AI Geopolitics: The Arms Race Is Won in Year Two

The geopolitics of artificial intelligence rewards those who remain standing in year two. Year one shows the capability. Year two shows the bill.

Here is the pattern, here are the precedents, here is where it leads. Anyone managing capital should watch the sustainability line item, never the demo.

The precedent the market forgets

In November 2002, in Mombasa, Kenya, terrorists fired shoulder-launched heat-seeking missiles at an Israeli airliner during takeoff. The attack failed. The mechanism that followed was clear: the United States assessed adapting military air defense technology to commercial fleets.

The technical idea already worked. Detect the incoming missile, then divert its guidance with a directed infrared countermeasure. The hard part came after the demonstration.

The Government Accountability Office documented how the military systems of the era achieved roughly 300-400 hours between failures, while the commercial target exceeded 3,000 hours, according to the account published by Geopolitical Monitor. The program aimed to cut operating costs to about $350 per flight and the unit price to about $1 million at large scale. By the end of the decade, deployment across the entire US commercial fleet was missing, airlines judged the costs unsustainable, and the fiscal year 2010 budget zeroed out the funding.

Ankara 2026, the same scheme

At the NATO summit in Ankara in July, allies announced over $40 billion in anti-drone investment over five years. The next day, the Alliance published a new industrial strategy revolving around three verbs: strengthen, scale, sustain. The third verb carries the weight.

The initial purchase reveals little about the strategic winner. The long-term advantage belongs to the country able to operate the system after the demonstration team leaves the field.

AI complicates the problem. Capability is shown with unusual ease and priced with unusual difficulty. The ownership architecture stays offstage:

• data center compute and energy
• data pipelines and integration
• cyber assurance and retraining
• model monitoring and operator support
• software updates and export restrictions
• cost of abandoning the vendor

NATO's own revised AI strategy flags the energy demand of compute and the difficulty of bringing dual-use systems into the military domain. These seemingly secondary details decide the durability of military power.

The causal mechanism

Modern power geopolitics rewards operational sustainability, never the initial spectacle. The Mombasa precedent and the 2026 anti-drone race share the same structure. The context differs, the architecture coincides.

Three precedents are enough to call it a pattern: post-2002 civilian air defense, the costly surveillance programs that collapsed under maintenance, the current wave of AI procurement. In all documented cases, the deciding factor becomes lifecycle cost.

The mechanism works like this: capability attracts the initial budget, then energy and retraining erode the operating budget, and finally the program shrinks. The divergence between purchase price and running cost always resolves. The question is how.

The semiconductor constraint

Here is my founding position applied to this case. The US-China geopolitical competition on AI is a problem of access to semiconductors, never of model capability. Whoever controls the fabs, TSMC and Samsung, controls the outcome.

Models are copied. Factories endure. The US GPU sanctions against Huawei remain the most geopolitically significant move of the past five years, more than the 2018 trade war.

Applied to defense: the anti-drone edge depends on sustainable compute, on low-cost energy, on the chip supply chain. A country that depends on foreign fabs for its AI arsenal carries a structural risk in its war plans. The market has yet to price this constraint into European defense budgets.

My thesis, and what would refute it

My position, explicit: the AI arms race is won in year two, when running cost separates durable systems from abandoned demos. The consensus looks at the capabilities announced in Ankara. I look at the sustainability line item.

What would change my reading? A leap in the energy efficiency of military compute, enough to bring operating cost below the threshold that sank the 2002 civilian defense program. Or an autonomous European semiconductor supply chain by 2028.

Absent these two conditions, the pattern holds. The history of civilian air defense provides the numerical precedent. Cost structure dominates demo structure.

Three implications for capital

Family offices and sovereign funds, 36-month horizon: overweight semiconductor makers and the utilities powering military data centers, underweight pure model providers. Value migrates toward hardware and energy.

CEOs and boards, 24-month horizon: the geopolitical risk absent from the plans concerns dependence on foreign fabs for critical AI systems. A European board should map its exposure to Taiwanese chips today.

Chief risk officers, 18-month horizon: the scenario missing from VAR models is the cut of a major AI defense program due to running costs, with knock-on effects on listed suppliers. Add it to the stress tests.

The forecast

The forecast, with horizon, confidence and verification indicator. By December 31, 2027, at least one major NATO AI or anti-drone procurement program, among those tied to the $40 billion Ankara commitment, will be publicly downsized or restructured for sustainability costs: energy, compute, retraining.

Confidence: 68%. Horizon: December 31, 2027. Verification: an official audit or a budget line documenting the cut. The signal that would refute the thesis: all anti-drone programs under the $40 billion commitment remain funded at the original level or above, with no restructurings tied to running costs, by the end of 2027.

What to watch

What to watch: three leading indicators will confirm or refute the reading. First: the fiscal year 2028 budget requests of European defense ministries, under the operations and support line for AI systems.

Second: dedicated energy supply contracts for military data centers, a signal that precedes the sustainability crisis by several months. Third: new export restrictions on advanced chips, which redefine the cost of abandonment for allies dependent on outside suppliers.

This data speaks before the official announcements. The rest is noise.