Aylin explores the various components of Supermicro's Data Center Building Block Solutions® (DCBBS), including the data center racks, in-rack and in-row CDUs, chilled doors, sidecars, and more! She also takes a deep dive into liquid-cooled solutions at the enterprise level, such as the Supermicro 4U liquid-cooled HGX™ server. Learn more about DCBBS: #DCBBS# #AI# #VeraRubin# #NVL72# #HGX#

Vera Rubin NVL72 Agentic Inference: 67x better Performance per Dollar Jensen Sandbagging Performance Again, 2x more Annual Profit Per GigaWatt, The More you Buy, The More you Earn, AgentX, InferenceX, Extreme Co-Design

$NVDA VERA RUBIN SUPPLY CHAIN Rubin is becoming a system-level buildout with rising content across memory, packaging, networking, power and cooling pushing the opportunity well beyond the GPU: • Memory sits with $MU, $SKHY & Samsung supplying the DRAM & HBM feeding Rubin accelerators while $SNDK adds exposure to storage layer around AI systems • Advanced packaging runs through $TSM, $AMKR & $INTC for foundry, packaging & assembly while $AMAT, $LRCX, $KLAC, $ASML, $TER & $AEHR provide equipment & testing needed to manufacture complex AI chips • Optical networking spans $COHR, $LITE & $AAOI in optical modules, $AVGO & $MRVL in switching silicon, $CRDO & $ALAB in high-speed connectivity & $GLW, $CIEN, $NOK & $ANET across fiber & network infrastructure • 800VDC power reaches from power semiconductors with $STM, $ON, $NVTS, $POWI & $WOLF into power management with $MPWR, $ADI, $TXN & $VICR • Power infrastructure & cooling sit downstream with $VRT, $ETN & $GEV supplying electrical distribution, backup power & thermal systems needed as rack densities keep rising • Compute systems bring full rack together through $DELL, $HPE & $SMCI which integrate Nvidia compute, networking, memory & power into deployable AI systems

TrendForce: Vera Rubin’s ASP is expected to be twice that of the GB300, while the combined production value of the GB300, VR200, and VR300 next year is projected to reach $710 billion.

NVIDIA Vera Rubin NVL72 production racks are here. The compute tray is engineered for fast compute, assembly, and serviceability. Manufacturing is 100% automated, and every tray goes together in one minute. Congratulations to @Microsoft on the first operational Vera Rubin NVL72 racks, now rolling off @HonHai_Foxconn Ingrasys lines.

The Supermicro @nvidia Vera CPU Rack is the latest addition to our DCBBS portfolio. Complementary to the now shipping NVIDIA Vera Rubin NVL72, the 100% liquid-cooled rack can be used either standalone or alongside NVIDIA Vera Rubin–powered AI factories, providing clusters with high-density CPU compute designed to address agentic AI and reinforcement learning (RL) workloads. Learn more: Watch the video update here:

NVIDIA's Vera Rubin NVL72 could train large mixture-of-experts AI models using 75% fewer GPUs than Grace Blackwell. Training a 10T-parameter model on 100T tokens in one month would just require roughly 16,000 Rubin GPUs, compared with 64,000 GB200 GPUs.

Jensen Huang says NVIDIA Vera Rubin is “in full production.” “No cables, no wires, completely 100% liquid cooled.” It takes less than five minutes to assemble a Vera Rubin node, compared with two hours for Grace Blackwell.