Despite continued expansion in capital expenditure within the artificial intelligence sector, NVIDIA's stock performance has cooled. Since the beginning of the fourth quarter, the AI chip giant has risen only about 1%, with a current price-to-earnings ratio of approximately 24 times, roughly in line with the Nasdaq 100 Index, indicating the market is reassessing its valuation premium.
Changes in the competitive landscape are a core driver of the cautious sentiment. NVIDIA CEO Jensen Huang's acquisition this month of a technology license from inference hardware startup Groq for around $20 billion, along with recruiting most of its chip team, itself confirms that other companies possess competitiveness in specific areas. Simultaneously, Cerebras signed a $10 billion agreement with OpenAI for rapid inference chip supply, while Anthropic has partnered with several non-NVIDIA chip suppliers.
These deals are reshaping market perceptions of the AI chip landscape. Multiple startups report a noticeable increase in interest from potential investors since the Groq transaction. SambaNova even abandoned discussions about selling the company at a valuation significantly below its previous round, opting instead to seek a new round of financing.
For investors, this series of signals implies that while NVIDIA remains the undisputed leader in the AI chip space, its monopoly position may no longer be as unshakeable as in the past. The market is shifting from "betting on a single leader" to "repricing competitive risks."
**Inference Chip Market Becomes Focal Point of Competition** In the AI chip arena, an increasing number of startups and investors are turning their attention to "inference"—the critical phase of running a model and generating answers after training is complete. This segment is seen as a potential breakthrough point for challenging NVIDIA's dominance.
Trading firm Jump co-led a $230 million funding round for inference chip startup Positron this month and has become one of its customers. The company's Chief Technology Officer, Alex Davies, stated bluntly: "Almost everyone uses NVIDIA for both training and inference today, but we see the industry changing; this situation won't last. We don't believe there will be only one winner."
NVIDIA dominates large-scale parallel training computations thanks to its high-bandwidth memory chips. However, a cohort of startups is attempting to achieve faster response times in inference scenarios by exploring different types of memory architectures. Meanwhile, the line between training and inference is blurring as inference AI models make real-time judgments when queried, rather than relying solely on pre-trained results, creating opportunities for novel chip architectures.
Sid Sheth, CEO of Microsoft-backed AI chip company D-Matrix, noted that interest in fast inference chips has significantly intensified since DeepSeek's debut early last year. His company completed a $275 million funding round in November.
**Tech Giants Accelerate In-House Chip Development** Major technology companies are racing to develop their own AI chips to reduce dependence on NVIDIA. OpenAI recently released a model running on Cerebras chips for the first time; Anthropic has secured usage agreements for Amazon's Trainium and Google's TPUs; Microsoft last month unveiled its second-generation in-house AI chip, Maia, and secured rights to use OpenAI's chip IP.
Startups are also actively positioning themselves. Inference chip company Etched raised approximately $500 million last month, directly targeting NVIDIA's dominance. AI model startup Simile emerged from stealth mode with a $100 million funding round led by Index Ventures, focusing on helping companies predict human behavior.
Nevertheless, despite the accelerated push for in-house development, companies like Amazon, Google, Microsoft, and OpenAI continue to purchase large quantities of NVIDIA GPUs to support their AI products and cloud services. This reality underscores that NVIDIA's position as the market leader remains solid, even as the competitive landscape quietly evolves.
**NVIDIA's Defenses and Market Outlook** NVIDIA has proven to be a formidable market leader. The company boasts a diverse product lineup and commits to a complete chip redesign annually. The Groq deal provides NVIDIA with an opportunity for further expansion. When asked if the agreement might lead to a new chip specialized for inference, Huang did not commit, only suggesting that "maybe somewhere we might create something unique."
Sheth anticipates NVIDIA will announce measures at its flagship conference in March to address demand for fast inference chips. According to Bloomberg, at various times, both startups and established companies have claimed they can compete with NVIDIA, but most have failed to do so, at least not at scale or comprehensively. However, cracks are beginning to appear in the market.
Davies commented: "If you look at the growth rate of this industry, you see specialized hardware. That's the pattern throughout engineering history. You start with something general-purpose, it grows wildly, and then someone discovers you simply can't have just one thing."
Comments