AI · Web3 · Tech trends and insights at a glance
AI · Web3 · Tech trends and insights at a glance
Renewed tension in the Strait of Hormuz is easy to file under "oil prices wobble again." But that framing misses how the explosive electricity demand of AI data centers has quietly retethered the digital economy to fossil-fuel chokepoints. This column traces the mechanism, and why the nuclear-and-renewables transition arrives too late to break the chain this decade.
The Strait of Hormuz is flashing red again. Oman's move to impose what amounts to a transit toll drew sharp objections from Iran, a commercial vessel was struck, and the International Maritime Organization paused an evacuation operation amid the rising risk. Markets did what they always do: oil ticked up, analysts dusted off their chokepoint explainers, and the story was filed under the familiar heading of Middle East volatility. That filing is a mistake. Read only as an oil-price headline, the episode hides the most consequential thing about it — that the fastest-growing force in the global economy, artificial intelligence, is now lashed to this narrow waterway in ways most of its boosters never mention.
For the past two years the real bottleneck in AI has not been silicon but power. Training and serving frontier models is no longer principally a question of buying enough GPUs; it is a question of where to find cheap, firm electricity to run them around the clock. A single large AI campus now draws as much power as a mid-sized city, and the hyperscalers are racing to lock up generation capacity, grid interconnections, and even dedicated plants. The trouble is that new supply cannot scale anywhere near as fast as demand. When demand outruns the available clean and baseload capacity, the marginal megawatt — the last unit needed to keep the cluster humming — still comes overwhelmingly from natural gas and oil.
This is where Hormuz reenters the picture. Roughly a fifth of the world's seaborne crude, along with a decisive share of the LNG that Qatar and its neighbors ship, passes through this strait. A blockade scare moves oil first, but the shock propagates quickly into gas prices and then into wholesale electricity. So long as AI infrastructure is exposed to the marginal price of fossil-fired power, a military standoff in the Persian Gulf is no longer the exclusive concern of refiners and airlines. It seeps, quietly, into the operating cost of GPU fleets and ultimately into the price of every inference token. AI looks like the emblem of a weightless digital economy, yet its physical foundation still trembles whenever a single tanker stops moving.
The standard answer is that nuclear and renewables will sever this dependency. And indeed the hyperscalers have signed small modular reactor deals, struck enormous solar and wind purchase agreements, and even contracted to restart shuttered nuclear plants. The direction is right. But being right about direction is entirely different from arriving on time. Small modular reactors face years of licensing and construction; intermittent renewables cannot carry a data center's twenty-four-hour baseload without massive storage and grid reinforcement that do not yet exist at scale. AI demand is exploding now, while the work of swapping that demand onto geopolitically insulated clean power will not be finished for five to ten years. That gap is the dangerous stretch.
What fills the gap is gas, and so AI infrastructure remains, for at least this decade, a hostage of energy geopolitics. The paradox is that the faster the AI industry grows, the deeper its near-term reliance on fossil fuels becomes, and the more exposed it is to the volatility of chokepoints like Hormuz. This is why data center siting strategy increasingly tracks, with almost embarrassing transparency, wherever cheap and stable power happens to sit — gas-rich West Texas, subsidized Gulf grids, the hydro and cold climates of the Nordics. The map of AI is being redrawn along the contours of power geopolitics, not model architectures. The Hormuz alarm, then, is not a regional security story but a floodlight on the most fragile seam of the AI era. While attention fixes on parameter counts and benchmark scores, the electrons that animate those models are still tied to a narrow strait, a tanker, and the political choices of petrostates. The decisive advantage will belong not to whoever builds the largest model, but to whoever breaks this chain first.
The Land-Permit Paradox of Korea's Chip Belt, When the Cluster's Boom Prices Out Its Own Engineers
Dongtan, Giheung, and Guri have been folded into Korea's land-transaction permit regime just as the AI chip capex boom reshapes the property market around the country's largest fabs. The very prosperity the cluster generates is raising the cost for the engineers it depends on to settle nearby. The real test of agglomeration may lie not in siting megafabs but in housing and labor mobility.
The Collapse of the Closed AI Moat and the Supply-Chain Paradox of Unverifiable Weights
DeepSeek-R1's open reasoning weights and Llamafile's single-file distribution are eroding the performance and distribution moats that closed labs once charged a premium for. Yet the same openness collides head-on with the gap exposed by the "250 samples to break an LLM" research: weight distribution that no recipient can verify. Democratized competition and accumulated security debt now sit on the same scale.
Forty-Year Yen Lows as the Hidden Subsidy Behind Japan's Chip Revival
As the yen slides into its weakest territory in four decades, Takaichinomics has entered uncharted monetary terrain. A cheap yen functions as a silent subsidy for Rapidus, Kioxia, and TSMC's Kumamoto fabs—yet the same currency inflates the cost of imported tools and materials and intensifies the talent war with Korea. The question is whether monetary policy can stand in for industrial policy, and what that means for Korea's memory champions.