Groq LPX Inference Racks Land SpaceX, Nebius as Early Nvidia Customers
The AMW Read
Converts the already-reported Nvidia-Groq IP-licensing and investment deal into named commercial customers (SpaceX, Nebius) for Groq's inference hardware, a concrete distribution proof point for the neocloud pivot rather than a new deal.
Named counterparties: Nebius Group, SpaceX
Groq LPX Inference Racks Land SpaceX, Nebius as Early Nvidia Customers
Nvidia named SpaceX and AI cloud provider Nebius as early adopters of its new Vera central processing units and the Groq LPX racks, inference-focused hardware built around Groq's chip technology. The announcement puts Groq's inference silicon into production at two well-known compute buyers, distributed through Nvidia's own hardware and customer channel rather than through Groq selling directly.
The placement extends a relationship AMW has tracked since Groq's neocloud pivot: after Nvidia's roughly $20 billion IP licensing deal and talent poach earlier this year stripped Groq of its standalone silicon differentiator, and after Nvidia moved to invest in Groq at a $3.5 billion valuation — down from $6.9 billion in 2025 — the open question was whether Groq's technology would keep commercial traction once folded into someone else's product line. Named customers at SpaceX and Nebius are an early answer: the chip lives on as an Nvidia SKU, and buyers are adopting it, but access now runs through Nvidia's sales and distribution rather than a direct Groq relationship. Groq is tracked in the AI Market Watch index with $1.75 billion in total funding under an AI Infrastructure classification (index coverage, not a full-market census).
For builders and investors, the arrangement is a real-world test of the absorption thesis: Nvidia gets a differentiated inference SKU and a marquee customer list without owning a standalone competitor, while Groq's upside now depends on deal terms with Nvidia rather than on winning enterprise accounts outright. Anyone evaluating inference-hardware options at hyperscaler-adjacent buyers like SpaceX or Nebius should treat Groq LPX as an Nvidia-distributed product, not an independent alternative to Nvidia's own stack.
