Tokenomics: smart strategies to deal with the costs of AI and infrastructure
Roula Khalaf, Editor of the FT, selects her favorite stories in this weekly newsletter.
TL;DR
- AI usage costs are rapidly increasing, moving from free consumer tools to paid subscriptions and token-based pricing.
- Tokens are pieces of information AI uses; their cost varies by model complexity and whether they are input or output.
- AI companies initially offered free access to gain users but are now monetizing due to surging training and infrastructure costs.
- The rise of AI agents and complex workflows significantly increases token consumption and unpredictability in billing.
- Enterprises are seeing substantial growth in AI spending, with median expenditure per employee per month rising significantly.
- Strategies to control costs include shopping around for models, using simpler models for basic tasks, and implementing model routing.
- Open-weight AI models offer cost savings and customization but come with security and hosting considerations.
- Companies are advised to fit the tool to the task, avoid over-reliance on the most complex models, and leverage institutional knowledge to reduce redundant queries.
- Building proprietary AI models is becoming an attractive option for cost control, data ownership, and independence from major AI providers.
- The future of AI pricing is uncertain, with potential commoditization due to capacity growth balanced against increased demand and IPO pressures.
- AI is shifting from an IT overhead to a primary operational driver, with a potential reallocation of headcount budgets into token spend.
- Successful AI deployment requires clear understanding of deployment goals, effective measurement of gains, and strict oversight of expenditure.