The AI Order and the Alliance Calculus: Washington's Parallel Recalibrations

The Trump administration is expected to sign an executive order on artificial intelligence and cybersecurity on May 21, 2026, according to two sources who spoke to Reuters. The order is anticipated to accelerate frontier AI development, push faster federal adoption, and expand US technical leadership through increased compute infrastructure and chip distribution frameworks. The timing, sources suggest, is partly driven by political pressure from the administration's base for visible AI policy wins.

That same week, the administration moved on a second front. Reuters reported on May 20 that US officials planned to inform NATO allies that Washington would scale back the forces it commits to the alliance during major crises. A separate Polymarket market implied an 8% probability of full US withdrawal from NATO before 2027. Two separate strategic recalibrations, announced within days of each other. The administration is simultaneously reducing commitments to its oldest security alliance while declaring AI a core national priority.

The executive order is not unexpected. US AI labs—Anthropic, OpenAI, Google DeepMind—have spent years arguing that frontier development requires sustained public backing: compute infrastructure, talent pipelines, regulatory clarity, and defense contracts. The order appears designed to deliver on at least some of that ask. Industry groups have pressed for faster federal AI deployment, streamlined procurement, and a coherent export-control framework that protects US model weights without strangling commercial collaboration. The incoming order addresses at least the first two items.

The geopolitical framing is harder to ignore. Announcing a major AI initiative during a week when the US is signalling it may reduce NATO commitments is not coincidental positioning. It reads as a statement: American technological leadership is the new alliance architecture. The administration appears to be betting that frontier AI—proprietary models, compute infrastructure, the industrial base underlying both—can function as a substitute for the network of formal security guarantees that has anchored US global influence since 1945.

Whether that bet is sound is another question. NATO's Article 5 guarantee is not merely symbolic; it has deterred aggression against member states for decades by tying American military power directly to allied territory. AI dominance, by contrast, is a more diffuse and less tested instrument of influence. The US can restrict chip exports, accelerate frontier models, and ink federal adoption agreements. Whether that translates into the same kind of credible deterrence that the NATO umbrella provided is uncertain. The Polymarket market on US NATO withdrawal—while not a calibrated prediction—suggests markets are pricing meaningful uncertainty about the alliance's future. Markets are not wrong to notice.

Anthropic's likely first quarterly operating profit offers one data point on the commercial health of the frontier AI ecosystem. The company has built substantial enterprise revenue through its Claude model family, with some estimates placing Anthropic's annual recurring revenue above $1 billion in recent quarters. For the first time, the company appears to be generating more revenue than it spends on operations—training runs, inference infrastructure, and headcount—without relying on external capital injection.

That is a genuine milestone. The AI sector has operated for years as a capital-intensive venture: large upfront infrastructure costs, uncertain return timelines, and a persistent dependence on investor appetite for frontier risk. Anthropic crossing the threshold where revenue covers operational costs marks a maturation moment for at least one major lab. The broader industry context matters here. US AI labs collectively burned through billions in 2024, with Anthropic alone reportedly spending roughly $2.8 billion on compute and infrastructure that year. Any operating profit, symbolically significant, remains fragile relative to those expenditure levels.

The strategic significance runs deeper than the balance sheet. A commercially sustainable Anthropic—alongside OpenAI's continued enterprise growth and the maturing inference economics across the sector—strengthens the case for AI as a domain where US leadership has genuine commercial foundations, not just government subsidy. That matters as Washington crafts its export-control and compute-infrastructure frameworks. Industries that can stand on their own commercial feet are easier to protect and export; those that depend on perpetual public funding are harder to scale globally.

The Chinese counterpoint is harder to dismiss. DeepSeek's R1 model demonstrated that frontier-level AI performance is achievable at a fraction of Western compute costs, using efficient architecture and open-source distribution. Beijing's policy apparatus has made clear that competitive AI development—not dependence on US chips or models—is a strategic priority. The US AI executive order will need to grapple with that pressure directly: commercial sustainability matters only if US labs can sustain competitive advantage against a Chinese ecosystem that has already proven it can close the capability gap at lower cost structures.

What the Anthropic milestone ultimately confirms is that commercial AI is no longer a speculative bet. The sector is generating real revenue, attracting real enterprise customers, and—at least for one major lab—approaching the economics of a functioning business. For the geopolitics of AI, that matters. A profitable Anthropic is a more durable Anthropic. A more durable Anthropic is a stronger data point in the argument that US AI leadership has structural foundations.

Those foundations still need to be built out. The executive order addresses compute, adoption, and chip distribution—but the harder strategic questions remain unanswered. Export controls require partner-country cooperation that NATO commitments have historically facilitated. Standard-setting authority depends on alliances of preference, not just market size. AI governance frameworks will need international buy-in to function. Reducing commitment to the formal alliance architecture does not automatically eliminate those needs; it may simply make them harder to meet through the conventional diplomatic channels that US influence has historically relied on.

The administration is making a bet on frontier AI as a primary instrument of national power. The executive order is the opening move. Whether that bet pays off will depend on whether AI dominance can be made to mean what the NATO umbrella once meant: credible, deterrence-backed, and backed by the willingness of allies to align with US standards. Anthropic's profit is a necessary but not sufficient condition for that outcome.