Tokenomics: Why enterprise AI economics are changing

Enterprise AI budgets built on today's flat-fee pricing are no longer sustainable, as the subsidy that made AI feel cheap is ending. In South Africa, this issue is compounded by billing intelligence in US dollars while the revenue needed to justify it is earned in rands. Technology leaders are budgeting AI similarly to their first cloud migration a decade ago, which worked reasonably well for cloud computing but won't last for AI. The reason is that AI is getting better and cheaper, but the bill is still going up.

Understanding this shift is now crucial for board-level decision-making, as it's no longer just an engineering concern. The first widely adopted price for frontier AI was around $20 per month, following a venture capital approach to building market share. However, the economics behind this model are unsustainable, with large hyperscalers investing billions in AI infrastructure and OpenAI spending $1.35 for every dollar of revenue earned.

Generative AI's short life can be divided into three phases, each with a different economic signature. The first phase was characterized by simple availability, such as ChatGPT. The second phase introduced reasoning models capable of handling multi-step problems, leading to the first real price competition. The third phase is agentic, where systems plan, call tools, and chain multiple steps together autonomously.

The blind spot in this phase lies in the fact that inference prices have fallen between 75% and 90% per year, while token volume - the actual number of tokens an agentic workflow consumes - has grown 500% to 1,000% over the same period.

The gap between falling prices and growing token consumption is the entire story. South African enterprises face fluctuating operational costs due to rand-dollar volatility, leading to real cloud bill shock unrelated to any change in usage. Local cloud services market is projected to grow significantly, but enterprises face a strategic fork: remain token consumers renting intelligence from providers or become token providers by running models on owned or dedicated infrastructure.

The single most important number in AI cost planning for 2026 is the multiplier that autonomous agents apply on top of the price per token. Agentic workflows can consume 50 to 500 times more tokens than simple chat interactions, with three forces compounding this: agentic volume multiplication, hidden background consumption from monitoring agents, document watchers, and compliance surveillance systems, and retrieval overhead.

Enterprises need to consider budgeting in a currency they don't control and forecast consumption patterns they cannot yet predict, making it a more challenging problem than most global AI cost guidance addresses.