Groq, an AI infrastructure startup, has raised $350 million in Series A funding to expand its GroqCloud platform, which offers customers a specialized chip-and-cloud alternative for running AI models.
The Groq 3 LPU, developed alongside Nvidia, handles specific parts of language model computation more efficiently than traditional GPU-only setups, allowing Groq to market a disaggregated approach where feed-forward calculations run on Groq's chip while attention mechanisms use Nvidia GPUs.
The company plans to grow its global data center footprint and power capacity significantly next year.
What happened
AI infrastructure startup Groq announced $350 million in Series A funding led by returning backer Disruptive, with Nvidia planning to join the round at an unspecified amount. This comes less than three months after Groq closed a $650 million round.
Why it matters
Groq operates GroqCloud, an AI-optimized public cloud platform powered by the Groq 3 LPU chip (co-developed with Nvidia). The chip is designed to handle inference workloads more efficiently by offloading feed-forward network calculations from Nvidia's GPUs, giving enterprises an alternative approach to running large language models in production.
What to watch
Groq plans to scale its cloud from 13 data centers worldwide and expand power capacity from 57 megawatts to more than 200 megawatts next year, using the funding to grow GroqCloud availability and capacity.
Ask the AI about this article →
Groq's $350 million Series A represents a significant bet on a disaggregated approach to AI inference. The company's pivot from hardware design to cloud infrastructure became possible after Nvidia licensed its core chip technology in a $20 billion deal last December, which also brought several Groq executives into Nvidia's ranks. Rather than compete head-to-head with Nvidia, Groq has positioned itself as a complementary service: it operates GroqCloud using racks of Groq 3 LPU chips paired with Nvidia Rubin GPUs, optimizing inference workloads by splitting computation between the two. The Groq 3 LPU excels at feed-forward network tasks—the components that store much of an LLM's learned knowledge and perform large portions of the generation work—while Nvidia's GPUs handle attention mechanisms. This split approach can also accelerate models with mixture-of-expert architecture or speculative decoding, broadening its appeal to enterprises seeking inference optimization.
The funding momentum—$650 million in one round followed by $350 million in the next within three months—signals market confidence in the infrastructure play. Nvidia's participation in the Series A, though at an unspecified amount, also underscores the strategic alignment: Groq's cloud platform drives demand for the very Rubin GPUs that Nvidia manufactures. Groq's expansion roadmap, scaling from 57 to over 200 megawatts of power capacity next year while growing from 13 to more data centers, shows the company is building toward meaningful scale in what remains a nascent market segment of specialized inference infrastructure.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Astromech, an AI startup co-founded by Ben Lamm and geneticist George Church, raised $20 million in funding le…
Alibaba is guiding its AI cloud revenue toward a US$10 billion run-rate in the next quarter, signaling that it…
Ode, a venture arm of Anthropic, has acquired Casper Studios to expand its enterprise artificial intelligence…
SK hynix, the world's largest supplier of high-bandwidth memory (HBM), has published a technical roadmap for c…
Alibaba Group's June quarter results show cloud AI revenue growing 45%, while capital expenditure (investment…
Anthropic launched Claude Academy on August 20, a free learning site that explains AI fundamentals and how to…
