Nvidia has considerable room to run after spending billions of dollars on assets from artificial intelligence accelerator chipmaker Groq — a deal that analysts think can give its graphics-processing units an edge over competitors. Groq announced its $20 billion “non-exclusive licensing agreement” with Nvidia on Wednesday, marking the largest such deal in the semiconductor manufacturer’s 32-year history. Under the agreement, Groq CEO Jonathan Ross, president Sunny Madra and other members of Groq’s core engineering team will join the AI giant. Nvidia will also tap the startup’s inference technology, which offers high throughput and low latency for running large language models and other AI workloads. By gaining access to top talent and technology, Nvidia is positioning itself to become more competitive in the AI chip market, particularly as Google attempts to push its tensor-processing units for large-scale AI training, according to Truist analysts. “We view this as a move by NVDA to bolster its competitive positioning in inference, specifically versus the TPU,” Truist analyst William Stein said last Friday in a note to clients. He noted that, “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.” Truist has a $275 price target on shares, implying 44.3% upside from Friday’s close. Nvidia has rallied 39% this year, and it’s up nearly 6% over the past month. NVDA YTD mountain NVDA year to date Although Google isn’t selling its chips as hardware, the company is opening access to its TPUs as a service through Google Cloud, which could help it capture market share from Nvidia. Last month, The Information reported Meta was in talks to use Google’s TPUs for its data centers by 2027, causing Nvidia stock to fall 4% . But as the Silicon Valley heavyweight’s chips are increasingly posing a threat to Nvidia, the AI player is taking note, according to Paul Meeks, head of tech research at Freedom Capital Markets. “NVDA seems to be making a savvy wager to protect and maybe even widen its moat,” Meeks said Monday in a note to clients, referencing the firm’s deal with Groq. “Groq claims that its LPUs can run LLMs (Large Language Models) for inference much faster than GPUs and with 10X the power efficiency…Everyone knows that we’re entering AI’s inference stage, where computing demand explodes (as if it hasn’t already). We’re also aware that accessing power is critical,” he said. ‘Chump change’ for Nvidia Nvidia’s $20 billion deal may add to investors’ growing concerns that the firm’s dealmaking is unsustainable. But while the cost of Nvidia’s move to acquire assets from Groq is expensive on an absolute basis, it’s “chump change” for the AI player, Meeks said in his note to clients. “Although on an absolute basis the deal is expensive with NVIDIA paying $20B in cash for a non-exclusive tech license and key personnel…of a start-up that just 3 months ago was valued at only about 1/3 of what the company paid, this is chump change for NVDA,” Meeks wrote. The analysts added that the $20 billion represents just 30% of Nvidia’s gross cash and less than half of its net cash. “While the reported $20b cost is significant in absolute terms, it’s small relative to NVDA’s cash position and cash flow generation,” Meeks added.
