Four new American reactors just 'went critical'. Here's what that actually means
The US government set a patriotic deadline: get new nuclear reactors running by 4 July 2026. Four made it, beating the goal, in some of the fastest reactor builds in decades (a fifth followed in August). It is a genuine milestone driven by AI's hunger for power. It is also not what the headlines imply: not one of them is yet putting a single watt onto the grid.
By 4 July 2026, four new advanced reactors in the US reached "criticality", a self-sustaining nuclear chain reaction, beating a government goal of three, and some were built in under a year. It is a real milestone and a symbol of a genuine nuclear revival, driven largely by AI datacenters' enormous appetite for power. But here is the part the headlines skip: these are zero-power test reactors. They proved the physics works; they produce essentially no heat and zero electricity, and none is connected to any grid. The bigger reactor deals that will actually power datacenters (Google, Microsoft, Amazon, TerraPower) are separate, larger, and still years away. Real momentum, not yet real power.
In May 2025, the US government told the nuclear industry to do something it had not done in half a century: build new reactor designs and get them running, fast, with a deliberately symbolic deadline of Independence Day 2026. The goal was three. On and before 4 July 2026, four made it. That is a genuinely impressive result, and it is being sold as proof that a nuclear renaissance has arrived to feed the AI boom. Both the achievement and the hype need unpacking, because what actually happened is real, important, and much narrower than "America switched on four reactors."
What actually happened?
The push came from an executive order signed in 2025 (Executive Order 14301, on reforming reactor testing) and a resulting Department of Energy Reactor Pilot Program, which set the target of getting "at least three advanced nuclear reactor concepts" to criticality by 4 July 2026, timed to the country's 250th anniversary.
Four reactors cleared the bar. As World Nuclear News reported, the fourth reached criticality in the early hours of 4 July, just ahead of the deadline. Several of these were built astonishingly quickly by nuclear standards, one company went from groundbreaking to criticality in under eight months, against an industry norm measured in years or decades. That speed, enabled by a streamlined federal testing pathway, is arguably the real story.
Wait, what does "criticality" mean?
This is the word doing all the heavy lifting, and it is widely misunderstood.
"Criticality" means a reactor has started and sustained a controlled nuclear chain reaction: enough neutrons splitting enough atoms to keep the reaction going on its own. It is the fundamental proof that a reactor design works. But every one of these four was a zero-power criticality test. As World Nuclear News defines it, that is "a self-sustaining chain reaction... but without reaching full operating temperatures or actively removing heat with a working fluid."
In plain English: the reactors lit the match and proved the fire will sustain itself, but they are being run cold, at essentially no power output. They generate no usable heat, no electricity, and are connected to no grid. One of the companies said as much, describing its milestone as a step toward reactors that would later be deployed to power AI datacenters, all in the future tense. So when you read that America "switched on" or "brought online" four reactors, hold on: they proved the physics. They are not power plants, not yet, and not for years.
The four reactors
The four are tiny experimental "microreactors," and it is worth being precise, because the details get flattened:
- Antares Nuclear's Mark-0 reached criticality on 4 June 2026 at Idaho National Laboratory, reportedly the first novel reactor design to go critical there in more than 50 years.
- Valar Atomics' Ward 250 followed in mid-June, notably not in Idaho but at a state energy lab in Utah.
- Deployable Energy's Unity went critical around 1 July at Idaho, under a separate DOE pathway rather than the main pilot program.
- Aalo Atomics' test reactor (its "CTR", also called Aalo-X) was the fourth, critical at 00:20 on 4 July at Idaho.
So "four beat the goal of three" is true as a count, though only three were formally in the pilot program that carried the deadline. It is a fleet of small proof-of-concept machines, not four power stations.
The run did not stop at the deadline. On 5 August 2026, Oklo's Groves Isotope Test Reactor became the fifth to reach criticality under the program, and the first to do so on private land rather than a national lab, per the Department of Energy. Groves is built to make medical and industrial isotopes rather than electricity, so it does not change the core caveat below, but it does show the sprint is still going.
Why now? The AI power crunch
The reason a decades-dormant industry suddenly sprinted is the same reason it is worth watching AI so closely: datacenters need staggering amounts of electricity, and the companies building them are desperate for clean, reliable power that solar and wind alone cannot guarantee around the clock. Nuclear is the obvious candidate, and the tech giants have moved:
- Google has partnered with Kairos Power to buy power from advanced reactors, targeting roughly 2030.
- Microsoft signed a 20-year deal to buy power from Constellation, which is restarting a reactor at Three Mile Island (now renamed the Crane Clean Energy Center) to feed Microsoft's datacenters.
- Amazon led a roughly $500 million investment round (later upsized to about $700 million) in the SMR developer X-energy.
The wider revival is moving too, beyond the datacenter deals: TerraPower, the reactor developer backed by Bill Gates, won the first US construction permit for a commercial-scale advanced reactor in March 2026, a grid plant in Wyoming. Notice the pattern: these are the serious, large projects, and every one of them is still under construction, in restart, or years from delivering power. They are separate from the tiny test reactors that just went critical. The demand driving the whole boom is real; the electricity is not here yet.
So is the nuclear renaissance real?
Cautiously, yes, and the honest verdict is the interesting one. Three things genuinely changed: the political will (a hard federal deadline the industry actually hit), the speed (sub-year builds that would have been unthinkable a decade ago), and the money (AI-scale demand finally making new nuclear bankable). That is a real shift after fifty years of stagnation.
But temper it with what has not changed. Proving a reactor's physics in a zero-power test is the beginning of a long road, not the end: scaling to full power, generating electricity, connecting to the grid, and doing it safely and economically are each their own multi-year challenge, and nuclear's history is a graveyard of projects that went over budget and behind schedule. The four criticality milestones are a real, verifiable down payment on a nuclear comeback. They are not, yet, a single watt of power for a single datacenter. The useful stance is to be genuinely impressed and to keep the calendar handy: the physics arrived in 2026, the power is a story for the end of the decade. For more from the frontier, see the Future section and our look at whether the AI boom itself is a bubble.
The four reactors, at a glance
| Reactor | Company | Critical | Where |
|---|---|---|---|
| Mark-0 | Antares Nuclear | 4 June 2026 | Idaho National Lab |
| Ward 250 | Valar Atomics | mid-June 2026 | Utah (state energy lab) |
| Unity | Deployable Energy | ~1 July 2026 | Idaho National Lab |
| CTR / Aalo-X | Aalo Atomics | 4 July 2026 | Idaho National Lab |
Google's Pixel 11 is here, and the real product is Gemini everywhere
At Made by Google 2026, Google unveiled a full Pixel 11 line, a folding flagship, a new watch and its first Bluetooth tracker, all built on a 2nm Tensor G6 chip. But the hardware was the supporting act. The headline features are AI ones, a camera that turns sign language into text and a voice input that understands rambling speech, and the prices crept up about a hundred dollars. Here is what launched and what it signals.
Fusion power just got a $4 billion vote of confidence, and a date to prove itself
The old joke is that fusion is always 30 years away. In 2026 the private leaders are trying to make it a dated bet instead: Commonwealth Fusion just raised its total to $4 billion, its reactor is most of the way built, and it is aiming for net energy gain in 2027. Here is what is real, and the caveat that separates a lab result from a power plant.
Quantum computing just had its 'this is a real industry now' week
In a few days in August, quantum computing companies posted triple-digit revenue growth, went live inside a big cloud platform, and landed government chip money. It is a genuine inflection, from lab curiosity to business. It is also not the same as quantum computers being broadly useful yet, and telling those two apart is the whole point.