How AI could impact geopolitics

Episode Summary

Executive Summary: Jared Cohen argues that AI’s biggest near-term constraint is not model quality or funding, but where the needed power- and chip-intensive infrastructure will be built. He says the U.S. can keep AI leadership only by using overseas capacity, with the Middle East emerging as the most practical overflow option despite long-term geopolitical risks and China’s competing ambitions.

Main Topics: AI infrastructure as a geopolitical issue (Priority: 5/5): Cohen frames AI not just as a technology or market story, but as a race shaped by energy availability, data center siting, chip access, and state power. U.S. power and data center constraints (Priority: 5/5): The U.S. faces low vacancy in existing data centers, mismatch between cloud and AI workloads, and pressure on the electric grid, making domestic buildout insufficient on its own. Three overflow regions for AI buildout (Priority: 5/5): Cohen evaluates democratic allies, the Global South, and the Middle East as potential locations for AI infrastructure, weighing infrastructure readiness against geopolitical risks. Middle East as the short-term favorite (Priority: 4/5): Saudi Arabia, Qatar, and the UAE combine cheap energy, land, capital, speed of execution, and coastal access, making them strong candidates for AI infrastructure expansion. China’s AI response and competition strategy (Priority: 4/5): China is pushing efficiency gains, data-center expansion, nuclear and clean-energy investment, and global partnerships to avoid falling behind in generative AI. U.S.-China policy hardening under Trump and Biden (Priority: 4/5): Cohen argues U.S. policy toward China has become consistently tougher across administrations, with tariffs and export controls likely to intensify further.

Key Arguments: The urgent question is not whether AI will change industries, but whether the U.S. can maintain leadership if AI software needs AI hardware and infrastructure beyond domestic capacity. Funding is not the main bottleneck; physical space, specialized data centers, and baseload power are the real constraints on U.S. AI expansion. Existing U.S. data centers are largely full and often built for cloud workloads, which cannot easily be retrofitted for ultra-high-density AI demand. AI workloads require concentrated, largely baseload electricity, making intermittent renewables insufficient as a primary solution. The U.S. will need roughly 35+ gigawatts of additional power capacity and an overflow location for AI infrastructure within 12-18 months. Democratic allies are geopolitically comfortable but may be too slow or politically constrained to deliver the needed scale quickly. Global South locations may offer cheap power, but AI capacity there could be diverted toward China, undermining U.S. strategic goals. The Middle East is the most practical short-term option because it has cheap energy, land, capital, speed, and willingness to align with U.S. chip requirements. Saudi Arabia, Qatar, and the UAE can leverage AI infrastructure to strengthen their geopolitical autonomy, but their long-term alignment is not guaranteed. China remains determined to compete through model efficiency, nuclear and clean-energy investment, and global data-center partnerships. U.S. policy toward China is increasingly bipartisan and likely to become even more protectionist, including expanded tariffs that could affect AI supply chains.

Data Points: Hyperscaler AI CapEx: roughly $1.1 trillion on the high end - Projected spending by hyperscalers to meet AI demand Global data centers: roughly 8,000 worldwide - Size of the global data center base U.S. data centers: roughly 3,000 - Share of global data centers located in the United States Data center vacancy rate: less than 3% - Indicates tight capacity in existing U.S. data centers Additional power needed: 35+ gigawatts - Estimated new baseload power required in the U.S. for AI demand Chinese clean energy investment share: about a third of global clean energy investments - China’s role in global energy investment China’s AI infrastructure investment: $6.1 billion - Investment in creating data center hubs around the world

Pivotal Quotes: "if AI software has to run on AI hardware somewhere, can the U.S. maintain a leadership position when it comes to generative AI?" — Jared Cohen: Defines the central strategic question of AI infrastructure leadership "the U.S. can maintain that leadership position. The infrastructure can keep up with demand, but not exclusively in the U.S." — Jared Cohen: Cohen’s core thesis on the need for overseas capacity "The third option, the Middle East. Middle East, probably more than any other countries, has probably the best attributes to accommodate all this." — Jared Cohen: His preferred short-term overflow region for AI buildout

Implications: AI leadership may hinge on international energy and infrastructure diplomacy, not just chip innovation. U.S. firms and policymakers will likely deepen ties with the Middle East while managing China risk, supply-chain exposure, and grid constraints at home.