Industry radar · curated · auto-fetched 2026-10-08
Industry radar
What is moving in liquid cooling: technology, products, investment and standards. Facts policy applies — every item links to its source; nothing is invented. The curated timeline below is the durable view; dated items rotate.
Milestones — the industry so far
| Year | Milestone |
|---|---|
| 2018 | First turnkey liquid micro DC (STULZ × CoolIT) · source ↗ |
| 2019 | ASHRAE framework for water-cooled servers — W-classes (W17-W45) established · on this site |
| 2022 | 3M announces PFAS exit (end-2025) — fluorinated fluid era gets an expiry date · source ↗ |
| 2024 | GB200 NVL72 makes 120 kW racks mainstream; OCP ORv3 blind-mate manifold Rev 1.0 lands · source ↗ |
| 2025 | 3M completes PFAS manufacturing exit — PG25 water-glycol consolidates as the DLC default · source ↗ |
| 2025 | GB300 NVL72 ships — the 140 kW-class rack (Blackwell Ultra, 288 GB HBM3e GPUs) · source ↗ |
| 2026 | SS 726:2026 (Singapore): first national standard for tropical liquid cooling · on this site |
| 2026 | Rubin generation ships: 2.3 kW GPUs, 45 °C warm water, 100% liquid racks — first deliveries Sept (Supermicro 9/23, CoreWeave 9/30) · source ↗ |
Dated items
Rubin racks move from launch to revenue workloads ~9 months after CES — the liquid-only generation is live. CoreWeave had done the industry-first bring-up/validation in June 2026.
72 Rubin GPUs + 36 Vera CPUs per rack; up to 6,000 racks/month at full production; 16-rack design = 1,152 GPUs / 331 TB HBM4. First volume deliveries of the 100%-liquid, 45 °C warm-water generation.
Covers direct-to-chip in tropical data centers; excludes standalone edge cabinets — scope matters when vendors claim compliance.
MI455X-powered Helios racks entered full production with first shipments late Q3 2026 — liquid cooling is now the default interface across both accelerator ecosystems, not just NVIDIA's.
The ramp between CES launch (Jan) and first rack shipments (Sept) — the supply chain's dress rehearsal for 2.3 kW-class GPUs.
2.3 kW-class GPUs (Max-P), 180-220 kW racks, 45 °C warm-water cooling, 100% liquid coverage with zero fans — the first AI platform generation with no air-cooled SKU at all.
Announced Dec 2022, completed end-2025. Single-fluid-class dependence is now a procurement risk — PG25 water-glycol is the default survivor.
Servers that mate on insertion, interoperable across vendors — the lock-in defense checklist item in our RFQ tool.
Announced at GTC 2024: rack-scale liquid cooling left the exotic category. The power cliff behind this guide's existence.
The announcement that redirected immersion-cooling fluid roadmaps for the entire industry.
Still the only full predecessor in the 20-50 kW turnkey cabinet space — see the gap statement in our cabinet comparison.
How this page works
- Sourcing: public announcements, standards bodies and vendor pages — each item carries its source link. We store titles, links and short notes only; full text stays with the publisher.
- Automation: 🚧 under construction until deployment — the RSS pipeline (scripts/fetch-news.mjs, daily GitHub Actions cron) activates when the site is hosted; until then this page is hand-curated. Machine items will merge in daily, deduplicated and channel-tagged; hand-curated entries are always preserved.
- Feed: machine-readable version at /newsfeed.xml.