The Power Beneath the Power

Yesterday I wrote about how the Pentagon subordinated its command infrastructure to Google’s operational tempo. Today we’re looking at the layer beneath that: the power generation crisis that Google itself depends on. When a supersonic plane startup pivots to building turbines because AI datacenters can’t get electricity fast enough, you’re not watching innovation—you’re watching civilizational triage.

Yesterday the Pentagon outsourced its operational tempo to Google. A supersonic airplane startup also announced it’s building natural-gas power plants to keep AI datacenters alive.

The through-line isn’t “innovation.” It’s that every layer of American infrastructure has lost control of its own clock.

Blake Scholl, CEO of Boom Supersonic, just dropped a 13-tweet thread that reads like a business announcement but functions as an infrastructure reconnaissance report. The headline: Boom is launching “Superpower,” a 42MW natural gas turbine optimized for AI datacenters, with a 1.21GW order from Crusoe AI already in hand.

The framing is entrepreneurial opportunity. The subtext is civilizational triage.

Scholl describes the problem plainly: “Our grid has flatlined while China has raced ahead. Hyperscalers have racks of GPUs sitting idle with no electricity to power them.” Then he explains the pivot: Boom was developing supersonic engines for aircraft, but “we decided it was time to pull energy forward in our product roadmap.” Translation: We can’t build the plane yet because the country can’t generate enough electricity for AI compute, so we’re selling power plants first.

This isn’t a pivot. It’s a symptom.

We’ve entered the phase where AI demand is outpacing grid capacity so quickly that aerospace engineers are repurposing Mach-1.7 hardware to keep transformers fed. Not metaphorically. Literally.

Here’s the timescale stack that matters:

Grid construction: 10–20 years
Transmission permits: 4–11 years (recently “streamlined” to a 2-year target that has yet to be achieved)
Turbine manufacturing: 2–3 years
AI datacenter expansion: quarterly
Model training cycles: 30–90 days
Benchmark races: weekly
VC fund pressure: 18–24 months

The scale of the mismatch is staggering. The IEA projects that global datacenter electricity consumption will double from 415 TWh in 2024 to 945 TWh by 2030—with AI identified as the primary driver. In the U.S. alone, datacenter demand could reach 100 gigawatts by 2030, adding massive load to a grid that saw flat growth for the previous 20 years.

To put that in perspective: 60 gigawatts of new datacenter demand is roughly equivalent to Italy’s entire peak hourly power consumption. And the grid moves at geological speed—transmission line projects average 10+ years from conception to completion, with one Idaho project still awaiting permits after 14 years.

The slowest layer—physical infrastructure—is now taking tempo from the fastest layer—AI development.

When acceleration outruns the substrate, you don’t get progress. You get panic engineering.

Boom’s turbine isn’t solving the energy crisis. It’s working around the fact that America can no longer build infrastructure at the speeds its technology sector demands. The grid can’t scale fast enough, so private companies are building their own generation capacity off-grid, outside the civic timeline, on timescales that match quarterly datacenter expansion plans instead of decade-long transmission projects.

This is Δt collapse at the industrial layer.

AI power plants are not innovation. They’re triage.

The “solution” being implemented is to bypass the grid entirely with private generation. Hyperscalers and AI labs can’t wait for transmission line permits, so they’re contracting with whoever can deliver electrons directly to the datacenter perimeter. Boom. Crusoe. Anyone who can spin up turbines faster than the grid can approve new substations.

This creates a parallel infrastructure system operating on a fundamentally different temporal regime than public utilities. The grid moves at regulatory speed. Private power moves at commercial speed. And increasingly, the AI stack depends on the latter because the former simply cannot keep pace.

When industry builds the grid faster than the grid can build itself, you no longer have a public infrastructure system. You have parallel fiefdoms with their own clocks.

The deeper problem is that this parallel system doesn’t just supplement the grid—it subordinates national energy priorities to compute availability. “Idle GPUs” are now treated as a crisis of state, not a business problem. Turbines are being prioritized over transmission lines. Power generation is being reorganized around inference workloads rather than civic needs.

The costs are already visible. Wholesale electricity prices have spiked dramatically near datacenter hubs—up 267% in some markets since 2020. Consumers served by PJM Interconnection will pay $16.6 billion to secure power supplies for datacenter demand from 2025 through 2027. Mark Christie, former FERC chairman, put it plainly: “The reliability crisis is here now; it’s not off in the distance somewhere.”

AI is now dictating national energy policy de facto, not de jure. And the policy is: whoever can deliver power to GPUs fastest wins, regardless of whether that creates a coherent or sustainable energy system.

This is not an energy strategy. It’s a civilization reacting to a tempo it no longer owns.

The symmetry with GenAI.mil is precise.

Yesterday I wrote about how the Pentagon subordinated its command tempo to Google’s operational infrastructure. Google Cloud sets the maintenance windows, the uptime guarantees, the API availability. The DoD’s decision-making capacity is now downstream of a commercial vendor’s quarterly priorities.

But Google’s operational tempo is itself subordinated. Google can’t run inference if the datacenters don’t have power. And increasingly, datacenter power availability is governed not by the grid, not by utilities, but by whoever can spin up turbines fastest—private generation vendors operating on commercial timescales.

The dependency stack looks like this:

Pentagon command decisions → depend on
Google Cloud availability → depends on
Datacenter compute capacity → depends on
Electricity generation → depends on
Whoever can build turbines faster than the grid can build transmission lines

At each transition, the timescale mismatch gets worse. The Pentagon thinks in doctrine cycles. Google thinks in quarters. GPU deployment thinks in weeks. Power demand thinks in days. The grid thinks in decades.

And now we have supersonic airplane startups becoming emergency power vendors because the grid can’t keep up.

Yesterday, the DoD subordinated to the cloud.
Today, the cloud subordinated to the grid.
Tomorrow, the grid will subordinate to whoever builds the turbines.

The state is downstream of all of it.

Scholl’s thread uses all the standard rhetoric. “America has fallen behind.” “China has raced ahead.” “We’re not sitting idly by.” It’s the same nationalist framing as GenAI.mil: crisis → urgency → proprietary solution → patriotic varnish.

But strip the rhetoric and you’re left with a confession: America cannot build power infrastructure faster than its AI sector demands it. The grid is too slow. Permitting is too slow. Transmission is too slow. So private companies are building parallel systems that bypass civic timescales entirely.

This is what it looks like when a civilization loses the ability to operate at its own infrastructure layer. You don’t get dramatic collapse. You get improvisation. Workarounds. Parallel systems built in backrooms by whoever can move fast enough to matter.

The Pentagon outsources command to Google. Google outsources power to Boom. Boom pivots from supersonic flight to turbine manufacturing because that’s where the acute need is. And everyone narrates it as innovation rather than admission that the foundational systems don’t work anymore.

The through-line is Δt collapse at every layer. Command, compute, power—all subordinated to faster external clocks, all losing coherence, all narrating the loss as modernization.

A nation that can’t build power lines in less than a decade ends up building civilization out of workarounds. The future won’t be lost in a war or a disaster—it will die in the compilation errors between mismatched clocks.

The power layer is now improvisational. That should worry us more than “AI risk.”

Every day, the gap widens.