October 31, 2026 • AI Costs & Budgeting • By Jeff Utecht

AI Budget for School Districts: A Worksheet for Your CFO

District business officer and superintendent reviewing an AI budget worksheet together at a conference table

Budget season is coming, and every AI budget for school districts I see has the same line item: “AI licenses, staff, renewal.” When I sit down with business officers, my first question isn’t about AI at all. It’s this: how many of the seats you bought last year were actually used?

That question matters because the pattern in ed tech is not encouraging. The 74 reported this spring that more than 65% of purchased ed-tech licenses typically go unused, citing a March 2026 report from Digital Promise and the Center for Outcomes Based Contracting. That figure covers ed tech broadly, not AI tools specifically. But AI seats are priced the same way, and that should make every CFO curious.

So here’s a worksheet you can take into your next cabinet meeting. It builds the budget two ways, per seat and per use, with prices I re-checked today, October 31, 2026, and arithmetic you can check yourself.

Key takeaways

  • Per-seat AI pricing charges the same for a staff member who never logs in as for your heaviest user, every year.
  • Commercial AI providers sell the same kind of models by the token, and a typical staff request costs about one to two cents at current published API rates.
  • A district-owned private server is a one-time $20,000 purchase with no per-user fees; open-source models on it have no token fee, and commercial API tokens are paid separately, only for what is used.
  • Token prices change too, so build a buffer into the budget and keep the freedom to switch models.

Step 1: Price the seats, then divide by real users

Let’s use a labeled example: a district with 300 licensed staff.

Google’s AI Pro for Education add-on lists at $20 per user per month on an annual plan, or $24 on a flexible plan. It’s licensed per user and available only to users 18 and over, so it covers staff, not students.

  • 300 staff × $20 per month × 12 months = $72,000 per year
  • Over five years, at today’s list price with no increases: $72,000 × 5 = $360,000

Now the CFO question. Suppose (hypothetically) 120 of those 300 staff use the tool regularly. Your real cost is $72,000 ÷ 120 = $600 per active user per year, or $50 a month, not $20.

Other tools follow the same pattern. MagicSchool lists its individual Plus plan at $12.99 a month, or $99.96 billed yearly; school and district Enterprise pricing is a custom quote. SchoolAI publishes no prices for its plans. I won’t guess at quote-only prices, and you shouldn’t accept a guess either.

To be fair, those subscriptions buy real things: polished teacher features, AI built into apps staff already use, support and curriculum tie-ins. The question isn’t whether they’re good. It’s whether paying for every seat, forever, is the best way to buy AI.

Step 2: Tokens explained, and what a request really costs

When AI providers sell to developers, they don’t sell seats. They sell tokens. A token is a small chunk of text, roughly three-quarters of an English word. Providers charge per million tokens the model reads (input) and per million it writes (output).

For this worksheet I’ll assume a generous 3,000 input tokens and 1,000 output tokens per request. That’s a working assumption, not a measurement. Here’s what one request costs at current published prices:

  • Claude Sonnet 5.5 ($2 input / $10 output per million, Anthropic pricing): (3,000 × $2 + 1,000 × $10) ÷ 1,000,000 = $0.016
  • OpenAI gpt-6.1-sol ($2 / $10, OpenAI pricing): same math = $0.016; gpt-6.1-terra ($1 / $5) = $0.008
  • Gemini 3.8 Flash (Google pricing): $0.75 / $3.75 through Dec. 31, 2026, then $1.50 / $7.50. At the 2027 rate: (3,000 × $1.50 + 1,000 × $7.50) ÷ 1,000,000 = $0.012
  • Claude Haiku 5.5 ($0.10 / $0.50 for prompts up to 100,000 tokens): $0.0008, less than a tenth of a cent

Now the same 300-staff district, paying per use. Hypothetically, 150 regular users make 150 requests a month for a 10-month school year:

  • 150 users × 150 requests × 10 months = 225,000 requests
  • 225,000 × $0.016 = about $3,600 per year if every request went to Claude Sonnet 5.5 or gpt-6.1-sol

Stress-test it. If all 300 staff made 300 requests a month for 12 months: 1,080,000 requests × $0.016 = about $17,280. That’s still under a quarter of the $72,000 seat bill.

Step 3: Put the private server into the worksheet

The PrivatEDU AI server is a one-time $20,000 purchase that includes pre-configured server hardware, 12 months of remote support and a year of virtual professional development run by LeadershipAIplus. There are no per-user licensing fees, and one server supports roughly 200 concurrent users. Two details, stated plainly:

  • Open-source models on the server have no token fee. Models like Llama and Mistral run locally on district hardware.
  • Commercial API tokens are not included in the $20,000. If the district connects Claude, OpenAI or Gemini, it opens its own API account with each provider and pays that provider directly. Those requests go to that provider under its API terms, and the district decides when and for whom.

Year one in the example: $20,000 + about $3,600 in tokens = about $23,600, against $72,000 in seats. Send 70% of requests to a local model and only 67,500 go to a commercial API: 67,500 × $0.016 = about $1,080 in tokens.

The server also serves students. Role-based access gives administrators, teachers and students separate environments, so students aren’t left out the way they are with an 18-and-over add-on.

One honest caution: token prices move too. Google’s page shows Gemini 3.8 Flash doubling on January 1, and its flagship model’s price doubling on December 1. That’s why I’d budget a buffer, and why owning the platform matters. When a price jumps, you can switch the model behind a teacher’s assistant instead of renegotiating a contract.

What district leaders can do this month

  1. Pull 30-day active-user counts for every AI tool you license, and calculate cost per active user.
  2. Run this worksheet with your own numbers: staff count, realistic users, requests per month, months per year.
  3. Ask every vendor for a written quote for quote-only plans, including renewal terms and price-increase caps.
  4. Add a token line with a buffer. If you connect commercial models, budget your estimate plus headroom for price changes.
  5. List the five-year questions for any platform: support after year one, hardware refresh, staff time and training. The PrivatEDU AI site doesn’t publish support pricing after the first year, so ask about it in a Systems Review.

For the non-financial planning questions (data classification, access, long-term support), my private AI servers planning guide at K12 AI Consulting pairs well with this worksheet.

Why a private server changes the AI budget for school districts

Ownership turns AI from a seat count into infrastructure. You pay once for the platform, pay nothing per token for local models, and pay only for the commercial tokens you choose to use. And because we recommend running the server on-site, prompts and logs stay on district-owned hardware inside your network.

Just as important is who’s behind it. The program runs through LeadershipAIplus, educational leaders supporting educational leaders, not a vendor chasing seats. Its courses include an AI Server user guide and a Server Admin Certification, so the year of included PD builds skills your team keeps. Tricia Friedman and I talked about why districts now want that kind of long-term support, not one-off workshops, in the Shifting Schools podcast episode AI in Schools Is Changing: What Leaders Need Now. For the companion one-year comparison, see AI Cost for School Districts: Paying for Seats vs. Paying for Use.

FAQ

How much does AI cost per teacher?

It depends on the pricing model. A per-seat add-on like Google AI Pro for Education lists at $20 per user per month on an annual plan. Pay-per-use API access at current published prices works out to roughly one to two cents per typical request.

Are API token costs included in the $20,000 private server?

No. The district pays each commercial provider directly for tokens it uses. Open-source models running on the server have no token fee.

What should a school board ask before renewing AI licenses?

How many seats were active in the last 30 days, what each active user really costs, what renewal terms and price caps apply, and whether students are covered.

The bottom line

Seats are easy to buy and hard to justify. Usage is easy to measure, if you own the platform that measures it.

So here’s my question: if you divided last year’s AI spend by the people who actually used it, would you sign the same renewal?

If you’d like help running these numbers for your district, Schedule a Systems Review.

Sources

Leave a Reply

Your email address will not be published. Required fields are marked *