DOJ Probes Nvidia's $20 Billion Groq Deal Over Antitrust Concerns

The U.S. Department of Justice has launched an antitrust investigation into Nvidia's approximately $20 billion arrangement with AI startup Groq. Federal regulators are examining whether the deal was intentionally structured to consolidate market power while avoiding the scrutiny typically applied to traditional mergers.

At the center of the probe is a deal where Nvidia paid approximately $20 billion for a non-exclusive license to Groq's Language Processing Unit (LPU) technology and the hiring of key engineering staff. This "acquihire" model allowed Nvidia to integrate Groq's specialized talent and intellectual property without a formal acquisition. The $20 billion price tag represents roughly 2.9 times Groq's previous valuation of $6.9 billion. As part of the integration, Nvidia's LPX racks now feature 256 Groq accelerators—specifically the Groq-3/LP30—to increase inference throughput.

The Shift Toward Specialized Inference

Groq specializes in SRAM-heavy dataflow accelerators, known as LPUs, which provide deterministic latency and high token-per-second speeds for Large Language Model (LLM) inference. This technology addresses a critical performance gap where traditional GPUs often struggle. Nvidia has been aggressively expanding its ecosystem to accommodate such specialized hardware through the Open Compute Project (OCP) MGX rack designs and the NVLink Fusion licensing scheme, which allows third-party chips to integrate directly into Nvidia's proprietary networking infrastructure.

A Test for Regulatory Loopholes

This case serves as a pivotal test for how antitrust laws apply to the modern AI economy. Regulators are questioning whether the use of licensing agreements and talent acquisitions can function as legal loopholes to absorb competitors without triggering mandatory merger reviews. If the DOJ finds that the deal was a merger in all but name, it could set a precedent that limits how Big Tech firms absorb smaller innovators in the AI sector.

The Competitive Landscape

Despite the investigation, the market for high-speed inference hardware is diversifying. Competitors such as Cerebras, which has partnered with AWS and AMD, and SambaNova, which is aligned with Intel, suggest that the industry is moving toward a multi-player ecosystem. Even if the DOJ moves to unwind the Nvidia-Groq arrangement, these existing partnerships indicate that the market for specialized inference hardware is no longer dependent on a single dominant provider. It remains to be seen if the DOJ will seek structural remedies or if Nvidia's claim that the deal promotes innovation will satisfy federal investigators.