Groq Halves Its Valuation...Shifts Focus from Competing with Nvidia to Data Centers

Valuation Drops from $6.9 Billion to $3.5 Billion
Raises $350 Million in New Funding
Pivots to Inference and Cloud Infrastructure
Chairman Amidi Sees Nvidia Deal as "Insurance"

Artificial intelligence (AI) semiconductor startup Groq has raised $350 million (approximately 490 billion won) in new funding. However, the company’s valuation has declined to $3.5 billion, roughly half of its peak last year. Since Nvidia recruited the founder and core personnel, Groq has rapidly shifted its business direction from an AI chip competitor to an operator of inference data centers.

Chip developed by Groq

According to Bloomberg News on August 17 (local time), Groq secured $350 million in funding at a $3.5 billion valuation. This round was led by investment firm Disruptive, headquartered in Dallas, Texas, with Nvidia also expected to participate as an investor.

Groq’s valuation dropped significantly over the past year. In September last year, Groq attracted a $750 million investment, reaching a $6.9 billion valuation. Its valuation, which had jumped more than twofold to $6.9 billion from $2.8 billion as of August 2024, has now dropped to around half its previous peak in this investment round.

The turning point came with a deal with Nvidia in December last year. Nvidia entered into a licensing agreement for Groq’s AI inference technology and at the same time recruited core personnel, including founder and former CEO Jonathan Ross and President Sunny Madra. Legally, Nvidia did not acquire Groq and the company remained an independent entity, but it was an unusual deal in that Nvidia secured both key talent and technology rights.

Groq was founded in 2016 by Ross, who had participated in the development of Google's AI semiconductor Tensor Processing Unit (TPU). By developing a Language Processing Unit (LPU) with a structure different from Nvidia’s Graphics Processing Unit (GPU), the company was recognized as a potential competitor to Nvidia in the AI inference market.

In particular, investor attention increased as the inference market—operating pre-trained models in real services as opposed to ‘training’ AI models—grew rapidly.

Saied Amidi, Chairman of Plug and Play, the Silicon Valley-based venture capital (VC) firm, said in an exclusive interview with The Asia Business Daily this past February that Nvidia’s deal with Groq could be seen as a kind of “insurance” taken out by Nvidia CEO Jensen Huang to prepare for future competition.

Plug and Play invested in Groq when the company’s valuation was around $2 billion. Chairman Amidi at the time said, “I believed Groq could become a big tech company even if it only secured a 3–5% share of the market,” citing Groq’s ability to perform a different type of computation from Nvidia as a key reason for the investment.

He also commented on the Nvidia-Groq deal, stating, “Nvidia likely thought it would be wise to have Groq’s solution in its portfolio in case it grows further.”

As a result, Nvidia was able to secure Groq’s core talent and technology without an outright acquisition, while Groq pivoted from being an independent AI chip competitor to focusing on data center and cloud infrastructure operations.

Groq is now focusing on expanding its inference data center business to address the growing demand for AI computing. In June, the company raised an additional $650 million to support its business transformation and reset its valuation, although specific figures were not disclosed at that time. With the recently disclosed $3.5 billion valuation, this is the first time detailed numbers have emerged since the Nvidia deal and subsequent valuation decrease.

Groq has also strengthened its cooperation with Nvidia recently. The company plans to implement Nvidia systems in its AI cloud service, GroqCloud, data centers. As a result, customers will be able to use both Groq’s LPUs and Nvidia technology within the same cloud service.

Part of the new funding will be used to expand Groq’s data centers, with plans to boost total facility capacity to more than 200 megawatts (MW) next year. Alex Davies, Groq’s Chairman of the Board, stated, “Inference will undoubtedly become the largest and most important layer in AI infrastructure,” adding that the company will focus on supporting major AI model developers.

This content was produced with the assistance of AI translation services.