In Houston, a Base Power customer pays $695 to install a battery, $19 a month, and 13.1 cents per kilowatt-hour—roughly the going rate in the region. That is the entire pitch: a knowable price for a physical thing that already sits in more than a few thousand backyards.
Hold that number. It is the most concrete fact in a week where three energy companies raised roughly a billion dollars each, all of them wearing the same jersey.
Valar Atomics: $1 billion led by Sequoia's Shaun Maguire, plus a $200 million credit line, at a reported $6 billion valuation. Base Power: another $1 billion, less than a year after its last billion, at $13 billion post-money. Commonwealth Fusion Systems: $1 billion, the largest fusion investment round worldwide since the $1.8 billion it raised in 2021, bringing total capital to $4 billion.
Three deals, one narrative. The round sizes are nearly identical; the evidence bases are not remotely comparable.
Three deals, one narrative. AI needs power. Everyone now agrees the AI-energy nexus is "real," and this trio maps onto it with almost suspicious neatness.
Last week, on the 07/28 nuclear-capital thread, the open question was whether any SMR could secure a regulator-approved deployment path. This week the market answered a different question entirely—not "will regulators sign off" but "will private capital front the money anyway"—and priced its answer at $6 billion for Valar and $4 billion cumulative for fusion.
So let me puncture the symmetry with a single test: what can each of these companies put on a truck today?
Start with Base, because the answer there is unambiguous. Over the past few years it has installed more than 500 megawatt-hours of storage. It is putting in about 100 batteries a day—roughly 8 MWh daily—and wants to double that by year's end.
The new unit, Base Core, stores 39.2 kilowatt-hours and is built at an Austin factory. Customers take one or two. The company operates in Texas and Illinois, owns the batteries, and sells stored electricity back to the grid during demand peaks.
None of that is a promise. It is a shipped product, a subscription line item, and a fleet you can count. The $13 billion valuation is rich, but it is rich on something delivered.
Now Valar. In June it demonstrated its Ward 250 reactor powering an Nvidia Blackwell system. It announced a deal with Nvidia to develop a waterless 30MW AI factory. That is a genuine, vivid milestone, and it is the closest thing this week to the deployment-path question the 07/28 thread left open.
But read the company's own framing carefully. "It took two years to complete the NOVA core. It took seven months to take Ward 250 critical," it said, adding that with each reactor "the tick rate will become smaller until Valar is producing tens, then hundreds, then thousands of reactors per year."
That sentence is the valuation. The $6 billion isn't priced on Ward 250 and a demo. It's priced on the fleet that comes after: tens, then hundreds, then thousands of factory-built reactors that do not exist yet.
The stated purpose of the raise confirms it. Sequoia's money helps Valar "shift from demonstrating small reactors to producing them in volume." You are paying for the shift, not the demonstration.
Then CFS, where the gap is widest. Khosla backed it in 2019 "when almost no other fund would touch fusion," and Vinod Khosla now calls it one of the most impactful and ambitious projects he's backed in his career. That is real conviction from a real track record.
But strip the endorsement and count the electrons on the grid: zero. The $4 billion cumulative total is underwritten by a milestone and a thesis, not by commercial fusion power delivered to a single customer. Canary Media's own headline on the round carried the operative caveat—"It'll take more."
So arrange the three on one axis: delivered engineering. Base has a fleet and a price. Valar has a reactor and a roadmap. CFS has a milestone and a belief. The round sizes are nearly identical; the evidence bases are not remotely comparable.
Here is the uncomfortable mechanism underneath. All three rounds are being sold on demand-side certainty—AI wants power, the grid is strained, PJM is reportedly strained by data centers. That part is well-evidenced; electricity demand is genuinely skyrocketing.
But demand-side certainty is being quietly laundered into supply-side certainty. "AI needs power" is a fact about load. "These specific reactors will deliver it" is a claim about manufacturing, regulation, and physics that no one has yet demonstrated at commercial scale.
Base is the only one of the three where those two things are the same statement. Its batteries are already absorbing and dispatching real electrons in real markets. When Lightspeed says it is backing "this generation of tech," the tech is on the ground.
For Valar and CFS, the demand story is doing load-bearing work the supply story cannot yet support. The Nvidia Blackwell demo is powerful precisely because it collapses the distance rhetorically—look, a reactor powering AI hardware—while leaving the manufacturing fleet, and the regulatory path the 07/28 thread flagged, entirely in the future tense.
And the demand anchor itself is now moving. States are beginning to claw back data-center tax incentives as lawmakers question whether the tech giants still need them. That is not a policy story I'll relitigate here—it matters only as a risk to the underwriting.
Because if the demand math softens even at the margin—if data centers get more expensive to site, if some projected load never materializes—the companies with delivered hardware and knowable prices can reprice. The ones priced on fleets-that-don't-exist and electrons-not-yet-generated have thinner cover.
That's the read. The symmetry of three billion-dollar rounds is an illusion produced by round size. What actually differs is how far each valuation reaches past what's been shipped.
Base is charging $695 and $19 a month for a battery you can touch. Valar is charging a reportedly $6 billion valuation for a tick rate that has to keep shrinking. CFS is charging $4 billion for a promise Khosla has held for seven years.
Only one of those is grid-storage-grade certainty. The other two are being priced as if they already were.