Nvidia says its inference accelerator Groq 3 LPX has entered full production and Nebius has signed on as the first customer, and SpaceX will deploy Vera CPUs
Chipmaker Nvidia Corp. says its dedicated artificial intelligence inference accelerator Groq 3LPX has now entered full production …
The customer list also extends the Vera story: Huang named Anthropic, OpenAI, and SpaceX as first big users of the Vera CPUs back in June, and SpaceXAI's adoption now adds another marquee account. Meanwhile Groq itself still operates independently, and bids for its GroqCloud platform were expected to exceed $1B after the license.
First-order effects
Nebius becomes the launch customer for an inference-only rack whose value proposition rests entirely on SRAM bandwidth rather than GPU generality, forcing it to position the LPX alongside — not instead of — its existing accelerator inventory.
Nvidia converts the Groq license into shipping revenue while Groq remains an independent competitor selling through its own GroqCloud channel, putting the two firms in direct competition using shared technology lineage.
Second-order effects
Rival neoclouds and hyperscalers now face a benchmark question: if Nebius's LPX deployments show better inference latency-per-dollar, competitors must either bid for their own allocations or lean harder on Groq's independent offering.
The expected $1B-plus bidding for GroqCloud gets repriced by this news — a fully produced Nvidia-branded LPX validates the technology's market readiness while simultaneously crowding the buyer's strategic options.
Third-order effects
If the LPX sells at rack scale, AI compute splits structurally into training GPUs and purpose-built inference systems, with value capture migrating toward whoever owns the low-latency serving stack — a shift the licensing-plus-talent-acquisition playbook was designed to pre-empt.
The pattern of incumbents licensing startups' architectures and absorbing their founders, rather than acquiring them outright, points to a consolidation model where inference silicon leadership concentrates even as nominally independent challengers survive.
The trend: Inference is separating from training as its own hardware market, with incumbents using licensing-and-talent deals to absorb startup architectures before they can scale independently.
The first CPU built for agents is now at work at massive scale. @SpaceXAI is deploying NVIDIA Vera to power agentic AI — faster agents, GPUs fully utilized, and a single architecture from Earth to orbit. Read more ⬇️ https://nvidianews.nvidia.com/ ...
NEWS: NVIDIA Groq 3 LPX is now in full production. NVIDIA Vera Rubin NVL72 is the foundation of every AI factory. Paired with Groq 3 LPX, it unlocks faster, smarter agents and breakthrough user experiences. Through extreme co-design across seven chips and five purpose-built
NVIDIA's embargo lifted. Three calls we dated shipped today: Groq LPX, Spectrum X at 128k GPUs a rail, Vera CPUs at SpaceX While the circular-financing crowd argued, $NVDA raised next year's prices 15% The Audit grades our record before Wednesday https://bepresearch.substack.com/…
The next wave of AI is about deploying incredible innovation at massive scale. Excited to work with @GroqInc and @nvidia to make that happen with @Dell. 🚀 https://x.com/...
HotChips warm up: NVIDIA just posted the slide. Groq 3 LPX. 3,431 tokens a second. 4x the next public endpoint. Gemma 4 31B at 100,000-token context. In full production. Faster than Cerebras 😮🔥 $NVDA
We are thrilled to announce that Groq will be among the first adopters of NVIDIA Groq 3 LPX, deploying it alongside NVIDIA Vera Rubin NVL72 in our purpose-built AI inference Cloud. Groq is working with Dell Technologies to deploy NVIDIA Groq 3 LPX. When Groq brings NVIDIA Groq 3
you read that right, Nebius is the first AI cloud to put NVIDIA Groq 3 LPX into production on @nebiustf Today NVIDIA also confirmed LPX is in full production, the first fruit of the Groq licensing deal from last year. Over the last year NVIDIA stopped just selling accelerators
SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028
NVIDIA's talk about Vera CPUs at Hot Chips is a series of chart murders like the one below, which is disappointing. The subtext for comparisons is important. “Comparison based on relative performance for individual SPECrate®2026_int_base benchmark ratios between an NVIDIA Vera
How potent are Nvidia's Groq-3 LPX racks? Very, but Gemma 4 31B tells an idealized story. The real trick will be how efficiency it scales to MoE. — My full 1,200 word analysis only @theregister.com — www.theregister.com/systems/ 2026...