NVIDIA's Groq Racks Deploy as AI Inference Race Accelerates
NVIDIA is shipping hardware from its $20 billion Groq acquisition, and the first deployment will arrive at a single cloud provider later this year. The Groq 3 LPX rack delivers 3,400 tokens per second in inference workloads.
That is more than 4x faster than OpenAI's Cerebras-powered solution. This gives NVIDIA a quantified speed advantage in the race to monetize low-latency AI inference.
The financial impact will be negligible in the near term. Only one customer, neocloud Nebius, is publicly committed to deploying the rack. But the move extends NVIDIA's reach across the AI compute stack.
It adds specialized inference capability to NVIDIA's dominant training franchise. This vertical integration…
Keep reading with a free account
Get tomorrow's institutional market briefing before markets open, produced by specialized AI analysts researching markets 24/7. Takes under a minute, no card required.
- groupsSpecialized AI analysts across ten research desks, working around the clock
- plagiarismAnalysis built from primary sources: SEC filings, Fed statements, and CFTC positioning data, not recycled headlines
- event_noteScheduled-event coverage: FOMC decisions, CPI prints, and earnings, written as they land
- monitoringCharts rendered from real market data, never stock imagery or invented numbers
Already have one? Log in · Want everything the day it publishes? Go Premium
Secure checkout via Stripe