The AI arms race has entered its long-war phase

Artificial intelligence was sold to us as a technology of knowledge, meant to help us learn faster, reason better, and extend human capability. Like so many decisive technologies before it, AI has entered the military arsenal. That should unsettle us. Not because it is surprising, rather because AI is one of the few strategic technologies that can, in principle, be programmed against certain uses and still be pressured into serving them.

 

That tension became visible in the recent clash between Anthropic, the White House, and the Pentagon. Anthropic, which was already working with U.S. national-security customers and deploying its systems in classified environments, refused to remove two hard limits from a new long-term contract being negotiated: it would not allow its models to be used for fully autonomous lethal warfare without human oversight, nor for mass surveillance of American citizens. When the Pentagon demanded broader access for lawful military uses, the relationship collapsed, and Anthropic was branded a “supply-chain risk”. Then, OpenAI stepped in with a defence agreement of its own, also containing guardrails, but embedded differently in contract language, architecture, and oversight. [Anthropic has subsequently sued the US government against the ruling of them as a supply-chain risk on the grounds that it goes against the US constitution and the right to free speech.]

 

This was not a minor procurement dispute. It was a strategic event showing that the AI race is no longer a technological sprint between labs chasing model performance. It is becoming a long war: a prolonged struggle over who defines acceptable use, who shapes the rules of deployment, who captures industrial advantage, and who controls the political narrative around intelligence, power, and risk. The point is not simply that AI is useful in war. The point is that AI is now part of the infrastructure through which states, firms, and societies will compete for decades to come.

 

Markets as battlegrounds 

In Business Warfare, I argued that markets increasingly resemble theatres of conflict in which information, positioning, alliances, and timing matter as much as products themselves. AI makes that argument harder, sharper, and more literal. The competition is no longer just between companies. It is between states, industrial coalitions, regulators, defence establishments, and private labs that possess capabilities governments urgently want but do not fully control.

 

Sun Tzu would have recognised the pattern immediately. The highest form of victory is not brute destruction. It is the shaping of conditions. It is to know more, decide faster, and force the adversary into an unfavourable position before open confrontation begins. That is precisely why AI matters in defence and geopolitics. Properly integrated, AI accelerates learning, fuses fragmented data into usable knowledge, shortens decision cycles, and turns explicit information into tacit operational advantage. In military terms, AI compresses the distance between intelligence and action.

 

That is why the Pentagon wants AI, because tactical agility wins battles. And, in practice, strategy always bends to tactical necessity. Commanders do not ask first whether a capability is philosophically elegant. They ask whether it improves the speed and robustness of decision-making under pressure, and AI promises exactly that. It can improve targeting, planning, logistics, simulation, intelligence analysis, cybersecurity, and operational prioritisation, among others. Once that strategic decision is taken at the state level, ethical discomfort does not disappear, but it becomes subordinate to the logic of capability acquisition.

 

This is where the Anthropic episode becomes revealing. Anthropic’s position was not anti-state. Dario Amodei explicitly said he believed in using AI to defend the United States and other democracies. But Anthropic also insisted that an AI company could retain a narrow zone of independent moral judgment over forbidden uses. OpenAI, by contrast, reached a deal with the Pentagon that also preserved red lines, yet did so through a more cooperative structure with a lawful-use baseline and a cloud-only deployment.

 

And that difference matters. Not because one company was ethical and the other was not, but because both claimed safeguards. The real fault line lies elsewhere, in who gets the last word when law, strategy, and ethics collide under conditions of national-security urgency. New technology creates grey legal loopholes, and, aware of this, Anthropic sought to remain a strategic partner while preserving a right of refusal for edge cases. The USA administration appears to have viewed that as an unacceptable intrusion by a private actor into sovereign decision-making. The blacklisting of Anthropic imposed immediate commercial costs and turned a contract disagreement into a broader warning to the sector.

 

 

The nature of long wars

This is what a long war looks like. It is not a single battle over one contract. It is a durable rivalry over the allocation of force, industrial leverage, and legitimacy. Ferdinand Foch, who understood modern war as an organised struggle of wills and resources, would have seen that the decisive issue is not isolated brilliance but concentration of effort throughout time. In this regard, the AI powers are already concentrating efforts with governments mobilising public procurement and defence budgets. And governments also define regulatory design, energy strategy, semiconductor capacity, and export controls. And firms are also aligning talent, capital, cloud infrastructure, and political access. Consequently, universities, chip manufacturing, data centres, and military use cases are no longer separate domains. They are one strategic system.

 

Clausewitz helps us go one step further. War, he wrote, is never merely a technical contest. It is a political conflict conducted through other means. The same is increasingly true of AI competition. The argument is not only about who has the best model. It is about who shapes the operating environment in which AI will be used.

The battlespace now encompasses standards, access rights, liability rules, licensing, export restrictions, safety requirements, and public narratives. In other words, regulation is no longer an external constraint on the AI race. It has become one of its weapons.

 

That is why perception management now matters as much as engineering. Machiavelli understood that power depends not only on capability but also on how others interpret it. One must command fear carefully, secure loyalty where possible, and avoid being seen as weak or hostage to other people’s judgments. In the AI arena, this means governments cannot appear dependent on private firms for sovereign choices. In contrast, AI companies cannot appear reckless, hostile to democratic interests, or unable to defend their own legitimacy. Every actor is therefore managing the two fronts of operational power and political image in parallel.

 

This is also why the AI race is becoming permanent. Where a sprint has a finish line, a long war does not; phases, escalations, and evolving doctrines characterise it. The current phase is one in which ethical, regulatory, and strategic priorities are colliding rather than aligning. Labs like Anthropic want room to innovate and some control over misuse. Governments want access, speed, and strategic freedom. And regulators want accountability before the consequences become irreversible. But militaries want advantages that can be operationalised now, not after philosophical consensus arrives. Because none of these priorities will disappear, they will be forced into uneasy coexistence.

 

 

Concrete implications

For business leaders, that has concrete implications.

 

First, AI strategy can no longer be treated as a matter of productivity tooling alone. It is now entangled with geopolitics, industrial policy, and regulatory sovereignty. The question is no longer whether AI will affect your industry. It is how firms will survive in markets where governments may shape AI capability.

 

Second, leaders must learn to think like strategists, not merely adopters. Sun Tzu’s lesson is to understand the terrain, which, in practical terms, means mapping dependencies. In the AI case, we are talking about models, chips, cloud providers, jurisdictions, data rights, and regulatory permissions. It also means asking where firms could be exposed if a supplier is restricted, if compliance obligations harden, or if geopolitical blocs begin to diverge technically and legally.

 

Third, force allocation matters. In a long war, advantage rarely comes from spreading resources thinly across fashionable experiments. It comes from concentrating effort where a durable strategic position can be built. In business, that means choosing where AI truly changes economics and decision superiority, meaning investing seriously where it matters.

 

Fourth, perception is part of power, and that is where Machiavelli is useful again. Firms must be credible to regulators, useful to governments, trusted by customers, and resilient in public debate. In a politicised AI environment, legitimacy is not decorative. It is strategic capital.

 

Finally, leaders need to grasp that regulation is not just a risk to be managed: it is a terrain to be shaped. Clausewitz would insist that the political layer cannot be separated from the operational one. The winning firms will not be those that merely comply after the fact. Winners will be those who anticipate the direction of doctrine, participate in its formation, and design their organisations for a world in which AI competition is continuous.

 

The Pentagon clash with Anthropic was therefore more than a dispute over guardrails. It was an early glimpse of the strategic order now emerging. So, AI is no longer just a field of innovation; it is a field of rivalry. Its decisive contests will not be limited to benchmarks and product launches. AI will unfold in procurement battles, legal doctrines, military integration, industrial alliances, and narratives of legitimacy.

The arms race for AI has matured and entered its long-war phase.

 

And in a long war, survival belongs neither to the fastest nor to the loudest. It belongs to those who understand that intelligence, force concentration, and perception are all part of the same campaign. 

 

---

 

Paulo Cardoso do Amaral is the author of Business Warfare, a globally recognised strategist, and an expert in competitive intelligence.

 

Main image courtesy of iStockPhoto.com and PeopleImages