- Token consumption is skyrocketing, with Goldman Sachs projecting 24x growth by 2030, making cost governance a top priority.
- Enterprises face runaway AI bills without observability; a single client incurred a $500 million monthly charge due to lack of monitoring.
- Industry leaders like AT&T, Cisco, and Accenture are adopting frameworks that tie token usage to business outcomes, not just volume.
- The next wave of optimization shifts from tokenomics to datanomics, emphasizing the cost of preparing data for AI.
The era of treating AI as a lab experiment is over. Enterprises are now grappling with the hard reality of operationalizing artificial intelligence at scale, and the financial implications are becoming impossible to ignore. Tokenomics — the economics of AI token consumption has emerged as a core operational discipline that separates successful AI deployments from costly failures. As organizations move from pilot projects to production systems, understanding and controlling token costs is no longer optional; it is a strategic imperative.
Tokens are the fundamental unit of currency in the AI economy. Every query, document summary, or code generation request consumes tokens, and commercial AI services from providers like OpenAI, Google, and Anthropic price their offerings accordingly. Goldman Sachs projects that token use could grow 24-fold by 2030, while the cost per token is expected to fall 60–70% annually. This combination makes AI feel cheaper at the unit level but drives up total enterprise spend dramatically, creating a financial paradox that demands new management approaches.
The Explosion of Token Consumption
The shift from static chatbots to autonomous agents is fundamentally changing demand patterns. Unlike chatbots that wait for human prompts, agents run continuously and increasingly communicate with other agents, pushing token consumption into uncharted territory. AT&T, for example, now burns over a trillion tokens per month, with that number growing by double digits. The telecom giant has more than 100 generative AI models in production, handling everything from customer care and fraud detection to network optimization and cell tower placement.
This explosion is not limited to telecom. Cisco’s president and chief product officer, Jeetu Patel, noted that fewer than 1% of potential users are currently deploying agents at scale, leaving enormous headroom for growth. Agentic AI workloads generate up to 450% more total traffic per task than a human performing the same work, according to industry research cited by AT&T’s CEO. By 2035, agentic adoption is projected to drive approximately 9x growth in enterprise traffic and 7x growth in consumer traffic.
The High Cost of Flying Blind
Without proper governance, token consumption can quickly spiral out of control. One AI consultant shared that a client received a half-billion-dollar bill in a single month simply because they failed to monitor their team’s AI spend against the value generated. This is not an isolated incident; it is the inevitable outcome of a fundamental misalignment between innovation speed and oversight. Traditional budgeting cycles take months, while AI agents can fan out and consume millions of tokens in minutes.
The problem is compounded by the fact that token counts alone are misleading. A million tokens could produce valuable research or an enormous quantity of confident nonsense. Amazon reportedly shut down a token usage leaderboard after an executive warned employees not to use AI just for the sake of using AI. The lesson is clear: higher token consumption does not equal better work. Organizations must connect token data to real business outcomes.
A Three-Pillar Framework for Token Governance
To move from reactive spending to proactive financial control, experts recommend a three-pillar approach. First, financial governance establishes centralized visibility into consumption, mapping token usage to specific business outcomes and departmental goals. This shifts the conversation from cost-tracking to value creation. Second, value-based performance optimization moves beyond raw volume to focus on the cost-to-accuracy ratio, reserving expensive frontier models for strategic tasks while using efficient alternatives for routine work.
Third, operational resilience ensures that infrastructure can handle the unpredictable load of AI workloads. Splunk provides observability tools that operationalize this framework, offering real-time insights into token consumption and infrastructure demands. Accenture has launched a similar offering called Accenture Tokenomics, which follows a three-stage process: see it, fix it, and keep it. Their AI Token Navigator includes seven specialized tools for forensics, model benchmarking, engineering efficiency, and real-time monitoring.
Real-World Strategies from Industry Leaders
AT&T has been particularly proactive in managing token costs. The company works closely with AMD to leverage open-source models and different chipsets, post-training models on AMD hardware to drive down token costs through what they call ‘moneyballing’ across models. AT&T’s CTO Jeremy Legg emphasizes that data sovereignty is key — not being tied to a specific chipset, model, or development tools. Their Open Telco AI model (OTel) has seen over 18 million downloads, and OTel 2.0 is now available via open source.
Cisco, meanwhile, has released Antares, a family of open-weight models designed to locate vulnerabilities in code without sending proprietary data to the cloud. This reflects a broader trend toward smaller, task-specific models that can handle 70% of use cases at a fraction of the cost. Cisco’s Cloud Control provides a unified management plane to track inference capacity across cloud, private data centers, and deskside computers, giving enterprises visibility into which agents are consuming tokens and whether usage has become excessive.
Beyond Tokenomics: The Rise of Datanomics
While tokenomics focuses on the efficiency of AI output generation, a parallel discipline called datanomics is gaining traction. Datanomics measures the efficiency of preparing the foundation that makes AI outputs valuable. Before an AI system generates its first token, enterprises must transform massive volumes of fragmented data into trusted, governed, AI-ready intelligence. Research from MIT found that roughly 95% of enterprise generative AI pilots deliver no measurable return, and Gartner predicts that 60% of AI projects will be abandoned due to lack of AI-ready data.
The rise of agentic AI makes datanomics even more critical. Gartner predicts that over 40% of agentic AI projects will be canceled by 2027 due to escalating costs, unclear business value, or inadequate risk controls. McKinsey found that eight in 10 companies cite data limitations as a roadblock to scaling agents. The organizations that win with AI will not necessarily be those with the largest models or the most GPUs, but those that can most efficiently transform their enterprise data into trusted intelligence.
The era of ‘AI at any cost’ is over. Success now belongs to organizations that treat observability not as a technical requirement, but as a strategic imperative. By mastering the economics of every token — and the data that fuels them — enterprises can transform AI from a volatile expense into a predictable, high-impact engine for sustainable growth. The next phase of enterprise AI will be defined by whether companies can make AI’s value compound faster than its consumption, shifting the conversation from ‘How much did AI cost?’ to ‘What business outcome did AI deliver, and how do we improve that ratio over time?’
