GitHub Copilot Cost per Seat: Finding Ghost Seats with Opslyft

GitHub Copilot cost per seat is $19 a month on Business and $39 on Enterprise, which is cheap enough that nobody questions a seat and expensive enough that the seats add up: a thousand-developer organization on Enterprise pays $468,000 a year before a single premium request. The problem is not the price. It is that the seat bills whether or not the developer behind it ever accepts a suggestion, and in most rollouts, assigned by org or team rather than by request, a third or more of the seats are not being used in any given month. The invoice shows seats. It does not show the ghost seats, the developers who left, moved to management, tried it and stopped, or never activated at all.

This page is about finding them, and about the larger point that AI tool seats are one line in a budget that also contains tokens, GPUs, and cloud compute, consumed by the same teams, and should be managed with them. It covers the 2026 seat prices, the ghost-seat math, why GitHub's own usage metrics are necessary but not sufficient, and how Opslyft shows usage per seat beside cost per seat and routes the reclaim. The pricing itself is in our GitHub Copilot pricing guide and the tool comparison in AI coding tools cost compared; this page is about the waste.

The Seat Prices, and the Ghost-Seat Math

The arithmetic that matters is seats times idle share. A 1,200-seat Enterprise rollout at a 35 percent monthly idle rate is 420 ghost seats, about $196,000 a year, and a reclaim-and-reassign policy that cuts the idle share to 10 percent saves about $140,000 of it without removing the tool from anyone who uses it.

Why Ghost Seats Persist

• The invoice shows seats, not use. Procurement sees 1,200 seats at $39. Nobody in that line sees that 420 produced nothing last month.
• Bulk assignment. Rollouts by org or team assign everyone, including the people who will never open an IDE, and the seats stay assigned through role changes and departures.
• Usage lives in a different tool from cost. GitHub shows activity; the finance system shows the bill; nothing joins them per seat, per team, per month.
• Review happens at renewal. Once a year, if then, which means eleven months of paying for seats nobody checked.
• AI spend is split four ways. Seats in procurement, tokens in an observability tool, GPUs in the cloud console, compute in a FinOps platform: four partial views of what the CFO sees as one AI line, the fragmentation our guide to

why AI costs are cloud costs now describes.

How Opslyft Solves It

Step 1: AI tool spend and seats arrive with everything else

Opslyft ingests AI tool spend and seat assignments from developer platforms alongside the cloud bill and AI provider spend, so Copilot and Cursor seats appear in Explorer as a line beside Bedrock, OpenAI, and the GPU instances the same teams run, with cost per seat and the premium-request credits broken out, tracked over time.

Step 2: Every seat has a team and a cost center

Dimension Studio attributes each seat to the team and cost center that pays for it, by the same rules that allocate cloud spend, so a manager's view shows their team's seats, their team's tokens, and their team's compute as one number, and finance's view shows AI spend by business unit including the tool line.

Step 3: Usage per seat beside cost per seat

Seat activity from the platform is joined to seat cost, so every seat shows what it cost and whether it was used in the last 30, 60, and 90 days. Seats with no activity in the chosen window are flagged as ghost seats with their monthly and annual cost, by team, and seats with low activity are shown separately so that a reassignment policy can be set rather than a blunt cull.

Step 4: Reclaims are tickets to the manager

Tickets routes each team's ghost seats to its manager as an assigned item in Jira, ServiceNow, or Asana, with the list and the annual saving, tracked to closure with the realized saving recorded. Reclaiming becomes a monthly workflow with an owner rather than a renewal-time audit, and reassigning a reclaimed seat to a developer who wants one is part of the same ticket.

Step 5: Anomalies and budgets on the AI line

Monitor flags a seat count that jumped, a team whose premium-request credits tripled after a model change, or an AI tool line that is growing faster than headcount, with the cause, routed to the owner in Slack; budgets on AI spend per team include the seats. Iris answers which teams have the most idle seats and what they cost in plain English.

Step 6: One AI line, so forecasts hold

Because seats, tokens, GPUs, and compute are allocated to the same teams in the same model, AI spend can be forecast per team on real patterns rather than on a procurement guess, which is what lets the forecasting benchmark described in our AWS cost forecasting walkthrough extend to AI-heavy workloads. The practice is set out in our guide to FinOps for AI.

A Worked Example

A company assigned 1,200 Copilot Enterprise seats in a company-wide rollout eighteen months ago, $46,800 a month, and reviews the contract at renewal. Its AI budget also carries about $60,000 a month in Bedrock and OpenAI tokens and $40,000 in GPU instances, tracked in three different places.

On Opslyft, the seats appear beside the tokens and GPUs, attributed to 41 teams. The seat view shows 1,200 assigned, 790 active in the last 30 days, 410 with no activity, and 140 of those with none in 90 days. The 410 ghost seats cost $16,000 a month. Tickets go to 41 managers with their team's list; within a month 350 seats are reclaimed, 60 are reassigned to developers on a waiting list, and the AI tool line falls to about $33,000 a month. The same view shows that two teams' premium-request credits have doubled since a model default changed, which becomes a second ticket. The AI budget is now one line, forecast per team, and the renewal conversation starts from 850 seats that are actually used rather than 1,200 that were assigned.

Proof: the discipline behind itAttributing every line of spend to the team that consumed it is what Opslyft does across cloud, Kubernetes, Snowflake, and AI: it returned 40 hours a month to Disprz with 100 percent automated attribution across six platforms, and it is the basis of the forecasting accuracy benchmark Opslyft tracks against actuals, which holds on AI-heavy estates only when seats, tokens, and GPUs sit in the same allocated model. Read the stories: customer stories

Use Opslyft If

• You have hundreds of Copilot, Cursor, or other AI tool seats and review them only at renewal.
• Seats were assigned by org or team rather than by request.
• Seat cost, token spend, GPU hours, and cloud compute are tracked in different tools by different people.
• Managers do not see their team's AI tool cost beside their team's other spend.
• Finance wants one AI line, by team, that can be forecast.
• You want reclaims to be a monthly ticket with a saving attached, not an annual audit.

If you have a few dozen seats assigned on request and one admin who checks last-activity dates monthly, GitHub's own seat management is enough. The platform earns its place when seats are in the hundreds and AI spend is in several places.

Conclusion

GitHub Copilot cost per seat is small and ghost seats are not: a third of a bulk rollout idling is six figures a year that the invoice will never show, because it shows seats and not use. Join usage to cost per seat, attribute each seat to the team that pays, flag the idle ones monthly, route the reclaim to the manager, and put the whole AI tool line beside the tokens, GPUs, and compute the same teams consume, so that AI spend is one budget that can be forecast. That is what Opslyft's cost visibility does for AI tools as it does for everything else; the broader practice is in our guide to FinOps for AI.