Abstract

Background: The way enterprises account for AI inference costs is fundamentally broken. Billions of tokens are consumed daily, yet the knowledge generated reasoning traces, domain workflows, agent execution plans are discarded the moment each task completes. Every subsequent similar task restarts from scratch, burning the same resources again. This paper argues that such a model is economically indefensible as AI adoption scales. Problem: The root cause is what is called token leakage, the systematic discarding of high-value knowledge artefacts after inference. In unmanaged multi-agent deployments, our simulations show leakage rates of 65-68%, meaning that most of an organisation's AI expenditure produces no durable value. Contribution: This paper introduces the Token Capital Framework (TCF) and the concept of Token Capital the cumulative, indexed, persistent stock of reusable knowledge assets produced through AI inference. TCF operationalises Token Capital through six integrated components: the Token Capital Lifecycle (TCL), Token Capital Stack (TCS), Token Capital Maturity Model (TCMM), Token Capital Engine (TCE), Adaptive Token Capital Scheduler (ATCS), and Token Capital Benchmark Suite (TCBS). Full TCF deployment reduces token leakage to 11%, raises the reuse ratio to 71%, delivers a 54% effective cost reduction, and produces a Knowledge Compounding Factor of 3.8× over 30 days.

References

• Chen Y, Chen J, He C: Token Economics for LLM Agents: A Dual-View Study from Computing and Economics. arXiv. 2026, 2605:09104. 10.48550/arXiv.2605.09104
• Amayuelas A: Self-Resource Allocation in Multi-Agent LLM Systems. arXiv. 2025, 2504:02051. 10.48550/arXiv.2504.02051
• Xu W: A-MEM: Agentic Memory for LLM Agents. arXiv. 2025, 2502:12110. 10.48550/arXiv.2502.12110
• Kerestecioglu D: Human-Inspired Memory Architecture for LLM Agents. arXiv. 2026, 2605:08538. 10.48550/arXiv.2605.08538
• Patil C: Beyond Per-Token Pricing: A Concurrency-Aware Methodology for LLM Infrastructure Cost Estimation. arXiv. 2026, 2606:11690. 10.48550/arXiv.2606.11690
• Token Economics and the Evolving Role of FinOps. (2026). https://www.finops.org/insights/finops-x-2026-day-1-keynote/
• Dai LP: A Token Adaptive Computing Resource Allocation Strategy for Large Language Models. arXiv. 2024, 2403:01425.
• Linux Foundation: Tokenomics Foundation Announced to Establish Open Standards for AI Cost Management. (2026). https://www.linuxfoundation.org/press/linux-foundation-announces-the-intent-to-launch-the-tokenomics-foundation-to-establish-open-standards-for-ai-cost-management
• Goldman Sachs Research: AI Agents Forecast to Boost Tech Cash Flow as Usage Soars. (2026). https://www.goldmansachs.com/insights/articles/ai-agents-forecast-to-boost-tech-cash-flow-as-usage-soars
• Vickrey W: Counterspeculation, Auctions, and Competitive Sealed Tenders. Journal of Finance. 1961, 16(1):8-37. 10.1111/j.1540-6261.1961.tb02789.x
• Zhao C: LLM-Auction: Generative Auction Towards LLM-Native Advertising. arXiv. 2025/2026, 2512:10551. 10.48550/arXiv.2512.10551
• Chhikara P: Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory. ECAI / arXiv. 2025, 2504:19413. 10.48550/arXiv.2504.19413.
• Lim B, Arik SO, Loeff N, Pfister T: Temporal Fusion Transformers for Interpretable Multi-Horizon Time Series Forecasting. International Journal of Forecasting. 2021, 37(4):1748-1764. 10.1016/j.ijforecast.2021.03.012