AI Strategy Needs a Better Story Than an Arms Race
AI strategy gets worse when leaders borrow military language for commercial decisions.
That is the real implication of Verity Harding’s argument in a recent WIRED essay on moving past AI nationalism. Harding, the former head of global public policy at Google DeepMind, argues that the arms-race metaphor feels clarifying but narrows thinking. For enterprise and public-sector leaders, that matters because the words used to describe AI often become the logic behind budgets, vendor choices, and rollout speed.
The market has spent the past two years importing geopolitical rhetoric into boardroom planning. That is a mistake. Nations may speak in terms of deterrence, export controls, and strategic advantage. Companies still need an AI roadmap, not a war plan.
Why the AI arms-race metaphor is failing leaders
Harding’s core point is not semantic. It is operational. If AI is framed as a race, then hesitation looks like weakness, cooperation looks naive, and governance looks like delay. That logic is attractive in headlines and dangerous in practice.
According to WIRED’s essay, Harding said the framing is “sexy” because it appears to simplify the stakes. But simplification is not the same as clarity. In enterprises, rivalry language changes behavior in predictable ways: executive teams over-prioritize speed, underweight interoperability, and treat every vendor launch as a strategic emergency.
That pattern now shows up across AI transformation programs. Instead of asking which workflows deserve automation first, leaders ask whether they are falling behind. Instead of defining success metrics, they chase parity with competitors whose use cases, margins, and regulatory exposure may be entirely different.
A race metaphor also distorts staffing. Teams buy AI consulting services before they have internal decision rights sorted out. They procure tools before they settle data boundaries. They mistake urgency for sequence.
Where the arms-race narrative came from
The shift Harding describes is real. In the late 2010s, much mainstream AI policy discussion emphasized international cooperation, safety, and shared standards. After late 2022, when ChatGPT made AI legible to non-specialists almost overnight, the frame changed.
Three forces converged.
First, the post-ChatGPT surge turned model capability into a public spectacle. OpenAI’s ChatGPT launch in November 2022 became a cultural marker, not just a product release. Second, pandemic-era politics had already revived national arguments about dependency, resilience, and borders. Third, the war in Ukraine made dual-use technology discussions feel immediate rather than theoretical.
The result was a narrative jump: from AI as a shared technical and governance challenge to AI as a civilizational contest. That jump was reinforced by export-control debates, including new US restrictions on advanced chips and model access discussed in policy analysis from the Center for Strategic and International Studies.
There is a steel-man case for this framing. Advanced models do have national-security implications. Compute concentration is real. Supply chains matter. Governments would be negligent if they ignored the strategic dimension of frontier systems. The OECD’s AI policy observatory and the Stanford AI Index 2025 both show how quickly capability and investment are concentrating.
But the rebuttal is stronger for most organizations: acknowledging geopolitical risk is not the same as adopting a zero-sum operating model. A company choosing AI implementation services for procurement, customer support, or internal search is not making nuclear doctrine. It is making workflow decisions under uncertainty.
Why smaller countries lose when AI becomes a race
The most under-discussed part of Harding’s argument concerns smaller powers. Once AI becomes a race between Washington and Beijing, everyone else is pushed into alignment politics. The same logic appears inside markets.
Mid-sized enterprises and public agencies often behave like smaller states in technology procurement. They assume standards will be set elsewhere. They over-commit to one vendor stack. They accept data terms, portability constraints, and integration assumptions that would look reckless in any other infrastructure category.
This is where AI strategy becomes materially different from AI enthusiasm. The useful question is not Who is winning? It is Where does bargaining power sit?
A practical example makes the point. In 2025, several enterprise software buyers accelerated copilots and knowledge-assistant deployments before settling identity architecture, document permissions, or retrieval quality thresholds. The technology worked well enough for demos. It failed in production because governance-light shortcuts created rework: duplicated connectors, conflicting data scopes, and low employee trust. That is what race framing does. It rewards visible movement over durable sequencing.
A better response is to build optionality into the AI roadmap: model portability where feasible, explicit data residency decisions, narrow use-case prioritization, and corporate AI training before broad access. Organizations that do this are not slower in any meaningful sense. They are reducing switching costs.
One natural way to structure that work is through a strategic advisory function such as a Fractional AI Director engagement, which fits here because the central problem is decision quality before scaling implementation, even if the market often mistakes it for a tooling problem.
What AI leaders should say instead of win and lose
The alternative vocabulary is not hard to find. The market is already splitting along five more useful terms: capacity, trust, resilience, interoperability, and fit.
Capacity asks whether the organization has the people, process, and data readiness to absorb AI into daily work. Trust asks whether outputs can be reviewed, explained, and corrected. Resilience asks whether the stack can survive policy shifts, outages, and vendor repositioning. Interoperability asks whether systems can move across tools and models without major rework. Fit asks whether the use case actually matters commercially or operationally.
This language sounds less dramatic because it is less dramatic. It is also more useful.
The strongest counter-argument is that softer language may hide the pace of change. If leaders stop talking in competitive terms, they may drift into bureaucracy while rivals deploy faster. That risk is real. McKinsey’s State of AI research has repeatedly shown a widening gap between companies experimenting with AI and those embedding it in operating models.
Yet speed and war framing are not the same thing. Fast organizations still define use-case economics, integration boundaries, and training requirements. Slow organizations often call themselves urgent while producing scattered pilots. The distinction is managerial, not rhetorical.
What the policy debate means for enterprise AI strategy
The policy story matters because enterprise language tends to mimic national language. When governments talk about sovereignty, companies start talking about owning the full stack. When governments talk about strategic dependence, buyers start overreacting to every platform shift. Some caution is healthy. Copying the rhetoric wholesale is not.
An effective AI strategy in 2026 should do four things.
First, separate frontier-model news from workflow value. Most organizations do not need the newest model for every task. Second, map where AI integration services will create irreversible dependencies, especially around proprietary data layers. Third, train managers to evaluate failure modes, not just demo quality. Fourth, create a simple decision cadence: what gets piloted, what gets governed, and what gets stopped.
The organizations that navigate this well usually treat AI transformation as an operating discipline rather than a status contest. They do not deny competition. They refuse to let competition write the entire script.
The takeaway: lead AI with strategy, not slogans
Harding is right about the danger of the arms-race metaphor, but the business implication goes further: bad framing produces bad sequencing. The leaders who will outperform in AI are unlikely to be the loudest. They will be the ones who choose language that leads to better decisions, cleaner implementation paths, and fewer expensive reversals.
Build AI strategy around bargaining power, trust, and fit—not around whether someone else says the race has started.
Martin Kuvandzhiev
CEO and Founder of Encorp.io with expertise in AI and business transformation
