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.

The MI300X's superior memory bandwidth and 192GB VRAM make it faster than H100s for non-FP8 dense transformers or MoE models. Quentin Anthony from Zyphra notes AMD's software has caught up, creating an under-appreciated arbitrage opportunity for teams willing to build on their stack.

The plateauing performance-per-watt of GPUs suggests that simply scaling current matrix multiplication-heavy architectures is unsustainable. This hardware limitation may necessitate research into new computational primitives and neural network designs built for large-scale distributed systems, not single devices.

The perception of LORAs as a lesser fine-tuning method is a marketing problem. Technically, for task-specific customization, they provide massive operational upside at inference time by allowing multiplexing on a single GPU and enabling per-token pricing models, a benefit often overlooked.

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.

Chinese AI models like Kimi achieve dramatic cost reductions through specific architectural choices, not just scale. Using a "mixture of experts" design, they only utilize a fraction of their total parameters for any given task, making them far more efficient to run than the "dense" models common in the West.

Fal maintains a performance edge by building a specialized just-in-time (JIT) compiler for diffusion models. This verticalized approach, inspired by PyTorch 2.0 but more focused, generates more efficient kernels than generalized tools, creating a defensible technical moat.

Instead of using high-level compilers like Triton, elite programmers design algorithms based on specific hardware properties (e.g., AMD's MI300X). This bottom-up approach ensures the code fully exploits the hardware's strengths, a level of control often lost through abstractions like Triton.

Today's transformers are optimized for matrix multiplication (MatMul) on GPUs. However, as compute scales to distributed clusters, MatMul may not be the most efficient primitive. Future AI architectures could be drastically different, built on new primitives better suited for large-scale, distributed hardware.