The explosive, profitable revenue growth of major AI labs like Anthropic invalidates the theory that AI is a niche toy. This growth is happening despite rising hardware costs, demonstrating that businesses are deriving massive, tangible value from AI and are willing to pay a premium for it.
In a striking economic anomaly, the cost to rent older NVIDIA H100 AI chips is increasing, not decreasing. This is because the growth in AI's usefulness is outstripping the tripling annual supply of compute. It signals that the value being generated by AI models is growing faster than our ability to manufacture the hardware to run them.
Benchmarks like META's task completion chart can overstate AI capabilities. They compare AI performance to human professionals who are new to the codebase. An experienced human staff member, intimately familiar with the software, could likely complete the tasks in a fraction of the time, making the AI seem far less impressive by comparison.
The significant performance jump from Anthropic's Mythos model wasn't a sustained acceleration but likely a one-time gain from training a much larger model from scratch. This suggests AI progress follows a pattern of punctuated equilibrium: steady, incremental gains followed by sudden leaps when a company invests in a new, larger pre-training run.
Even if AI fully automates coding tasks at a lab like Anthropic, it may not dramatically accelerate overall research. The real constraint will become access to compute for training and experiments. With human labor effectively infinite, the scarcity of chips becomes the primary bottleneck, limiting the speed of recursive self-improvement.
Despite superhuman coding skills, AIs struggle to run real-world businesses like cafes. They can manage daily operations but fail at long-term strategy, prioritization, and open-ended thinking. This "messy task problem," stemming from a lack of dense feedback loops for training, is a primary blocker to full job automation.
An AI model disproved a mathematical conjecture not through a flash of creative genius, but by methodically applying a known technique from a different math subfield. This highlights AI's current strength: synthesizing vast, disparate human knowledge rather than generating truly novel, alien ideas. It's an exhaustive librarian, not an intuitive genius.
AI task completion costs have consistently held at just 3% of the human equivalent. This means there's a huge, untapped lever for performance: we could let an AI 'think' 30 times longer on a critical problem, boosting its capabilities, and it would still only cost as much as hiring a person. This economic runway negates concerns about inference scaling costs.
The key uncertainty splitting AI bulls and bears isn't about data or compute, but 'spillover.' Will training models to be superhuman coders (a clean, verifiable task) also make them proficient at messy, open-ended tasks like business strategy? The size of this spillover effect is unknown and is the biggest determinant of AGI timelines.