Key Takeaways

  • Electricity travels at near light speed, requiring supply and demand to match continuously because almost all power is consumed at the exact moment it is generated.
  • Tokyo requires a baseload of roughly 20,000 MW just to keep normal life running, leaving an 18,000 MW gap of dynamic load that grid operators scramble to fill during peak hours.
  • Nuclear plants can stop a fission chain reaction in seconds, but reactor steam keeps turbines spinning for hours, locking those assets into continuous baseload economics.
  • Flexible consumers like Bitcoin miner Riot earn revenue by curtailing electricity use during price surges, but AI data centers face severe chip depreciation schedules that force them to consume baseload power around the clock.

The Physics Problem Behind Power Markets

Software engineers take asynchronous queues for granted. If an API receives ten thousand requests per second during a product launch, it buffers the traffic, holds jobs in Redis or Kafka, and burns through the backlog over the next twenty minutes. The electrical grid has no buffer.

As Patrick McKenzie points out, “the supermajority of all electricity is consumed approximately contemporaneously with being generated.” Electrons move through copper and aluminum conductors at a significant fraction of the speed of light. Meanwhile, mechanical generation operates on human time. McKenzie notes: “The physical processes required to generate electricity frequently require hours of time to spin up or spin down. These are fundamentally incompatible time scales, and yet we somehow make it work anyway.”

Consider a nuclear plant. Operators can drop control rods into the core to halt the chain reaction almost instantly. Yet that does not mean the power generation stops immediately. As McKenzie notes, “You can turn off a nuclear reaction in an order of seconds, but the steam produced by your reactor, which turns the turbines to generate power, will continue being hot due to its own energy and remaining energy in the system for a few hours.” Because these plants cannot stop on a dime, they form the baseload: the continuous floor of electricity that never turns off.

Dynamic Load and the Peaker Squeeze

Baseload keeps hospitals lit and refrigerators cold at 3:00 AM. In a massive metropolitan center like Tokyo, the baseline energy required to prevent society from falling apart is immense. “If you draw a line around 20,000 MW, you will see that Tokyo has a certain level of work per second required to sustain civilization,” McKenzie explains. “That is called baseload. The 18,000 MW gap between that and peak capacity is dynamic load.”

Dynamic load fluctuates wildly based on the time of day and the weather. When temperatures spike on a Tuesday afternoon in July, millions of office air conditioners turn on at once. Baseload plants cannot spin up fast enough to absorb that 18,000 MW surge.

To bridge the gap, grids turn to peaker plants. These units, usually running on natural gas, sit cold and empty for thousands of hours every year. McKenzie observes: “Finally, there are peaker plants, which spend most of their time off and economically justify themselves by keeping society running for a few hours every day and, in some cases, only a few weeks out of every year.” Because they operate so rarely, they charge extortionate spot rates to pay off their capital costs.

This dynamic creates an opportunity for elastic industrial loads. Bitcoin miners like Riot built a business model around demand response: when peaker prices spike to four digits per megawatt-hour, they cut power instantly and sell their grid contracts back to the operator for massive profits.

AI labs cannot replicate this trick. An H100 GPU depreciates heavily every hour it sits on a server rack. If an AI cluster powers down when grid stress arrives, the saved electricity bill rarely offsets the crushing depreciation loss of idle silicon. As a result, AI data centers enter the market as captive baseload buyers, locked into bidding wars for continuous, unyielding power contracts.

What to Do With This

Audit your cloud or compute infrastructure contract this week to identify whether you are paying for continuous baseload availability or interruptible capacity. If you run batch training, testing pipelines, or background analytics jobs, move those workloads to spot instances or flexible compute tiers that clear at off-peak rates. Do not pay continuous baseload prices for computational work that can afford to wait three hours.