NVIDIA’s projection of generating over $1 trillion in revenue from its Blackwell and Rubin chips between 2025 and 2027 is no longer mere market exuberance—it has become the central anchor for global AI infrastructure investment. Its latest quarter reported $81.6 billion in revenue, a staggering 85% year-over-year increase, with data center sales accounting for more than 90% of total revenue. Behind these figures lies near-insatiable demand from hyperscalers, sovereign AI funds, and enterprise clients. Yet despite this dominance, NVIDIA’s stock has shown signs of fatigue—not due to weakening fundamentals, but because investors are increasingly wary of valuation overextension, supply chain concentration, and the fragility of a single-technology trajectory.
Meanwhile, AMD is quietly executing an asymmetric counterstrategy. Rather than directly challenging NVIDIA’s H100 or B200 in general-purpose AI training, AMD is focusing on inference optimization, custom accelerators, and an open software ecosystem. While the MI300X still lags behind Blackwell in raw performance, its improvements in power efficiency, memory bandwidth integration (notably through Infinity Fabric paired with HBM3e), and ROCm compatibility have already attracted Meta, Microsoft, and others to deploy AMD solutions for specific AI workloads. Crucially, AMD is leveraging TSMC’s capacity flexibility at the 3nm node and beyond to secure more agile wafer allocation—potentially sidestepping delivery bottlenecks that could arise from NVIDIA’s overwhelming order concentration.
Analysts at The Motley Fool repeatedly asking “Is NVIDIA stock a buy?” reflect a growing market schism: undeniable earnings momentum on one side, and skepticism toward the “sole winner” narrative on the other. This skepticism is not unfounded. The AI chip market is shifting from a unipolar explosion to a multi-track evolution. Training remains NVIDIA’s stronghold, but inference, edge AI, domain-specific models (e.g., healthcare, finance), and sovereign compute initiatives—particularly in Europe and the Middle East—are increasingly adopting hybrid or multi-vendor architectures. AMD’s full-stack capability (CPU + GPU + interconnect) gives it structural advantages in these niches.
Geopolitical dynamics further amplify this divergence. Ongoing U.S. export controls on high-end AI chips compel non-U.S. customers to seek alternatives. Although AMD’s MI300 series is also restricted, its product portfolio offers greater configurability—such as tuning HBM capacity or interconnect bandwidth—to create “compliant” variants that meet regulatory thresholds. This technical modularity enhances AMD’s negotiating leverage in government-backed AI projects across South Korea, Japan, and parts of Europe. NVIDIA’s solutions, by contrast, are highly integrated and less adaptable.
A deeper challenge lies in capital efficiency. NVIDIA’s ~75% gross margin depends on sustained process leadership and CUDA’s software moat. Any strain on TSMC’s 3nm/2nm capacity or meaningful erosion of CUDA’s dominance by open frameworks (e.g., MLX, Triton) could undermine its pricing power. AMD, meanwhile, pursues a “performance-cost-openness” triangle: not chasing peak specs, but delivering sufficient performance-per-dollar and developer freedom. As AI investment shifts from frenzy to rationalization, this approach may prove more resilient.
I judge that over the next 18 months, the AI chip market will settle into a “one-superpower, multiple-strong” configuration—but the definition of “superpower” is evolving. It’s no longer just about raw compute dominance, but about ecosystem control, supply chain resilience, and geopolitical adaptability. NVIDIA will retain leadership in high-end training, but AMD could capture 15–20% share in inference, customization, and sovereign projects—far exceeding its current ~5%. This shift won’t dethrone NVIDIA, but it will recalibrate investor expectations of perpetual hypergrowth.
The real question may not be whether NVIDIA remains a good stock, but whether the AI compute dividend is transitioning from hardware exclusivity to a shared value between software and silicon. As Meta, Google, and even Tesla develop in-house AI chips, and as open-source models reduce reliance on proprietary hardware, chipmakers’ value proposition must shift from “selling FLOPS” to “empowering developers.” In that deeper contest, whoever wins developer mindshare holds the ticket to the next era.