Understand this story
On October 23, 2026, xFusion delivered a keynote speech at the 2026 OCP Global Summit, elaborating on the three physical constraints facing AI data centers—cooling, power supply, and electrical interconnection—and their compound effects. According to a report by The Next Platform, xFusion pointed out that AI rack power has risen from 20 kW in 2023 to 240 kW in 2026, with single-chip power exceeding 2,000 watts, and the three bottlenecks stacking simultaneously. On thermal dissipation, its diamond-copper composite cold plate has a thermal conductivity exceeding 650 W/m·K, with single-chip heat dissipation capability exceeding 4,000 watts, leaving about 33% headroom for 3,000-watt-class chips. On power supply, xFusion stated that 54V power distribution at a 1 MW rack reaches 18,519 amps, and busbar copper loss can exceed 50 kW. Currently, the content of this speech is reported by a single source, with no subsequent factual developments yet.
AI-generated from public reports
Developments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 23
The Next Platform
xFusion Analyzes the Triple Physical Bottlenecks of AI Data Centers: the Compounding Effects of Cooling, Power Delivery and Interconnect
At the 2026 OCP Global Summit, xFusion pointed out that AI rack power has risen from 20 kW in 2023 to 240 kW in 2026, with single-chip power exceeding 2,000 watts, so the three bottlenecks of cooling, power delivery and electrical interconnect now compound simultaneously. Its diamond-copper composite cold plate has a thermal conductivity exceeding 650 W/m·K and a single-chip cooling capacity of over 4,000 watts, leaving about 33% headroom for 3,000-watt-class chips. xFusion says that with 54V power distribution, a 1 MW rack draws 18,519 amps, and busbar copper losses can exceed 50 kW.