New AI models are designed to perform well on available, dominant hardware like NVIDIA's GPUs. This creates a self-reinforcing cycle where the incumbent hardware dictates which model architectures succeed, making it difficult for superior but incompatible chip designs to gain traction.
Related Insights
The performance gains from Nvidia's Hopper to Blackwell GPUs come from increased size and power, not efficiency. This signals a potential scaling limit, creating an opportunity for radically new hardware primitives and neural network architectures beyond today's matrix-multiplication-centric models.
While purpose-built chips (ASICs) like Google's TPU are efficient, the AI industry is still in an early, experimental phase. GPUs offer the programmability and flexibility needed to develop new algorithms, as ASICs risk being hard-coded for models that quickly become obsolete.
Large tech companies are buying up compute from smaller cloud providers not for immediate need, but as a defensive strategy. By hoarding scarce GPU capacity, they prevent competitors from accessing critical resources, effectively cornering the market and stifling innovation from rivals.
Google successfully trained its top model, Gemini 3 Pro, on its own TPUs, proving a viable alternative to NVIDIA's chips. However, because Google doesn't sell these TPUs, NVIDIA retains its monopoly pricing power over every other company in the market.
Top-tier kernels like FlashAttention are co-designed with specific hardware (e.g., H100). This tight coupling makes waiting for future GPUs an impractical strategy. The competitive edge comes from maximizing the performance of available hardware now, even if it means rewriting kernels for each new generation.
For a hyperscaler, the main benefit of designing a custom AI chip isn't necessarily superior performance, but gaining control. It allows them to escape the supply allocations dictated by NVIDIA and chart their own course, even if their chip is slightly less performant or more expensive to deploy.
Model architecture decisions directly impact inference performance. AI company Zyphra pre-selects target hardware and then chooses model parameters—such as a hidden dimension with many powers of two—to align with how GPUs split up workloads, maximizing efficiency from day one.
Google training its top model, Gemini 3 Pro, on its own TPUs demonstrates a viable alternative to NVIDIA's chips. However, because Google does not sell its TPUs, NVIDIA remains the only seller for every other company, effectively maintaining monopoly pricing power over the rest of the market.
The current AI landscape mirrors the historic Windows-Intel duopoly. OpenAI is the new Microsoft, controlling the user-facing software layer, while NVIDIA acts as the new Intel, dominating essential chip infrastructure. This parallel suggests a long-term power concentration is forming.
A key component of NVIDIA's market dominance is its status as the single largest buyer (a monopsony) for High-Bandwidth Memory (HBM), a critical part of modern GPUs. This control over a finite supply chain resource creates a major bottleneck for any potential competitor, including hyperscalers.