Groq

Groq, Inc. is a US artificial-intelligence hardware company that designs an AI-accelerator application-specific integrated circuit (ASIC) optimized for inference. The architecture was introduced as the Tensor Streaming Processor (TSP) and later rebranded the Language Processing Unit (LPU) to reflect its use serving large language models. Groq positions the LPU and its hosted service as a high-throughput, lower-latency alternative to Nvidia GPUs for inference workloads, alongside other non-Nvidia entrants such as Cerebras and specialty inference providers (Source: en.wikipedia.org).

Background

Groq was founded in 2016 by a group of former Google engineers led by Jonathan Ross, one of the designers of Google's Tensor Processing Unit (TPU), and Douglas Wightman, a former Google X engineer who served as the company's first chief executive. It is headquartered in Mountain View, California, with additional offices in San Jose, Liberty Lake (Washington), Toronto, and London (Source: en.wikipedia.org).

The company's central technical bet is a deterministic, single-core, functionally sliced microarchitecture in which memory units are interleaved with vector and matrix compute units and all execution is controlled by the compiler rather than by reactive hardware such as branch predictors, caches, and reordering buffers. Groq argues this design yields predictable latency and high energy efficiency for inference. The first-generation TSP was fabricated on a 14nm process; the second-generation LPU is being manufactured on Samsung's 4nm node at the foundry in Taylor, Texas, which took Groq's order as its first (Source: en.wikipedia.org).

Funding history

Groq received seed funding from Social Capital's Chamath Palihapitiya beginning in 2017. It raised a $300 million Series C in April 2021 led by Tiger Global Management and D1 Capital Partners, reaching unicorn status at a valuation above $1 billion, and a $640 million Series D in August 2024 led by BlackRock Private Equity Partners at a $2.8 billion valuation. In February 2025 it announced a $1.5 billion commitment from Saudi Arabia tied to a GroqCloud data center in Dammam, part of a broader Gulf push into sovereign AI infrastructure (Source: en.wikipedia.org; reuters.com).

Nvidia licensing deal (December 2025)

In December 2025, Nvidia agreed to license Groq's AI-inference technology and to transfer several senior Groq executives — including founder Jonathan Ross and president Sunny Madra — to Nvidia, in an arrangement reported at approximately $20 billion and described as Nvidia's largest deal on record. Groq characterized the agreement as non-exclusive and said it would continue operating independently (Source: reuters.com; techcrunch.com). The transaction folds a leading independent inference-silicon design into the dominant GPU supplier through licensing and talent transfer rather than acquisition, a structure that bears on AI antitrust questions about consolidation in the compute layer.

The licensed technology reached production hardware in August 2026. Nvidia said at the Hot Chips conference on August 24, 2026 that Groq 3 LPX — a dedicated inference accelerator built on the technology licensed from Groq, extending the Vera Rubin platform by offloading token generation from the GPUs handling context ingestion — had entered full production, with Nebius Group as its first committed customer (Source: siliconangle.com). Reporting of August 20, 2026 separately described a China-tailored variant of Nvidia's language processing unit, also built with Groq-licensed technology, planned for small-batch shipment by the end of 2026 (Source: theinformation.com).

Justice Department investigation

A Justice Department investigation into whether Nvidia structured the licensing arrangement to avoid antitrust scrutiny became public on September 9, 2026. The department opened it shortly after the arrangement was announced in December and has sent Nvidia a formal request for information, according to two people familiar with the matter. An Nvidia spokesperson said "the Groq story is a prime example of the American system working as designed to promote innovation, reward entrepreneurs, and benefit consumers." The department could fine Nvidia if it finds the deal was mishandled but is unlikely to seek to unwind the transaction (Source: reuters.com). That report gives the deal's value as $17 billion, against the approximately $20 billion reported at announcement; the divergence is unresolved. The investigation tests the structure directly — a non-exclusive licence plus executive transfers, rather than an acquisition — and is the first regulatory action recorded against it. See AI Antitrust.

2026 funding rounds and pivot to inference cloud

Groq confirmed a $650 million funding round on June 22, 2026, led by Disruptive and Infinitum and structured as a pro-rata offering backstopped by existing investors, roughly six months after the Nvidia deal removed its founder and president. Now led by chief executive Doug Wightman, the company said it is concentrating on its "neocloud" inference-cloud business — reported at 13 data centers serving more than five million developers — and named Alan Rice (formerly of xAI) as chief operating officer and Sinclair Schuller as chief technology officer (Source: techcrunch.com; axios.com).

A further round followed within two months: on August 17, 2026 Groq announced it had raised $350 million at a $3.5 billion valuation, again led by Dallas-based Disruptive, with Nvidia also set to invest according to a Groq representative. The valuation is roughly half the $6.9 billion Groq reached in September 2025, before the Nvidia licensing deal and the departure of Jonathan Ross and other key employees. The company plans to expand its data-center capacity past 200 megawatts in 2027 (Source: bloomberg.com).

Policy relevance

Groq is one of several companies whose existence shapes the inference-economics debate over whether non-Nvidia silicon can become a credible procurement option for frontier labs and enterprises, a question central to the AI-bubble debate and to the durability of Nvidia's position in the AI compute supply chain. Its Saudi commitment situates it within the Gulf-states build-out of AI capacity, and the Nvidia licensing deal makes it a reference point for how compute-layer consolidation is structured.

Relationships

• related: Nvidia & TSMC — AI Compute Infrastructure, Cerebras Systems, Specialty Inference Hardware — Cerebras, Groq, SambaNova, Inference Economics and Token Pricing, AI Bubble Debate, Semiconductor Supply Chain, AI Antitrust, Sovereign AI.