Doomberg on why oil is still cheap despite the Strait of Hormuz closure and what it means for AI data centers

Doomberg on cheap oil, natural gas, and AI data centers

Doomberg is an anonymous team of former industry executives who write about energy markets under a pseudonym — nearly 400,000 email subscribers on Substack, built over five years with a deliberately brand-conscious approach. Their core audience holds Bloomberg terminals. The green chicken is not an accident.

Why oil is still cheap

The Strait of Hormuz has been closed for roughly 70 days, and oil is still trading well below where any serious energy analyst would have predicted. Doomberg's best working theory involves China. Throughout 2023–2024, China was quietly buying sanctioned and shadow-fleet oil at scale — Doomberg estimates 1.5 to 2.5 million barrels per day above normal demand — and is now bleeding those reserves into the market to keep a lid on prices. The geopolitical leverage is deliberate: China is cutting refined fuel deals with Australia, positioning itself as the stable adult in the room while Trump posts on Truth Social.

The IEA released 400 million barrels from strategic reserves shortly after the Strait closed, which helped. But Doomberg is direct that when you run the math, prices should still be higher. A full accounting will have to wait for the after-action report.

“If there's no funding, there's no big computer. The AI revolution is powered by coal in China and natural gas in The US. And natural gas in The US has been made cheaper by this war for reasons that we can explain. Natural gas is so cheap and abundant — so then, okay, what's the constraint downstream from being able to produce electricity from natural gas? That's gas turbines sold out.”

The coproduction dynamic that matters for AI

For the US AI build-out, the more important story is natural gas, and here the mechanism is counterintuitive. The shale revolution twinned oil and gas production in the same wells, particularly in the Permian Basin. When oil prices spike and drilling accelerates, natural gas floods out as a byproduct. The Permian currently has negative spot prices for natural gas — drillers are giving it away to get to the oil.

Henry Hub is trading around $3 per million BTU, roughly equivalent to $18 per barrel of oil. That is the fuel powering US AI data centers, and the Strait closure has made it cheaper, not more expensive. Coal in China is similarly insulated from oil markets. The AI compute build-out's energy costs are, at least for now, structurally protected from the Middle East conflict.

Off-grid is the destination

Cheap fuel is not the binding constraint. Gas turbines are sold out. Grid connection queues are measured in years. Doomberg argues the US grid's operating model is structurally incompatible with the pace Silicon Valley wants to move, and most hyperscalers — except Elon Musk, who built a private natural gas plant for xAI in Tennessee by bypassing the normal permitting process — are not going to break rules to get there faster.

The practical model Doomberg envisions is buildings where natural gas goes in one end and data comes out the other, fully off-grid with a grid connection kept only for backup. Old coal plant sites in Appalachia are attractive for exactly this reason: the transmission infrastructure already exists, local natural gas from nearby shale patches is abundant, and there is no queue to get connected.

Cold geography is an underrated variable. Roughly half the energy consumed in a data center goes to cooling, which makes British Columbia, Alaska, and Iceland structurally cheaper to operate than Texas — assuming the permitting bureaucracy cooperates.

Nuclear and solar

Doomberg dismisses fusion as a solution to problems that don't exist. The case that nuclear waste and meltdown risk are unsolvable is weak given current AP1000 and CANDU designs. The actual constraint is that building what already works is not a fundable Wall Street story. For venture capital, Doomberg thinks there are better uses of the money.

On solar, the core argument is that intermittency costs are systematically misallocated. Every other technology that has to stand ready when solar drops off the grid absorbs costs that are never attributed to solar itself. Beyond a certain penetration threshold, grid stability starts to break, and the accounting still doesn't reflect the full cost.

The China-US summit as the white pill

Doomberg's more optimistic read centers on the Trump-Beijing summit. A presidential visit of this profile, with a reported delegation of major US CEOs, does not happen without substantive deals already negotiated in advance. Doomberg is speculating, but the inference is that a broader settlement — potentially touching Iran and Ukraine — is in play. A diplomatic resolution that reopens the Strait would remove the one scenario where energy costs genuinely threaten the US data center build-out.

The base case for AI infrastructure remains intact: natural gas cheap, coal cheap, grid workarounds increasingly well understood. The risk is a prolonged conflict that eventually overwhelms the coproduction buffer and forces the IEA reserve math to catch up with reality.