Alpha School Says Kids Can Learn in Two Hours a Day. The Public Data Says: Verify First.
Alpha School is expanding toward 50 campuses on a 2-hour learning AI model. Several charter boards said no. Arizona approved an affiliate. Here is what the public evidence shows.
Anshika Verma · Founder, GrowWise · Last updated
Alpha School is the most talked-about school in America right now.
The pitch is simple. Students use AI-driven software for a maximum of two hours a day on core subjects. The rest of the day goes to life-skills workshops. Adults on campus are "guides," not licensed teachers, and are paid at least $100,000 a year. Tuition runs $45,000 to $75,000.
According to Axios (August 2, 2026), Alpha is growing from 12 locations in California, Texas, and Florida to roughly 50 campuses for the 2026 school year, across 27 new markets including Atlanta, Boston, Charlotte, Chicago, Denver, Nashville, and Raleigh.
The growth is real. The evidence is the open question.
Several states looked at the charter model and said no
The Texas Tribune reported on September 2, 2026 that Alpha's charter applications have been rejected in Texas, Arkansas, North Carolina, Pennsylvania, Utah, and other states.
Arizona is the exception that reporting often gets wrong. The state approved an Alpha-affiliated charter, Unbound Academy. A rejection list that includes Arizona is not accurate.
Texas is still the notable "no." Its State Board of Education had approved most charter applications over the previous decade. It rejected Alpha's in summer 2025. Board member Keven Ellis said he was "a little skeptical of those types of gains."
Then Alpha did something the Tribune called a "backdoor": it registered 30+ private campuses in Texas for the state's new voucher program. The comptroller's office checks only four things: Texas operation, accreditation, two years of operating history, and standardized testing. No curriculum standards. No outcome verification.
One board member described the result: "At this point, it's a franchise."
The Arizona numbers belong to Unbound Academy
The most concrete public outcome data comes from Unbound Academy's first-year state scores, not from a private Alpha campus.
Projected English proficiency: 65%. Achieved: 28%. Projected math proficiency: 60%. Achieved: 10%.
One campus is not a verdict on the model. But it is the kind of number that a charter board is supposed to ask about, and the kind a voucher application does not require.
The critique that matters
Penn State professor Gerald LeTendre, quoted by Axios, put the concern precisely. The model risks "conflating teaching and tutoring." His view: "AI tutoring can help, but it has to be within an atmosphere where there's deep human contact with a highly qualified teacher."
That is not an anti-AI position. It is an argument about where AI belongs.
Tutoring software can deliver practice. It can adapt difficulty. It can move a motivated student through content quickly.
What it cannot do is notice that a student's fraction errors come from a place-value gap two grades back, or that a fluent reader is skipping comprehension entirely, or that three students in the room share the same misconception and should be retaught together. Those are teaching decisions. They require a human who can see the whole picture.
The performance cliff is the evaluation test this model still has to pass: what can students do when the software is removed?
The question every school should take from this
Alpha is expensive and private. Most schools will never copy it.
But every school is being sold some version of the same promise: put students in front of AI and outcomes will follow.
The Alpha story suggests a better set of questions:
- What is the outcome evidence, and who verified it?
- What happens to learning when the software is removed?
- Is AI replacing the teacher's judgment, or informing it?
- Which students does the model fail, and how would you know?
Where we land
At Omnix360, we made a deliberate architectural choice: the AI never interacts with students.
It runs behind the teacher. Plan → Teach → Diagnose → Verify. The teacher decides. The student does the work. The system makes the evidence visible in the same lesson cycle, not at report-card time.
This is not a rejection of Alpha's ambition. Faster, more personalized learning is the right goal. It is a disagreement about method: personalization should sharpen the teacher, not route around them.
See teacher-facing AI and misconception radar and NYC's student-facing pause.
The takeaway
Growth is not evidence. Tuition is not evidence. A voucher approval that checks four boxes is not evidence.
Several charter boards asked Alpha for outcomes and were not satisfied. Arizona approved an affiliate whose first-year scores missed the projection. That is a standard every school should hold for every AI product, including ours.
See Teacher-Facing AI in Action.
Sources
Frequently asked questions
- What is Alpha School's two-hour learning model?
- Students use AI-driven software for a maximum of two hours a day on core subjects. The rest of the day goes to life-skills workshops. Adults on campus are guides, not licensed teachers. Tuition is reported at $45,000 to $75,000.
- Did every state reject Alpha's charter applications?
- No. Charter applications were rejected in Texas, Arkansas, North Carolina, Pennsylvania, Utah, and other states. Arizona approved an Alpha-affiliated charter, Unbound Academy. Approval and evidence are not the same thing.
- What do Unbound Academy's first-year scores show?
- Public reporting of Unbound Academy's first-year state scores showed English proficiency at 28% against a 65% projection, and math at 10% against a 60% projection. One campus is not a verdict on the whole model.
- What should schools ask of any AI learning product?
- Who verified the outcomes? What happens when the software is removed? Does AI replace the teacher's judgment or inform it? The performance-cliff test applies here too.
See teacher-facing AI in action
AI that helps teachers see misconceptions and decide what to reteach next, without student chatbots.