For years, AI has concentrated most of the attention in markets. But the recent events in the Persian Gulf and the ongoing war between the US and Israel, Gulf states, and in some way, NATO too, all against Iran have taken the top spot in interest.

And for good reason. But how does a potential long war in Iran affect AI?

I’ve analyzed how AI is being used, the war’s impact on the AI industry, and, perhaps most importantly, I’ve built a financial model I will give you that connects the dots between oil price increases, the financial metric the entire world is paying attention to, and how it impacts the most important metric in AI: infrastructure spending.

And here’s a hint for you: soon things might not be looking good, my dear reader. Let’s dive in.

The most important point I want to make clear is that ‘AI in war’ is not about killer robots, weapons, mass destruction, or other fanfic. It’s actually much less spectacular.

Right now, its real wartime role is to speed up human-led operations by compressing the kill chain: turning surveillance, intercepted communications, and other data into target lists, threat rankings, and operational options much faster than before.

Doing what AI does best: crunching data at scale.

That said, public reporting does say Israel used AI alongside espionage to help prepare strikes inside Iran, and US forces are using AI tools to handle huge volumes of operational data in the Iran campaign. Interestingly, despite the public fall-off, Reuters reported that Anthropic’s Claude has supported US military operations tied to Iran, including intelligence analysis and planning.

Beyond being used for targeting and planning, it is also reshaping drone warfare, missile defense, and command systems, which can leverage on-device AIs if they lose connectivity with the command centers.

But the big use case that is often neglected is that AI is now a major weapon in the information and cyber war surrounding the fighting.

AI-generated and altered videos are flooding the information space around the Iran war, often amplified by state-linked propaganda (mostly from the Iran side), which makes it harder to verify casualties, battlefield outcomes, and escalation claims in real time.

So the condensed takeaway is: AI in this war is being used for analysis, targeting, defense, cyber operations, and propaganda, and its main strategic effect is faster decisions on a larger scale, with less transparency and higher risk of error. Again, doing what AI does best: crunching data at scale.

So, for now, you can rest easy in that we can leave humanoid robots to Hollywood. However, the conversation gets much more interesting when we start discussing the economic impact of the war on AI.

Having analyzed how AI is being weaponized, let’s now turn to the economic side: is this war making AI more expensive? The answer is mostly about common sense, but let’s give more detail than simply stating the obvious.

And we’re starting at the core of investors’ worries: oil.

Due to Iran’s big presence in the area, concentrating a vast amount of the Persian Gulf coastline in a gulf packed with oil refineries, Iran has announced it will burn any ship that tries to go through the Strait of Hormuz, the maritime path all ships have to go through to leave to other areas, and one of the most important oil supply chain chokepoints in the world.

But why is this Strait so important?

The main buyers of this oil are China, India, South Korea, and Japan, in that order, but the actual impact on each economy is much more nuanced.

In fact, despite being the largest importer by nominal count, China’s exposure to oil supply shortages coming from this area seems to be much more limited than people imagine, putting Singapore and Taiwan in a more “interesting” position, losing more than 1% GDP if oil prices stay at or above $85.

Nevertheless, the potential loss of access to 20% of daily supply has automatically caused price surges, with today’s prices already going above the danger zone of $100 a barrel of Brent to a peak just over $120, rising more than 30% in days, only to come down back to almost flat in the day by dropping another 30%, one of the biggest drop on record (Brent price at the time of writing is just over $89).

We can celebrate price drops, but this unforeseen volatility is not welcomed. Furthermore, Iran knows this is a fight they can’t win by brute force, so they are simply trying to make the war too expensive for the US and the world in general, even reaching the point of attacking other Gulf states, like the UAE, Bahrain, or Saudi Arabia, even forcing Bahrain to drive entire oil output to a halt due to force majeure.

Okay, but why are oil prices so important? The most obvious impact is overall inflation, as oil price hikes translate directly into higher living costs. Gas and diesel become more expensive, transport and construction costs skyrocket,…

Moreover, the potential shutdown of the Strait has also led to price rises in Liquefied Natural Gas and Helium. The former puts huge inflationary pressure on energy prices, too, especially in Europe, while the latter has strategic implications for AI.

What implications? Industrial gas suppliers describe helium as being used at “hundreds of points in the fab” for cooling. In particular, it’s used for backside wafer cooling, loadlock cooling, as a carrier gas in some cases, and sometimes to help support plasmas.

Hence, a sudden shortage of Helium could seriously disrupt advanced node manufacturing (the production of advanced chips used in AI), affecting everything, particularly yield (the ability to manufacture chips at scale without the majority coming out broken).

The problem? Qatar, a Gulf state, is home to one of only two plants that produce semiconductor-grade helium, which is ionized and used to etch silicon wafers.

But the less explored implication is toward AI: how does this conflict affect the AI industry as a whole?

The first noteworthy point is that AI is seen as a strategic investment by a large part of the Gulf states (if not all) in their efforts to diversify away from oil, which is seen as a ‘dying asset’ (although it’s not quite as clear these days, but the pivot is still clear).

Iran knows this, as it has since begun targeting data centers in other Gulf states, hitting up to three Amazon Web Services campuses: two in the UAE and one in Bahrain.

We aren’t talking about peanuts here. Currently, there are up to $300 billion in announced investments in AI data centers in the region, with even Saudi Arabia’s most bold project, The Line, being repurposed as a data center.

Some announcements include:

Therefore, a prolonged conflict could risk it all. This is terrible news not only for the Gulf States but also for US Big Tech, which has turned to the Gulf as a way to expand its computing footprint, especially given the notorious internal resistance to new data center developments in the United States, with just 25 data center projects canceled in 2025 alone.

The biggest reason for this resistance is the belief (which is correct, I have to say) that big data center projects that require gigawatt-scale power draws raise prices in the region where they are constructed, making large training clusters a PR mess.

Predicting these growing pressures, most Hyperscalers sought investment and land in the Middle East to displace their training clusters out of the US.

A crucial point to understand here is that large data centers are only required for training, as large training runs require hundreds of thousands of tightly connected GPUs. Furthermore, training is not latency-sensitive, so compute doesn’t have to be close to the user, so you can place them elsewhere, even outside America.

On the other hand, inference can’t be offloaded to the Middle East because latency is crucial, meaning these data centers have to be built onshore.

However, inference clusters can be deployed at very low megawatt-scale sizes and thus much more distributed across the entire country because inference workloads follow the opposite scaling law: the fewer GPUs per workload, the better.

Sadly, the Iran conflict puts this entire strategy at risk if training clusters can’t be built there, and coupled with onshore resistance, could put data center buildout progress in serious risk.

Energy prices are also lower in the Gulf, but this isn’t a major factor considering that energy costs of training are a tiny percentage of the total cost of ownership of AI training clusters, which are dominated by capital costs, as we’ll see later.

A third leg is financial; some data centers are being delayed simply because the money to pay for them is missing, with examples such as Oracle/OpenAI’s Abilene, Texas data center expansion being canceled. And although this may seem unrelated to today’s discussion, it’s actually relevant to the oil debate due to the growing importance of debt, as we’ll learn later.

Knowing all this, let’s finally quantify and model the impact of this war on AI and your related investments.

Let’s start by stating the obvious. The most direct way this war affects AI is in markets. An already volatile, fear-driven market before the outbreak of the war, as this market is largely driven by narratives of promised future revenues rather than real, hard evidence, markets are going to massively overreact to the slightest concerns.

Most investors think an eventual, sizeable crash of stock markets is long overdue, and many will bet that this is the time it materializes.

Perhaps no market is more exposed to and volatile from this war than Korea, due to the obvious geopolitical risk; shortages of Gulf oil directly affect South Korea, one of the key importers of that oil, as mentioned earlier.

Added to the debatable tendency of Korean investors to operate on margin (using debt to increase their stock exposure, putting them at risk of getting liquidated by lenders if stocks fall below certain thresholds), has turned the Kospi index into the most volatile market by far in the world right now, severely impacting two stocks that have been instrumental to our AI investment thesis: Samsung and Sk Hynix.

Both are still incredibly profitable businesses and will remain so, but their geopolitical exposure to this conflict is not small.

For Big Tech, whose revenues are, to this day, largely uncorrelated with AI ROI (return on investment), except for NVIDIA, their risks are also mostly about being so large that they cannot be affected in some way, even if it’s simply being dragged down by index selloffs (i.e., people owning Vanguard funds or ETFs and thus causing overall market selloffs).

Furthermore, Hyperscalers have to consider the risk of becoming targets in the area, as we have explained with Amazon’s case, but their exposure doesn’t seem to be large either.

The biggest question here is how the inevitable oil price impact AI infrastructure buildouts and consumption. To do that, I’ve built a financial model that estimates impact through scenario analysis of several key variables.

In particular, the model analyzes the impact of oil in AI across three factors:

And how do we quantify each? And more importantly, should you do something about it?

For our estimations, we will be working with a base facility load of 1 GW, using the industry assumption popularized by Jensen Huang that every gigawatt of IT-critical load requires around $50 billion in total capital investment.

Additionally, we are adding an additional 6% CapEx growth estimated by JLL, which we’ll use to factor in supply shortages in labor, construction materials, and LNG price hikes, among others, giving us a total base facility CapEx expenditure of $53 billion. In layman’s terms, the increase costs caused by oil price hikes will be applied over this value.

In an effort to estimate cumulative price increases throughout 2030 (this is one hell of a stretch, though), we are expecting up to 100 GW in new IT critical load deployed over the next five years, at an average of 20 GW per year, which will allow us to estimate the cumulative costs over a five-year period.

And without further ado, let the nightmare begin. Besides scaring you a bit, this model is going to teach you a hell lot of how intertwined everything really is, even in ways I myself did not predict until I did the numbers myself.