At the end of the month, a support team's AI avatar has talked to customers for a total of 500 hours. Most of those hours were ordinary: a delivery date, a package delay, and a refund that hasn't arrived.
Drawn in the cloud, those 500 hours keep a data-center GPU busy for about 100 hours. Drawn on the customer's own phone or laptop, they keep it busy for none. Those 100 hours, with the power and cooling behind them, stay with the data center as headroom: spare capacity for other work.
Why you pay by the minute
An AI chatbot can share one GPU among many users, since each answer needs it for only a second or so. A live avatar drawn in the cloud can't. It has to draw 25 to 30 new images of the face every second, for as long as the conversation lasts, even while the customer is the one talking. Even the most efficient model with a published benchmark fits only about five such conversations on one NVIDIA A10, an entry-level data-center GPU.
A GPU costs the provider money for every hour it runs. When each conversation keeps a fifth of one busy from start to end, the cost of serving a customer grows with every minute they talk. So the price is set the same way: by the minute.
At a fifth of a GPU per conversation, 500 conversation-hours hold 100 GPU-hours for the faces alone. That load grows with every minute of conversation: a contact center running 10,000 conversation-hours a month holds about 2,000 GPU-hours.
How we got these numbers: 500 hours × 20% of a GPU = 100 GPU-hours. The 20% comes from Hedra's published Livatar-1 benchmark (141 frames a second on one NVIDIA A10). It is a share of the GPU's frame output, not a measured utilization. Larger photoreal models on bigger GPUs hold several times as much.
MeshLite draws the face on the user's own device, so no data-center GPU is held for it during the conversation.
What the support desk pays
At 500 hours a month, our support desk pays more than four times as much on a per-minute plan as on a plan with a fixed monthly price. The per-minute bill comes to $2,635 a month. MeshLite Business is $499 a month. Over a year, that is $31,620 against $5,988.
The difference grows as the avatar gets more use. Add an evening shift or make it available on another customer portal, and a per-minute bill rises with the extra conversations. A fixed monthly price stays the same within the plan’s limits. The team can serve more customers without paying for every extra minute.
To keep the comparison fair, we picked a plan that covers the same ground as MeshLite: the avatar only. In both cases the customer brings speech recognition, the language model and the voice. One large AI avatar provider sells this plan for $475 a month with 100 hours included, then $0.09 a minute, or $5.40 for every extra hour.
| Avatar hours a month | Per-minute plan | MeshLite Business, billed monthly | MeshLite Business, billed annually |
|---|---|---|---|
| 100 | $475 | $599 | $499 |
| 200 | $1,015 | $599 | $499 |
| 500 | $2,635 | $599 | $499 |
| 2,000 | $10,735 | $599 | $499 |
Avatar only. Speech recognition, the language model and the voice cost extra in both setups. The same provider's plans with its own voice and language model charge more per minute. MeshLite Business covers up to 50 conversations at the same time; on our estimate, this desk needs about 12 at its busiest.
When per-minute is the better deal. At low volume, the per-minute plan can be a more economic plan. The two types of billing cost the same at about 104 hours a month, roughly three and a half hours of conversation a day, or 123 hours if MeshLite is billed monthly. We think that a team that uses its avatar for an hour a day should stay on per-minute pricing. To see where your own volume lands, try our cost calculator.
What an app with 600 users pays
Take a training app with 600 monthly users who each practice with its avatar twice a week, for about 10 minutes at a time. That comes to about 87 minutes per user a month, or roughly 870 conversation-hours. Rendered in the cloud, the faces hold about 175 GPU-hours a month, around 2,100 a year.
On the same per-minute plan, those 870 hours cost about $4,615 a month, or $55,380 a year. MeshLite Starter covers up to 1,000 monthly users in a single product for $11,988 a year ($999 a month). The crossover is about 197 hours a month, roughly 135 users at the same 87 minutes a month. If the app grows from 600 users to 1,000, the per-minute bill grows by two-thirds. The fixed monthly price doesn't move until the product outgrows its tier, and then it steps to the next one.
Can a discount close the gap?
A discount could change the price. It does not change the work: a face drawn in the cloud still keeps part of a GPU busy for every minute of every conversation.
Drawing the face on the device removes that work, and that is the headroom. The GPU time, power and cooling the face would have used stay free for other jobs. AI is already short of all three, so that matters beyond any one invoice.
Next
How much of a GPU one conversation holds is our own estimate, because none of the AI avatar providers we checked publishes it. Next, we turn that gap into questions any buyer can ask a provider before signing.
Sources: AI avatar providers' published pricing and billing documentation (read 2026-10-19) · MeshLite pricing · Hedra, *Livatar-1: Real-Time Talking Heads Generation with Tailored Flow Matching* (2025), arXiv 2507.18649 · NVIDIA A10 specifications.