Nvidia Licenses Groq’s AI Chip Tech and Hires Founder Jonathan Ross in $20B Deal

In a landmark move shaking up the AI chip industry, Nvidia has entered a $20 billion non-exclusive licensing agreement with Groq for its high-speed inference technology, while hiring founder Jonathan Ross and key team members like President Sunny Madra.[1][2][3] Announced on December 24, 2025, this “acqui-hire” deal—Nvidia’s largest ever—allows the chip giant to bolster its inference capabilities without fully acquiring Groq, which will operate independently under new CEO Simon Edwards.[1][3]

The Deal Breakdown: Licensing, Talent, and Assets

This isn’t a traditional merger. Nvidia is licensing Groq’s Language Processing Unit (LPU) architecture—a 144-way VLIW design built at GlobalFoundries since 2019—known for delivering blazing-fast single-user inference speeds with low latency and energy efficiency thanks to embedded SRAM memory.[1][2][3] Unlike GPUs, which excel in training massive AI models, Groq’s LPUs shine in inference, the phase where models generate real-time responses to user queries.[3]

Key elements include:
– $20 billion valuation: Surpassing Nvidia’s prior record $7 billion Mellanox buyout in 2019, the deal covers licensing rights, talent acquisition, and Groq’s physical assets—but not its IP or full company.[2][3]
– Talent exodus: Jonathan Ross, ex-Google TPU pioneer, joins Nvidia to integrate the tech into its “AI factory architecture,” alongside Madra and other executives.[1][2][3]
– Groq’s continuity: The startup retains independence, with GroqCloud serving over 2 million developers uninterrupted, including Middle East deployments of 19,000 chips backed by a $1.5 billion Saudi commitment.[1][2][3]

Nvidia CEO Jensen Huang emphasized synergies: “We are adding talented employees… and licensing Groq’s IP” to extend platforms for broader AI inference workloads.[3] Groq will keep servicing its neocloud business and regional deals.[2]

Why Groq’s Tech Challenges Nvidia—and Why It Matters

Groq emerged as a Nvidia rival in inference, touting top token-per-second rates for single users via its only-local-memory architecture.[2] Founded by Ross from Google’s original TPU team, Groq raised $1.8 billion, hitting a $6.9 billion valuation post-$750 million Series E in 2025.[2] Its first-gen chip, cheap to scale, powered rapid growth from 356,000 to 2 million developers yearly.[3]

Nvidia dominates training but faces stiffer inference competition from players like Cerebras, which eyes an IPO.[3] Groq’s edge: faster deployment, lower energy use, and manufacturing wins amid resolved substrate shortages (now 12-week deliveries).[3] A planned second-gen SF4X chip at Samsung Foundry was slated for 2025 revenue, but this deal accelerates Nvidia’s access.[2]

This table highlights complementary tech, positioning the licensing as a strategic win.[2][3]

Broader AI Chip Market Boom

The deal underscores explosive growth: US AI chip startups snagged $5.1 billion VC in H1 2025; the market hit $20 billion this year, projected to $52 billion by year-end and $311 billion by 2029.[3] Hybrid “acqui-hires” like Meta’s $15B Scale AI and Microsoft’s $650M Inflection deals dodge regulations while grabbing talent and IP fast.[3]

Middle East plays intensify rivalry—Groq and Cerebras inked massive contracts there.[3] Nvidia’s Rubin CPX chip was teased earlier in 2025, but Groq tech integration could supercharge inference in its stack.[2]

Competitive Shifts and Future Outlook

For Nvidia, this fortifies inference defenses, blending Groq’s low-latency processors with its training prowess for end-to-end AI factories.[3] Ross’s LinkedIn post confirms he’ll drive integration while GroqCloud hums along.[3]

Groq benefits too: IP retention, new leadership (Edwards from finance), and Nvidia ecosystem ties preserve its edge in niche inference.[1][3] No full acquisition means agility for upcoming chips and deals.

Risks linger—integration hiccups or regulatory scrutiny—but the structure minimizes them.[3] This signals consolidation in a cutthroat arena where inference lags training in maturity.

What It Means for AI Innovation

Enterprises gain: faster, cheaper inference scales global AI without Nvidia lock-in.[1] Developers keep GroqCloud perks, now potentially Nvidia-enhanced.[1]

Ultimately, the Nvidia-Groq pact accelerates AI at global scale, marrying incumbency with upstart speed. As Ross jumps ship, expect inference benchmarks to leap, fueling the next wave of real-time apps from chatbots to autonomous systems.[2][3]

(Word count: 812)

Original source: TechCrunch – Nvidia to license AI chip challenger Groq’s tech and hire its CEO