Thirty-five gigawatts. The number is loose in the wild now — crypto Twitter, tech newsletters, at least one utility earnings call — a project pipeline so vast it would, on paper, cover every AI data center currently planned on American soil. There is a catch. Nobody has told you what the number means.
In-service capacity? Under construction? Signed power purchase agreements? Does it include storage, inverters, or nothing but land options? A backlog without a timestamp is not a backlog. It is a rumor with a spreadsheet. After eleven years reading smart contracts line by line, I have learned the most dangerous document is never the one missing a signature — it is the one missing a definition.
NextEra Energy runs the largest wind and solar fleet in the United States. The company holds roughly 35 GW of what it calls project backlog, and the story writes itself fast: hyperscalers want carbon-free electricity, NextEra builds carbon-free electricity, the two meet at a substation somewhere in Texas. The thesis reached crypto through vertical media — a Crypto Briefing item that compressed a dense electricity story into a single sentence and moved on. One sentence is all the substance on offer. No timestamp. No split between in-service and contracted. No primary source.
That compression is the trap. When a reporter without a utility background covers a power company, the output reads like a token whitepaper — bold claims, zero methodology. And a crypto audience, trained by a decade of unaudited tokenomics to accept the headline and skip the appendix, swallowed it whole. I have run that experiment from the other side. In 2017 I tore apart forty-plus ICO contracts in a weekend and found reentrancy bugs nobody had bothered to look for, because the only number anyone cared about was the raise. The 35 GW figure deserves identical suspicion. The difference: a buggy contract drains a wallet; a fictitious megawatt drains a grid. Code is law, but audits are mercy.
Start with physics, because narrative does not override it. A data center is a brutal customer — load factor above ninety percent, near-full draw around the clock, zero tolerance for interruption. An outage is not an inconvenience; it is a breached SLA. Wind and solar are intermittent by construction. Lithium-iron-phosphate storage, the chemistry dominating grid-scale BESS today, with roughly six to eight thousand cycles and a levelized cost that has collapsed over three years, solves intraday shifting. Two to four hours. It does not solve multi-day firmness, and it does not solve the base-load reliability a hyperscaler is actually buying.
So "renewables will quench AI's power thirst" is a category error. The correct sentence is "renewables plus firm capacity quench AI's power thirst," and firm capacity means nuclear, gas, or long-duration storage that has not yet proven a business model. Here is where the reporting gets lazy: NextEra is not a pure-play renewable developer. It owns nuclear — Point Beach, Seabrook, the Duane Arnold restart — plus a substantial gas fleet. The label on the box and the asset inside the box are different things. Value NextEra off its wind-and-solar headline and you have mispriced the most defensible part of the company.
Then trace the materials. Thirty-five gigawatts converted to hardware becomes an ordering schedule for lithium, copper, transformer-grade electrical steel, rare-earth magnets, and — if the nuclear thesis holds — uranium. Every one of those sits on a concentrated, geopolitically sensitive supply chain. The 35 GW number is not really a capacity figure. It is a materials call option with an unspecified delivery date, and that is before you count transformers, currently the tightest single component in the American grid build.
Now the delivery problem, which crypto readers should recognize instantly. Backlog is not power. Backlog is a promise. The U.S. interconnection queue holds roughly 2.6 terawatts of projects waiting for grid access, with average waits stretching three to five years. That queue is the true gatekeeper, and it behaves less like a market than like a DAO whose upgrade rights sit with a handful of multi-sig admins — a few regional operators and transmission owners deciding who connects and when. No amount of capital overrides that signature threshold.
Layer-two fragmentation hands us the cleanest analogy. Dozens of chains, the same thin liquidity, each promising scale while slicing the same small pool of users. Count the AI-power "solutions" now — green hydrogen, gravity storage, sodium-ion, compressed air — and the pattern repeats: dozens of answers, one scarce input, firm dispatchable megawatts with a live interconnection. Splitting liquidity does not scale it. Splitting power solutions does not generate it. The queue is a filter, and filters do not disappear just because demand is loud.
And the cost line nobody touched. Battery cells, rare-earth magnets, and transformers are concentrated supply chains, and U.S. policy is actively trying to de-risk them through tariffs and foreign-entity rules. De-risking carries a price, and that price lands on the developer's hurdle rate. The entire clean-energy buildout leans on IRA tax credits and tax-equity financing; trim the credits and the math on 35 GW changes overnight. Entropy increases until someone audits it — and almost nobody audited this.
Here is the angle the aggregators skip, because it requires holding two ideas at once. The word "renewables" hides a violent divergence. Upstream, module and cell manufacturers are being gutted by a price war — margins near zero, capacity glutted. Downstream, asset owners who buy those cheap components and sell scarce electricity into a tightening capacity market capture a historic spread. Same word, opposite fates. Anyone who read "renewables" as one bullish theme erased the most important structure in the trade.
Then the crypto layer, where I actually live. The AI-agent economy I spend my days mapping does not care about green labels. An autonomous agent negotiating machine-to-machine value — compute, bandwidth, energy — optimizes for uptime and cost per joule, not marketing copy. That is the convergence nobody is pricing. If by 2027 a majority of on-chain volume originates from agents, the settlement rail for energy becomes the product, and any token clearing it needs physical backing and verifiable metering, not a narrative. Volatility is the tax on uncertainty, and right now the market is paying that tax on a number it cannot define.
Three signals to watch, none of them the headline figure: the baseload share inside new hyperscaler PPAs, the conversion rate from interconnection queue to energized asset, and whether the next marquee signing is nuclear-first. Speculation is just data with a heartbeat. So the next time a gigawatt number lands in your feed, ask for the timestamp before you ask for the ticker. The pool remembers what the ticker forgets.