West Asia war has exposed the fragility of our trillion-dollar AI investment

Depending on how prolonged and intense the war is, there will be second- and third-order impacts on AI development and adoption. This may broadly take three forms.

First, an immediate effect already being felt and one that will continue to worsen in the next few weeks is short-term supply chain disruptions due to increased uncertainty and higher sourcing costs for input materials.

Consider helium. Its uses range from magnetic resonance imaging machines to semiconductor manufacturing, thanks to its cryogenic cooling properties, which are essential for maintaining the ultra-low temperatures required in semiconductor fabrication. Its supply chains are fairly concentrated, dominated by the US, Russia and Qatar, and the gas is also too light to be retained by gravity, making its stockpiling expensive and technically demanding.

However, the Iran war is unlikely to trigger a helium crisis. The element is one input in a complex supply chain, and high-tech industries command enough purchasing power to secure supply at elevated prices and from other sources, even as other buyers get squeezed.

Rising energy prices, however, are likely to have more significant ramifications. South Korean firms like Samsung Electronics and SK Hynix account for nearly 80 per cent of the global high-bandwidth memory (HBM) and 70 per cent of the dynamic random access memory (DRAM) market.

While they account for around 40 per cent of South Korea’s stock market cap, the country is heavily dependent on fossil fuels for its energy needs, with 70 per cent imported from West Asia through the Strait of Hormuz. A similar story plays out in Taiwan, where 90 per cent of the world’s most advanced semiconductors are manufactured.

The current supply shortage of memory chips is not a product of a war so much as a market distortion that the war has exposed. Over the past year, the world’s three largest memory chip manufacturers, which account for over 95 per cent of the DRAM market, have systematically shifted production to HBM, driven by record-high investments in AI infrastructure and higher profit margins in the category compared to more commoditised DRAMs.

Meanwhile, DRAM prices had already surged multifold before the war, with major electronics OEMs (Original Equipment Manufacturers) warning of consumer price hikes by 15-20 per cent. Reportedly, the memory market has become a seller’s market with the three manufacturers rejecting long-term agreements and preferring quarterly contracts in anticipation of stepwise price escalations.

The logistical costs, high insurance premiums, and geopolitical risks triggered by the Iran war only further complicate the situation. Because increasing supply capacity takes years, the price upswing in DRAMs is now expected to take longer to stabilise.

Second, should this war drag on, there may be implications for the attractiveness of West Asia as a location for data centre capacity buildout. The region has so far been considered lucrative due to its extensive land availability, inexpensive energy, government support, favourable geography relative to other emerging markets and access to capital from sovereign wealth funds. Saudi Arabia and Dubai, having signed access-for-investment deals with the US, also benefit from geopolitics.

However, the attack on AWS data centres in the region, the first such confirmed military attack in history, with the IRGC claiming these were “legitimate targets” since AWS data centres were used for military applications, has set a precedent for a new category of infrastructure in need of military protection from kinetic strikes, in addition to cyber threats.

Given the high costs, including long-term power contracts, land agreements, and fibre connectivity, invested in already operational facilities, it’s unlikely that hyperscalers will relocate built capacity. Moving out existing AI workloads would also be difficult in light of data localisation rules. But the marginal investment decision for new capacity rollouts would factor in this added geopolitical risk, alongside other location-evaluation metrics such as tax incentives and power-grid stability.

Third, in the scenario of a prolonged war including the Gulf countries participating, the world may be headed for stagflation pressures like during the 1980s and 1990s: High energy prices contributing to global inflation while economic activity stagnates. This is likely to have an impact on the profitability of frontier AI labs and hyperscalers.

Just this year, five American tech giants are expected to make $700 billion in capital expenditures on AI infrastructure. The scale of the investments by US companies exceeds that of momentous past capital expenditures, such as the Manhattan Project, the moon landing, or the broadband buildout. Returns on AI investments now need to price in greater geopolitical and market concentration risks.

This is likely to stretch the timelines between investment and return, even leading to valuation compressions for AI companies and concentrating market power in favour of companies with deeper pockets. Whether AI adoption timelines will be compressed for productivity gains or stretched due to a declining global GDP growth rate remains to be seen.

Ultimately, one of the spillover effects of this war, should it be long-term, may be that the impact of generative AI on the workforce and the broader economy, unfolding simultaneously with the effects of the war, is difficult to mitigate. On the heels of the aftereffects of a global pandemic and supply chain readjustments forced by the tariff wars, the war in Iran might serve as an additional confounding variable, making it harder to gauge AI’s true impact.

This will have implications for the technology’s governance and policy responses.

Shobhankita Reddy and Bharath Reddy are researchers in technology geopolitics at the Takshashila Institution, Bangalore. Views are personal