AuraTracer智迹闻
中文

EVENT DOSSIER

Why Scaling AI Compute Performance Requires a New Power Architecture

2026-08-11 23:00 Chips & Hardware 🔥 28.9 heat score
1sources
1days unfolding
28.9heat score
4mentions
SummaryAI generated

On August 11, 2026, NVIDIA stated in its official blog that as the demand for artificial intelligence computing performance continues to rise, existing power supply architectures can no longer meet these requirements. The article emphasized that to effectively expand AI computing power, it is necessary to develop entirely new power supply architecture technologies.

Related eventsRELATED EVENTS
Key entitiesKEY ENTITIES
GoogleMicrosoftNVIDIAOpen Compute Project

Coverage · reports per dayLANGUAGE SPLIT

Entity relations
Google × Microsoft1Google × NVIDIA1Google × Open Compute P…1Microsoft × NVIDIA1Microsoft × Open Comput…1NVIDIA × Open Compute P…1

SignalsSIGNALS

Keyword heat
  • NVIDIA1
  • Google1
  • Microsoft1
  • Open Compute Project1

All reports (1)SOURCES

N NVIDIA Blog en 2026-08-11 23:00

Why Scaling AI Compute Performance Requires a New Power Architecture

The new generation of accelerated computing places higher demands on infrastructure, with the bottleneck being the power transmission efficiency from the grid to GPUs. 800 VDC provides direct DC power supply, reducing the number of conversion stages and increasing available power. NVIDIA, Google, and Microsoft jointly developed this architecture and released related specifications and reference designs. More than 80 device manufacturers are producing products according to this standard. Existing AC facilities do not need to be modified; NVIDIA MGX-compatible hybrid architecture power cabinets will be delivered in the second half of 2026, allowing connection to existing systems and providing 800 VDC within the cabinets. Row-level power centers are expected to be available in 2027, capable of supplying 2 megawatts per row; new facilities will use single-step conversion DC power cabinets for large-scale direct medium-voltage conversion. Open standards have facilitated the formation of supply chains, and NVIDIA DSX reference designs provide operators with a complete blueprint from power architecture to computing facilities.