The Army Is Burning Through Its AI Tokens

Executive Summary

The recent experience of the US Army highlights a crucial lesson for private equity-backed companies: while AI promises significant operational uplift, an unmanaged, "unlimited" approach to its adoption can quickly lead to unforeseen costs, diminished returns, and even operational setbacks. Strategic AI enablement and robust data transformation are paramount to converting AI's potential into tangible EBITDA growth and measurable ROI, rather than merely burning through budgets with limited value.

Introduction

The buzz around Artificial Intelligence has promised a new era of productivity, efficiency, and competitive advantage. For private equity firms and their portfolio companies, the allure of margin expansion and accelerated growth through AI is undeniable. Yet, as with any transformative technology, the path to value realisation is paved with strategic decisions and potential pitfalls. A recent report detailing the US Army's rapid depletion of its AI "tokens" – essentially, its computational budget for generative AI tools – serves as a stark, commercial-grade cautionary tale. It underscores that unchecked enthusiasm, without a clear strategy for effective usage and underlying data readiness, can quickly turn a promising investment into an unexpected drain on resources, threatening projected payback periods and overall EBITDA impact.

Key Developments: The Army's AI Token Challenge

In May 2026, the US Department of Defense (DOD) proudly announced that nearly half of its 3.5 million employees were actively using AI. Shortly after, an internal email revealed that the Army’s Combat Capabilities Development Command (DEVCOM) had rapidly exhausted its allocated AI tokens – a unit representing the computational cost of generative AI outputs. Despite an initial promise of "unlimited tokens," the Army CIO pool was depleted within weeks, forcing a re-establishment of usage limits.

The Army relies on platforms like Ask Sage, which provides access to various large language models (LLMs). An annual enterprise subscription granted the Army 100 million tokens, an amount anticipated to last the entire year. Yet, this entire year's allocation was reportedly burned through for just one service in a fraction of that time. This wasn't an isolated incident; similar overruns in AI usage budgets have been reported by major tech firms like Meta and Uber.

Adding to the concern, an anonymous Army employee indicated that despite being encouraged to "lean into" generative AI, the tools were often unreliable, sometimes even falsely asserting task completion. This highlights a critical operational challenge: simply providing access to AI tools does not equate to effective, reliable, or value-generating usage. It risks significant rework, diminished trust in new technologies, and a net negative impact on productivity.

What This Means for PE-Backed Companies

The Army’s experience offers invaluable insights for private equity-backed companies navigating their own AI journeys, particularly for non-technical executives focused on the bottom line.

• • Operational Efficiency and Margin Expansion: The rapid depletion of AI tokens translates directly to unforeseen operational expenditure (OpEx). For a commercial enterprise, this would mean unexpected cloud computing costs or increased software licensing fees, directly eroding margins. Without a clear strategy for efficient AI utilisation, the promised operational efficiencies can easily turn into budget overruns. Furthermore, unreliable AI outputs, as reported by Army personnel, necessitate human oversight, verification, and often rework, nullifying productivity gains and inflating labour costs.
• • EBITDA Impact and Measurable ROI: Every pound spent on AI must contribute to EBITDA. Uncontrolled "token burn" without commensurate value generation directly impacts profitability. The Army’s situation underscores the importance of a clear ROI framework for AI investments. Simply encouraging usage without measuring its true impact or ensuring reliability means capital is being deployed with uncertain, and potentially negative, returns, extending payback periods far beyond initial projections.
• • Risk Reduction:
• • Data Quality: The unreliability reported by Army users points to a critical underlying issue: the quality and structure of data. AI models, particularly LLMs, are only as effective as the data they process. Poor data quality can lead to inaccurate outputs, compliance risks, and flawed decision-making, increasing operational risk.
• • Compliance & Governance: Even with "Controlled Unclassified Information," the scale of AI usage demands robust governance. For commercial entities, this extends to data privacy, intellectual property, and regulatory compliance. An "unthinking application" of AI, as noted in the article, can expose companies to significant regulatory and reputational risks.
• • Workforce Capability and Trust: Pushing employees to use unreliable tools can erode trust in new technologies and leadership. True workforce productivity uplift comes from enabling employees with tools that genuinely enhance their work, backed by comprehensive AI upskilling and support. Otherwise, enthusiasm quickly wanes, and investments in new tech sit idle or are misused.

The Epoch AI Perspective

The challenges faced by the Army are not unique to government agencies; they resonate deeply within the commercial sector. At Epoch AI Consulting, we recognise that unlocking AI's true potential for PE-backed companies requires a holistic, strategic approach that addresses both technology and people.

Firstly, the Army's token burn highlights the urgent need for targeted AI enablement. It's not enough to provide access to AI; employees, from the factory floor to the executive suite, need to understand how to use it effectively, efficiently, and responsibly. Our custom AI training portal delivers tailored learning paths, ensuring that workforce AI training is relevant to a business's specific tools and operational context. This focused AI skills development helps prevent wasteful usage, improves output quality, and ensures that every interaction with AI contributes to productivity. We provide comprehensive corporate AI training, including AI training for executives, fostering AI literacy and accelerating AI upskilling for portfolio companies to drive tangible business outcomes, not just token consumption. This ensures that valuable computational resources are used judiciously, directly supporting margin expansion.

Secondly, the reported unreliability of AI outputs points directly to the foundational importance of data transformation. AI models are powerful, but they are dependent on high-quality, well-structured data. Our expertise in data architecture, data engineering, and AI engineering ensures that a business's data is captured, moved, and processed in a way that makes it AI-ready. This includes building a robust modern data stack, improving data quality, and creating the necessary infrastructure for reliable AI application. By tackling data transformation for PE-backed companies, we reduce the risk of inaccurate AI insights, enhance decision-making, and ensure the data foundation is solid for measurable ROI, offering data architecture for AI readiness. Our capabilities in data science, analytics consultancy, and data visualisation further empower businesses to derive clear, actionable intelligence from their data.

Finally, while generic LLMs offer broad utility, the Army's struggle with unreliable outputs underscores that a one-size-fits-all approach is often insufficient for specific operational challenges. This is where software engineering becomes critical. We develop bespoke software development solutions – internal tools development that are purpose-built to solve specific operational problems. Examples like AI-powered sales tools integrated with CRMs or intelligent stock management systems offer a much higher probability of delivering precise, reliable results and a clear ROI. These custom applications are designed for maximum operational efficiency, directly addressing the pain points that off-the-shelf or general-purpose AI tools might miss or exacerbate, ensuring that AI investment drives genuine EBITDA impact.

Conclusion

The US Army's experience serves as a clear warning: the promise of AI is immense, but its strategic implementation is paramount. For PE-backed companies, the board-level decision is not if to adopt AI, but how to ensure it delivers measurable ROI, expands margins, and reduces operational risk. This demands more than just providing AI access; it requires a structured approach to AI enablement, foundational data transformation, and targeted internal tools development. By investing strategically in these areas, companies can transform AI from a potential cost centre into a powerful engine for sustained growth and competitive advantage, safeguarding future EBITDA and accelerating speed to value.

Source: The Army Is Burning Through Its AI Tokens