The Compute We Cannot Power

On September 9, Oracle reported a cloud backlog of $664 billion — up $209 billion from a year earlier, an order book the size of a mid-cap national economy on hand for delivery over the next several years. That same week, PJM Interconnection, the grid operator serving 65 million Americans across thirteen states, was still processing a December 2025 capacity auction it had cleared 6,625 megawatts short of its reliability target — the first shortfall in the operator's history — at a record clearing price of $333.44 per megawatt-day. The order book has grown faster than the grid can keep up with.

This is not a permitting problem or a temporary shortfall. American companies are contracting artificial-intelligence compute faster than the country can build the grid to power it, and the mismatch is arriving inside a coordination architecture that has assigned pieces of the problem to six regional grid operators, fifty state public utility commissions, and a small set of federal agencies whose authorities do not overlap on the composite. In June, the Federal Energy Regulatory Commission ordered each of the six regional operators to write its own rules within sixty days. Commissioner David LaCerte warned publicly that FERC would “dictate solutions” if the operators cannot converge. Convergence is not on the calendar. No entity above the assigned authorities has been named to arbitrate the composite-against-composite demand.

The critical-minerals track shows the same asymmetry in miniature. On July 10, 2025, the Department of Defense used Defense Production Act Section 303 authorities to take a $400 million equity stake in MP Materials and to guarantee a ten-year price floor for neodymium-praseodymium at roughly double the current market price. One company, one mineral, one price floor. On November 10, China’s extraterritorial rare-earth licensing regime returns under Announcement No. 61 — the 0.1-percent rule — covering rare earths, magnets, and downstream technology inputs across automotive, aerospace, semiconductor, and artificial-intelligence supply chains. The International Energy Agency estimates full implementation would place $6.5 trillion of downstream production outside China at risk. Beijing’s move is compound, systemic, and licensed centrally. The American answer is per-vendor, per-input, and negotiated bilaterally.

The Middle East corridor added the third register this week. A major east-west Saudi pipeline was forced offline. Transit disruptions extended simultaneously across the Strait of Hormuz and the Bab el-Mandeb. American diesel crossed $6 per gallon at retail, and the transmission landed directly in the August Producer Price Index at 5.4 percent — a tenth above consensus, driven by a 24.1 percent monthly jump in diesel. For the first time this year, the corridor shifted from a regional flashpoint to a domestic macroeconomic input. Data centers need power; power needs fuel; the corridor is where the fuel comes from. Every hyperscaler private-generation contract and every backup-power assumption in a 2027 AI capital plan now runs against a diesel price on the goods side of the American inflation print.

The Pentagon’s autonomy doctrine sits on the same footing. Drone swarms, autonomous undersea platforms, and the tactical-edge inference the Joint Fires Network is being procured to deliver each presume a compute supply chain that runs from mineral to chip to installed platform at operational speed. Enterprise capital plans priced against frontier compute delivery presume the same. Sovereign artificial-intelligence partnerships with Korea, Japan, and the Gulf presume the same. If those inputs cannot be delivered at the pace assumed by the contracts, none of those presumptions hold.

The three-week arc through this diagnosis now points to a single operational conclusion. The Army America is buying cannot be commanded on the doctrine as written. The artificial intelligence it is buying cannot be deployed on the mission as scoped. The compute at the head of the growth forecast cannot be powered on the grid the country has built. Beijing is running the integration architecture the United States has not written.

Picture the October 2026 planning session. A hyperscaler chief financial officer is closing the 2027 capital plan against compute-delivery schedules her contracts already fixed. Her power-procurement lead tells her that large-transformer lead times have moved from 24 to 36 to 48 months. Her regulatory counsel tells her the six regional show-cause returns FERC ordered in June come due between sixty and one hundred and fifty days, and whether the rule that clears her interconnection comes from a converged regional set or a FERC-imposed replacement is unresolved. She calls the Department of Energy. She calls FERC. She calls her state public utility commission. Three offices; three pieces of the answer; no single voice authorized to warrant the composite. She signs the plan against megawatts no institution has vouched for.

That is the pattern of every arbiter she reached. DPA Section 303 was built for targeted intervention in specific supply chains, not for compound systemic response. FERC sits above the regional grid operators but not above the states. The Department of Energy directed FERC to write rules in October 2025 and got six show-cause orders in return. Nothing in the current apparatus assigns integration authority. The signal to watch is not another Oracle print or another PJM auction. It is whether any American institution names an accountable integrator above the assigned authorities — one office empowered to arbitrate grid, minerals, and integrated federal decisions in response to composite demand.

If you are running defense industrial-base capital, hyperscaler capital, or duration exposure priced against the AI cycle, do the work now. For your highest-conviction 2027 plan, name the specific power source, the specific mineral supply chain, and the specific federal decision whose failure would break the plan. Then name the person or institution accountable for delivering each. Where the second name is missing, the risk in the first has not been sized.

The last three weeks have named the same problem three times — the Army we cannot command, the AI we cannot deploy, the compute we cannot power. Each name points at the same missing office. The next capability America fields will not fail because it never arrived. It will fail because it arrived on inputs no one had been assigned to deliver.

Richard Berry is the founder of Stratnova Advisors and the editor of Strategic Horizons, a weekly geopolitical assessment for senior executives and national-security professionals. His forthcoming academic work examines the supervisory function in human-AI decision architectures.

The Cipher Brief is committed to publishing a range of perspectives on national security issues submitted by deeply experienced national security professionals. Opinions expressed are those of the author and do not represent the views or opinions of The Cipher Brief.

Have a perspective to share based on your experience in the national security field? Send it to Editor@thecipherbrief.com for publication consideration.

Read more expert-driven national security insights, perspective and analysis in The Cipher Brief