceability was shaped by production conditions, not modeling assumptions. That experience is now built into every EcoCore deployment. See the EcoCore platform: https://lnkd.in/gnKfeWCn #DataCenters #LiquidCooling #AIInfrastructure 

Retrofits account for 65% of new CDU installations across North American colocation facilities this year. The economics explain why. Capex runs 20 to 40% below greenfield. Time-to-market compresses from 24 to 36 months down to 8 to 14. For operators with existing grid contracts, permitting overhead drops 30 to 50%. The constraints are structural, not financial. Facilities built 10 to 15 years ago often cannot support overhead coolant piping without reinforcement. Raised floors congest quickly. Cut-over to coolant loops on a live facility carries genuine downtime risk. The operators executing retrofits well are not the ones with the largest budgets. They are the ones who selected modular CDUs with quick-disconnect couplings before the structural survey returned. CDU architecture is a decision that precedes engineering kickoff, not one that follows it. Our full analysis of greenfield versus retrofit CDU deployment: https://lnkd.in/gqP2nnce #DataCenterDesign #LiquidCooling #Colocation #AIInfrastructure

📣 We have some really big news... 📣 We have completed independent performance testing and validated our EcoCore Facility Cooling Distribution unit (FCD) capabilities exceed stated specifications and demonstrates readiness for Vera Rubin class AI infrastructure and the next generation of high density liquid cooled deployments.  In fact, our FCD demonstrates readiness for the 2026, 2027 and 2028 GPU roadmaps. And no, you don't have to wait months for the EcoCore FCD. In fact, typical lead times are 12-16 weeks from order to install. If you are needing future-ready liquid cooling, Nautilus EcoCore must be in your evaluation set. https://lnkd.in/gVKK9pe8

Most retrofit liquid cooling projects do not fail on density or capex. They fail on an assumption nobody examined: where the coolant water comes from. Greenfield teams specify their water source upstream. Retrofit teams inherit municipal contracts, building geometry, and decade-old facility decisions. A CDU locked to a single water pathway becomes a facility-wide constraint that governs operations, maintenance, and flexibility for the next five to ten years. Water source agnostic design is not a premium specification. It is the difference between optionality and lock-in, and it belongs in the day-one CDU evaluation, not in late-stage discovery. We break down exactly what to evaluate in a retrofit CDU spec here: https://lnkd.in/gbac9v3n #LiquidCooling #DataCenters #ColocationServices #AIInfrastructure

Looking forward to #DatacloudGlobalCongress next week. If you'll be there, be sure to stop by and see us at Booth 103. 

About one week until we head to #DatacloudGlobalCongress. If you're making the trip to Cannes, stop by Booth 103 and say hello! Robert Pfleging Kindra Martone Agustin Roca Mark Anderson Datacloud - a techoraco brand

Looking forward to #DatacloudGlobalCongress the week of 01 June. This is always one of the most exciting shows of the year. If you'll be there, be sure to stop by and see us at Booth 103.