On Tuesday, Microsoft said it had signed agreements with four of India’s largest information-technology services firms, each committing to deploy more than 50,000 licenses of Microsoft 365 Copilot. Tata Consultancy Services, Infosys, Wipro and Cognizant together account for more than 200,000 seats.
At the list price of $30 per user per month, that order is worth about $72 million a year in subscription revenue. It is a small sum against the four firms’ combined revenue, and a large one against the number of seats any single employer has deployed anywhere.
The deal has two audiences. The first is the customers of those four firms, mostly banks, insurers, retailers and manufacturers in the United States and Europe, who will be told that their outsourcing partners are running the software they are being sold. The second is the Indian firms’ own investors, who have spent two years asking how artificial intelligence changes a business built on billing for people.
Microsoft’s announcement described the deployments as a step toward selling the same tools to clients. The four companies have built practices around moving enterprise workloads to the cloud and now around automating parts of the work those workloads generate. A seat count is the clearest measure of internal adoption available, and it is the number the announcement used.
The economics of the order depend on which tier the seats are bought on. Microsoft lists Copilot at $30 per user per month for its larger enterprise plans and sells a lower-priced business tier that carries an annual commitment. Volume agreements of this size rarely settle at list.
Seats sold and seats used are different figures. Software companies report licenses first and engagement later, and the gap between the two has decided the fate of more than one enterprise product. Microsoft said the deployments will scale in phases, which leaves the question of daily use open.
In the same week, Microsoft published its October security update, which fixed 1,413 vulnerabilities, the largest number in the company’s history. The release included a flaw in Windows Shell that allows an attacker to raise privileges and that the company says has been exploited in the wild.
The research group that tracks the patch count attributed part of the growth to artificial intelligence, which speeds up the discovery of flaws for defenders and attackers at the same time. Vendors now find more of their own bugs because their tools are better, and they face more pressure because everyone else’s tools are better too.
That pairing is the uncomfortable part of the week’s news. The same category of software being deployed to 200,000 employees is the category generating a record repair load, and the firms signing the orders are also the firms that manage patching for other companies’ systems.
The patch load is its own kind of operating cost. Fixes have to be tested against the applications a client runs, scheduled into maintenance windows and deployed without breaking a workflow that produces revenue. Most large organizations install a fraction of what vendors ship, and the fraction falls as the count rises.
The Indian services model rests on a simple trade. Clients hand over work that can be standardized, the vendor performs it with a large trained workforce, and the contract is priced by the hour or by the unit of effort. Margins come from moving work to lower-cost locations and from improving the repeatability of what is delivered.
Automation has always been the friend of that model when it lowered the cost of doing the same work. Generative tools are less comfortable, because they reduce the number of hours a project needs and the hourly rate is where the revenue lives. Chief executives at the largest firms have spent several quarters describing a shift toward contracts priced on outcomes, and none of them has yet shown how much of the old model that replaces.
A seat order of this size is, in that light, a way to buy time and information. It gives the firms a controlled population on which to measure productivity, a credible reference when they pitch the same tools to clients, and a place to learn what breaks before they promise results. The risk sits with the client, who is expected to pay for the deployment, and with the vendor, who must deliver savings that justify the fee.
Shares of the four firms have moved this year in part on how investors judge that transition. The market has rewarded those that report growing order books for AI-related work and has questioned those whose headcount and revenue move together. License counts are easier to announce than earnings from them.
What remains unresolved is the sequence. The seats are being bought to change how the work is done, and the work is what pays for the seats. If adoption is slow, the cost is a rounding error. If adoption is fast, the four firms must sell their clients a different kind of contract, and the clients must agree to buy work measured by results rather than by hours.
