The key measure of leverage for AI-powered developers is no longer GPU utilization (FLOPs) but the volume of tokens processed by agents. Karpathy feels nervous when his token subscriptions are underutilized, indicating he's the bottleneck, not the system.

Current AI models are priced too cheaply, leading to inefficient consumption like using powerful models for simple tasks. As prices rise to reflect true costs, companies will need to optimize usage. This may create a new role, the 'Chief Token Officer,' responsible for allocating AI compute resources versus human capital.

Unlike traditional software, OpenAI's growth is limited by a zero-sum resource: GPUs. This physical constraint creates a constant, painful trade-off between serving existing users, launching new features, and funding research, making GPU allocation a central strategic challenge.

The core resource allocation question will evolve from budgeting for AI tools to choosing between hiring humans and buying compute tokens. Answering this requires a "software factory" with quantitative feedback loops to determine where each incremental dollar adds the most business value.

Escalating compute requirements for frontier models are creating a new market dynamic where access to the best AI becomes restricted and expensive. This shifts power to the labs that control these models, creating a "seller's market" where they act as "kingmakers," granting massive competitive advantages to the highest corporate bidders.

According to Mike Cannon-Brookes, advanced enterprises are not tracking AI success by counting tokens. Instead, they are asking harder questions about overall output, such as engineering productivity and quality. They understand that high token usage doesn't always correlate with high productivity, shifting focus from raw usage to tangible business outcomes.

For entire countries or industries, aggregate compute power is the primary constraint on AI progress. However, for individual organizations, success hinges not on having the most capital for compute, but on the strategic wisdom to select the right research bets and build a culture that sustains them.

Jensen Huang argues that elite AI engineers should not be constrained by compute costs. He proposes a heuristic: if a $500k engineer isn't consuming at least $250k in tokens annually, their talent isn't being leveraged effectively. This reframes compute from a cost center to a critical force multiplier.

As demand for AI far outpaces compute supply, costs will rise. Only labs with the most lucrative algorithms, like OpenAI and Anthropic, can afford it. They reinvest massive revenues into the next training run, creating a self-reinforcing loop that raises the barrier to entry for any potential competitor, solidifying their duopoly.