This is interesting. A quick calculation with AI suggests pricing for cents/core/GB/hour suggests cost for power+net is 0.5cent/core/GB/hour assuming "average load" (web browsing, installing packages). Or in otherwords ~$3.60 a month for a "tiny machine" on this load 24/7. I'm guessing with actual "per human" usage the cost per person is 10 - 30x lower. So you can serve 1000 people for free a month if you have $300 to pay for power/net.

I took this further and developed a table of breakeven and profit costs for different user sizes:

  Total Users | Tiny Machines (15× / 25×) | Residential Cost/mo | Colo Cost/mo    | Break-even Paying @ $3/mo | Profit Factor @ 10% pay
  ----------- | ------------------------- | ------------------- | --------------- | ------------------------- | -----------------------
  100         | 7 / 4                     | $14 – $25           | $20 – $35       | 5 – 12                    | 0.86× – 2.08×
  1,000       | 67 / 40                   | $144 – $241         | $200 – $335     | 48 – 112                  | 0.90× – 2.08×
  10,000      | 667 / 400                 | $1,440 – $2,401     | $2,000 – $3,335 | 480 – 1,112               | 0.90× – 2.08×
  1,000,000   | 66,667 / 40,000           | $144k – $240k       | $200k – $333k   | 48k – 111k                | 0.90× – 2.08×

All figures use May 2026 US averages (EIA power rates + CBRE wholesale colo data); power+internet only. Profit factor is if exactly 10 % of all users pay the $3/month tier, how many times does the revenue cover your total power + internet cost?

My gut says these prices are still 20-30% too high at the mid-high end tho.