The tech industry wrongly compares AI to software, which has near-zero marginal costs for new users. In reality, providing access to frontier AI models is a zero-sum game during compute crunches because of immense computational requirements. Servicing another user is expensive, leading to rationed access.

The enormous compute budget for the original AlphaGo was not about finding the most efficient training method, but about proving a method could work at all. Once a breakthrough is made and the path is clear, subsequent efforts can focus on optimization and achieve similar results with far less compute.

With frontier models costing over 100x more than competent alternatives ($56 vs. 50¢ per million tokens), companies are burning cash. An estimated 98% of tasks sent to top-tier models don't require that power, an inefficiency driven by engineers who are disconnected from cost implications.

Unlike previous years where the path forward was simply scaling models, leading AI labs now lack a clear vision for the next major breakthrough. This uncertainty, coupled with data limitations, is pushing the industry away from scaling and back toward fundamental, exploratory R&D.

Scientists constrained by limited grant funding often avoid risky but groundbreaking hypotheses. AI can change this by computationally generating and testing high-risk ideas, de-risking them enough for scientists to confidently pursue ambitious "home runs" that could transform their fields.

A "software-only singularity," where AI recursively improves itself, is unlikely. Progress is fundamentally tied to large-scale, costly physical experiments (i.e., compute). The massive spending on experimental compute over pure researcher salaries indicates that physical experimentation, not just algorithms, remains the primary driver of breakthroughs.

OpenAI spent an estimated $15-20M to solve the Navier-Stokes problem, which had a $1M prize. This demonstrates a new paradigm of using immense compute to solve complex problems, a method out of reach for most academic or commercial entities and highlighting a growing resource disparity in scientific research.

Unlike many AI fields obsessed with compute, the primary bottleneck in materials discovery is the speed and cost of running physical experiments. Progress depends on experimental throughput, not just bigger models or more GPUs.

OpenAI's effort to create 'SWE-bench-verified' demonstrates the immense cost of quality benchmarks, requiring millions of dollars and multiple human annotators per task. Despite this, a later audit revealed that 59% of the unsolved problems were actually impossible to solve due to inherent flaws.