Nvidia Stock Slips as Groq Deal Raises Questions

Nvidia chips are the favored choice for training artificial-intelligence models. (Courtesy NVIDIA)

Key Points

• Nvidia’s stock declined Monday, influenced by broader tech sector concerns.
• Nvidia agreed to license technology from Groq, an AI chip start-up specializing in inference chips.
• Analysts expressed surprise at the reported $20 billion valuation for Groq, given its estimated revenue between $90 million and $500 million, but noted the deal’s strategic benefits.

Nvidia was dropping early on Monday. Broader jitters about the technology sector looked to be hurting the stock, while Wall Street assessed the reported $20 billion price tag put on Nvidia’s agreement to license technology from artificial intelligence chip start-up Groq.

Nvidia shares were down 1.6% at $187.45 in early trading, while the technology-heavy Nasdaq Composite lost 0.4%. Among other chip makers, Advanced Micro Devices was down 0.8% and Broadcom lost 1.2%.

The main news in the AI chip sector is Nvidia’s recent agreement to purchase a nonexclusive license for technology from privately held Groq.

The deal valued Groq at around $20 billion and will involve Nvidia hiring many of Groq’s key employees, according to reports by CNBC and The Information. Nvidia didn’t immediately respond to a request for comment on Monday.

Groq specializes in chips for inference—producing output from AI models. It calls its hardware language processing units, or LPUs, and was last valued at $6.9 billion in a $750 million September funding round.

“It is not clear what sparked the rapid, significant increase in value to Nvidia’s $20 billion price,” wrote Seaport Research analyst Jay Goldberg in a research note. “We have met with Groq in the past and are familiar with their offering, but we are not aware of any changes in their market position in recent months that would merit such an increase.”

Goldberg is an extremely rare voice on Wall Street in having a Sell rating on Nvidia stock, with a target of $140 for the price. However, he wasn’t alone in his surprise at the reported valuation of the Groq deal.

“We find it hard to believe there aren’t better assets in the same market for Nvidia to take a look at,” wrote D.A. Davidson analyst Alex Platt in a research note, arguing the limited memory capacity of Groq’s current chips means its hardware isn’t suitable for many inference tasks.

As Barron’s has previously noted, while the deal isn’t formally an acquisition and Groq will continue to exist as a stand-alone company, it looks like a potentially defensive move to stave off competition.

“The (unconfirmed) $20 billion price that Nvidia is paying is certainly large in absolute terms, and even relative to Groq’s revenue, which we have seen estimated at between $90 million and $500 million,” wrote Truist Securities analyst William Stein.

However, Stein also noted the deal represents less than half of Nvidia’s net cash and less than its expected free cash flow for the current quarter, while potentially fortifying its competitive position against the threat from Google’s Tensor Processing Units, or TPUs.

“Because Groq’s leadership formerly worked on the TPU project, we believe Groq’s LPU architecture is likely similar to that of the TPU, and designed for better latency and energy performance in large scale inference,” wrote Stein. “Nvidia’s further development of Groq’s technology could make Nvidia’s capabilities more appealing to high volume inference customers.”

Write to Adam Clark at adam.clark@barrons.com