TL;DR
Nvidia paid $20 billion for Groq's intellectual property and hired its executive team while explicitly not buying the operating company. The deal was structured as a non-exclusive license plus talent acquisition, leaving GroqCloud independent and avoiding traditional regulatory reviews.
What happened
On December 24, 2025, Nvidia agreed to a $20 billion transaction that transferred Groq's patents and other intellectual property to Nvidia and brought Groq's CEO Jonathan Ross and the senior leadership team into Nvidia. The deal is framed as a non-exclusive license to Groq's inference technology rather than a corporate acquisition; Nvidia's CEO said the company was adding talent and licensing IP but not buying Groq itself. Crucially, GroqCloud — the cloud infrastructure arm and holder of lucrative contracts, including a $1.5 billion Saudi commitment tied to a Dammam facility — was left independent under CFO Simon Edwards. Groq had raised $750 million at a $6.9 billion valuation just months earlier, making Nvidia's payment roughly $13.1 billion above that post-money figure. The arrangement avoids the typical acquisition triggers cited by regulators: CFIUS review, extended antitrust scrutiny and shareholder votes, while transferring the core technology and team to Nvidia.
Why it matters
- Regulatory precedent: structuring IP-and-talent transfers as non-acquisitions can reduce or bypass CFIUS and antitrust processes.
- Market consolidation: Nvidia acquired Groq's core technology and leadership while leaving the operating cloud business independent, shifting competitive dynamics in inference hardware.
- Inference economics: Groq's SRAM-centric LPU design targets low-latency, high-throughput inference for mid-sized models, a growing commercial workload.
- Geopolitical risk management: carving out GroqCloud avoids inheriting foreign government contracts that could trigger national security reviews.
- Competitive positioning: buying the IP and people removes a potential alternative supplier or partner for other large cloud and chip vendors.
Key facts
- Nvidia paid $20 billion for Groq intellectual property and patents and licensed Groq's inference technology under a non-exclusive agreement.
- Jonathan Ross (CEO), Sunny Madra (President) and Groq's senior leadership team joined Nvidia as part of the transaction.
- Nvidia explicitly did not acquire GroqCloud; GroqCloud remains an independent entity under CFO Simon Edwards.
- Groq had raised $750 million in September 2025 at a post-money valuation of $6.9 billion; Nvidia's deal implies a $13.1 billion premium over that valuation.
- Groq's LPU architecture centers on large on-chip SRAM instead of external DRAM/HBM, with quoted per-chip specs of ~80 TB/s bandwidth and ~230 MB capacity.
- Performance claims in the source: Llama 2 7B at ~750 tokens/sec (2,048 context), Llama 2 70B at ~300 tokens/sec (4,096 context), Mixtral 8x7B at ~480 tokens/sec (4,096 context).
- LPUs are described as inference-only devices with deterministic execution and limited model-size capacity (source notes a 14GB per-rack constraint preventing very large models).
- Groq operated 13 facilities across multiple regions and built a large Dammam inference cluster in eight days; the Saudi commitment referenced in the source was $1.5 billion.
- Nvidia's previous largest acquisition was Mellanox for $7 billion in 2019; the Groq transaction is Nvidia's largest to date, per the source.
- The non-acquisition structure avoids typical triggers for CFIUS review, extended antitrust scrutiny, shareholder voting and lengthy regulatory disclosure.
What to watch next
- not confirmed in the source: whether regulators (CFIUS, DOJ, FTC) will reclassify or challenge the deal structure after the fact.
- not confirmed in the source: whether GroqCloud will remain viable as an independent business without the IP and leadership team.
- not confirmed in the source: if Nvidia will integrate LPU technology into future Blackwell-era products or adopt a chiplet integration strategy.
- not confirmed in the source: whether other cloud or chip companies will license Groq IP or hire displaced Groq engineers.
Quick glossary
- LPU (Language Processing Unit): A specialized inference accelerator design that emphasizes on-chip SRAM to host model weights and activations, aiming for low-latency, deterministic execution.
- SRAM: Static RAM, a type of on-chip memory with lower latency than external DRAM/HBM but higher area and cost per bit; used for fast local storage.
- DRAM/HBM: Off-chip memory technologies (dynamic RAM and high-bandwidth memory) commonly used by GPUs and TPUs to store model weights and activations during compute.
- Inference: Running a trained machine learning model to generate outputs (for example, responding to prompts); typically prioritizes low latency and cost efficiency in production.
- CFIUS: The U.S. Committee on Foreign Investment, which reviews foreign investments in U.S. companies for potential national security risks.
Reader FAQ
Did Nvidia buy Groq as a company?
No. According to the source, Nvidia licensed Groq's IP and hired its leadership but did not acquire GroqCloud or the corporate entity.
Is GroqCloud part of the deal?
No. The source states GroqCloud was left independent under CFO Simon Edwards and was not purchased by Nvidia.
Why did Nvidia structure the deal as a non-exclusive license?
The source describes the structure as a way to avoid lengthy CFIUS and antitrust reviews, shareholder votes and certain disclosures.
Will LPUs replace GPUs for all AI workloads?
Not confirmed in the source. The source notes LPUs are optimized for inference and have model-size limitations and cannot train models.
"You're taking on a giant. What gives you the audacity?" On November 5th, 2025, Groq CEO Jonathan Ross was asked why he was even bothering to challenge Nvidia. He didn't…
Sources
- Nvidia's $20B antitrust loophole
- Nvidia's $20B Groq Deal: Strategy, LPU Tech & Antitrust
- NVIDIA's $20 Billion Groq Acquisition and the Dawn of …
- Nvidia's Strategic Expansion in AI Inference: What the Groq …
Related posts
- Six prominent creators leave FaZe Clan as new management reshapes the group
- Nvidia’s $20B Groq Deal: IP and Team Acquired, Company Left Intact
- OrangePi 6 Plus review — 12-core ARM64 SBC with high-end IO and NPU