Shares of NVIDIA (NVDA) are under pressure again this week, falling roughly 2% ahead of the company’s highly anticipated Feb 25 earnings report, while rival AMD continues to climb. NVIDIA Corporation (NVDA) closed at $182.81, down $4.13 (-2.21%) at 4:00 PM EST. Advanced Micro Devices, Inc. (AMD) closed at $207.32, up $1.38. The two stocks moved in opposite directions on Friday after Arista Networks’ CEO told investors that the company is observing a shift in some customer deployments toward AMD-based solutions.
At the center of the split is a broader narrative pivot in artificial intelligence investing, from AI training dominance to the “Second Wave of AI,” where inference efficiency and cost control are taking priority.
From AI Builder to AI Maintainer
For the past two years, NVIDIA (NVDA) has been the undisputed market leader, riding demand for its AI training chips to a market capitalization near $4.5 trillion. Its next-generation Blackwell and Rubin architectures remain in high demand, but many analysts argue that those expectations are already “priced in.”
While NVIDIA’s CUDA software ecosystem continues to serve as a powerful moat, preventing easy switching by enterprise customers, investors are beginning to question how much incremental upside remains in the stock at current levels. NVIDIA trades at roughly 25x forward earnings, cheaper than AMD’s 40x multiple, yet sentiment analysis shows that traders perceive AMD as having higher “velocity.”
That perception shift has intensified ahead of earnings, what some traders call the “lull before the storm.” Short-term weakness appears tied to portfolio rotation and tax-driven positioning, rather than deteriorating fundamentals.
AMD Emerges as the “Value Challenger”
Meanwhile, AMD has positioned itself as the execution-focused alternative. CEO Lisa Su has emphasized roadmap delivery over spectacle, spotlighting the MI355X and upcoming MI400 accelerators as cost-efficient solutions for AI inference workloads.
The shift toward inference, running trained models at scale, is reshaping capital allocation across data centers. Inference chips are optimized for efficiency rather than brute-force training power, a dynamic that plays directly into AMD’s value proposition.
A key inflection point came during CES 2026, where AMD’s detailed data center roadmap contrasted with NVIDIA CEO Jensen Huang’s forward-looking vision centered on humanoid robots and autonomous AI systems. While Huang’s long-term ambitions captivated headlines, some institutional investors appeared to gravitate toward nearer-term execution metrics.
Further momentum for AMD followed comments from Arista Networks, which indicated that roughly 20% to 25% of recent AI cluster deployments now incorporate AMD hardware, a signal that hyperscale customers are diversifying away from a single-vendor strategy.
Arista Networks, Inc. (NYSE: ANET) reported its fourth-quarter and full-year financial results for the period ending December 31, 2025, posting annual revenue of $9.006 billion, a 28.6% increase compared with fiscal 2024.
Jayshree Ullal, Chairperson and CEO of Arista Networks, stated, “2025 was the year of validation of our Arista 2.0 momentum, as we hit the milestone of shipping a cumulative of 150 million ports”.
Diversification Pressures Build
The diversification theme extends beyond networking partners. Reports suggest that OpenAI is broadening its hardware stack to reduce reliance on NVIDIA GPUs. While NVIDIA remains deeply embedded in large-scale training environments, customers increasingly seek multi-vendor flexibility to manage costs and supply risks.
Export policy remains another wildcard. The U.S. Dept of Commerce continues to oversee semiconductor export restrictions to China. Any easing of chip bans could unlock fresh demand for NVIDIA’s premium training chips, introducing significant volatility into the stock.
The Second Wave of AI
The deeper story, however, may be structural rather than cyclical.
The first AI wave rewarded companies that built the infrastructure to train large language models. NVIDIA dominated that era. The second wave rewards those who can deploy AI models economically at scale. Inference efficiency, memory optimization, and power consumption are now the new battlegrounds.
AMD is increasingly perceived as the cost-effective “king of inference,” benefiting from enterprises seeking performance per dollar rather than maximum raw throughput.
This does not mean NVIDIA’s competitive edge has evaporated. CUDA remains a powerful lock-in mechanism, and demand for Blackwell-based systems remains robust. But as AI workloads mature, buyers are scrutinizing the total cost of ownership more closely.
Earnings as a Catalyst
All eyes now turn to Feb. 25, when NVIDIA reports earnings. Investors will be listening for commentary on inference demand, Blackwell supply constraints, and customer diversification trends.
If NVIDIA delivers another blowout quarter, yet the stock remains muted, it could reinforce the current rotation into perceived laggards like AMD. Conversely, stronger-than-expected forward guidance, particularly on inference revenue, could quickly reverse the narrative.
