In December 2025, Nvidia executed a non-exclusive licensing agreement with AI chip startup Groq, simultaneously hiring away CEO Jonathan Ross, President Sunny Madra, and key engineering talent. This transaction mirrors a broader Big Tech strategy of acquiring startup intellectual property and personnel without triggering formal antitrust acquisitions.
The Architecture of the Deal and Regulatory Maneuvering
On December 24, 2025, Nvidia announced an agreement to license chip technology from Groq and onboard its executive leadership. According to a blog post published by Groq, the transaction involves a non-exclusive license and the departure of founder Jonathan Ross—who previously helped Google start its AI chip program—alongside President Sunny Madra and other members of its engineering team.
While CNBC reported that Nvidia had agreed to acquire Groq for $20 billion in cash, neither Nvidia nor Groq commented on the report. Groq confirmed that it would continue to operate as an independent entity under new CEO Simon Edwards, maintaining its cloud business operations.
This structural arrangement mirrors recent moves across the tech sector. Microsoft secured a top AI executive via a $650 million startup deal framed as a licensing fee, Meta spent $15 billion to hire Scale AI’s CEO without a full corporate buyout, and Amazon onboarded founders from Adept AI. Analysts note that these transactions often bypass the formal Hart-Scott-Rodino Act review processes typical of major mergers.
“Antitrust would seem to be the primary risk here, though structuring the deal as a non-exclusive license may keep the fiction of competition alive (even as Groq’s leadership and, we would presume, technical talent move over to Nvidia),” wrote Bernstein analyst Stacy Rasgon in a note to clients following the announcement. Rasgon also noted that Nvidia CEO Jensen Huang’s “relationship with the Trump administration appears among the strongest of the key US tech companies.”
Hardware Realities: SRAM Versus High-Bandwidth Memory
The technical foundation of the deal lies in how Groq approaches AI computation. While Nvidia dominates the market for training large language models using Graphics Processing Units paired with external High-Bandwidth Memory (HBM), the inference market—where trained models respond to user queries—presents different engineering challenges.
Groq’s Language Processing Units utilize on-chip Static Random-Access Memory (SRAM) instead of external HBM chips. This architectural choice frees the startup from the memory crunch affecting the wider semiconductor supply chain. On-chip SRAM offers significantly faster access latency and boosts interactions with AI chatbots, though it imposes limits on the size of the model that can be served.
Following a $750 million funding round in September, Groq had previously seen its valuation spike from $2.8 billion in August of last year to $6.9 billion prior to this agreement. The startup competes in the SRAM-based accelerator space with firms like Cerebras Systems, which Reuters reported plans to go public as soon as next year. Both companies have signed large deals in the Middle East.
Market Shifts as Inference Demand Accelerates
As the artificial intelligence industry pivots from training to inference, Jensen Huang dedicated a significant portion of his primary 2025 keynote address to explaining how Nvidia plans to preserve its dominant market position. As deployment scales globally, some estimates suggest inference will eventually represent 70-80% of AI compute spending.
By absorbing Jonathan Ross—who was previously a TPU architect at Google—alongside key engineering staff, Nvidia integrates alternative inference paradigms directly into its engineering ecosystem.
