Copilot's 50% Discount at 10,000 Seats: What the Volume Cliff Says About Demand

A Microsoft 365 Copilot seat that lists at $30 a month reportedly drops to $15 once a customer buys more than 10,000 of them. The reported Microsoft Copilot discount is 30% above 1,000 seats and 50% above 10,000. If the discount applies to every seat on the contract, a customer with 9,999 seats pays about $720,000 a year more than one with 10,000. That cliff tells you more about Copilot demand than any press release does.

What Microsoft reportedly changed in Copilot pricing

On 22 September 2026, The Information reported that Microsoft had authorised large discounts on Microsoft 365 Copilot for big corporate buyers. The story relied on one person with direct knowledge of the change. Microsoft has not published the tiers.

The reported terms are simple. Customers buying more than 1,000 Copilot seats get roughly 30% off. Customers buying more than 10,000 seats get up to 50% off. The discounts could start as soon as October, and they come attached to commitments to additional usage-based payments.

The Information sits behind a paywall, so this post works from two outlets that relayed it the next day. IT Home reported the same tiers on 23 September 2026, and so did Analytics India Magazine. Both are Tier 2 relays of one anonymous source. Read every number below with that in mind.

The list price the discount comes off has not changed since launch. Microsoft announced Copilot in July 2023 at $30 per user per month for E3, E5, Business Standard and Business Premium customers. It is an add-on. You pay it on top of the Microsoft 365 licence you already have.

So the headline question is narrow. What does a seat actually cost you at each tier, and what does the shape of the discount say about how Copilot is selling?

Copilot volume discount: the effective seat price by tier

Start with arithmetic that needs no assumptions. A 30% discount on $30 is $21 a month, or $252 a year per seat. A 50% discount is $15 a month, or $180 a year. At list price, a seat costs $360 a year.

Discounts as reported by The Information, 22 September 2026, relayed by IT Home and Analytics India Magazine. List price from Microsoft's July 2023 announcement. The final column is our arithmetic: 1 divided by (1 minus the discount). Usage charges are excluded throughout.

The last column is the one to keep. It tells you how many more seats Microsoft has to sell to earn the same seat revenue after the discount. At 30% off it needs 43% more seats. At 50% off it needs twice as many.

The all-units assumption

None of the reports say whether a tier discount applies to every seat on the contract or only to seats above the threshold. That detail decides whether a cliff exists at all. The reporting describes discounts "for customers with more than 10,000 seats", which reads as a price for the whole contract rather than a marginal rate.

This post models the all-units reading because it matches that wording and because enterprise software volume bands commonly work that way. If your Microsoft account team says the discount is marginal, the cliff disappears and the dead zones below do not apply. Ask before you plan around it.

The two dead zones

Under the all-units reading, 999 seats at list price cost $359,640 a year. One more seat drops the bill to $252,000. Any customer holding more than 700 seats would pay less by buying 1,000.

The top threshold is far larger. 9,999 seats at $21 cost $2,519,748 a year. 10,000 seats at $15 cost $1,800,000. Any customer above about 7,143 seats pays less by rounding up to 10,000.

Why a seat cliff tells you what Microsoft wants

A pricing schedule is a statement of intent. Smooth volume curves reward size gradually. A cliff with a dead zone underneath tells customers near the line that rounding up is the obvious move.

That is the design working, not a mistake. A procurement lead holding 8,000 seats who sees a lower bill at 10,000 will buy 10,000. Microsoft gets 2,000 additional seats deployed, and the customer's total spend falls. Both sides can call it a win, which is exactly why cliffs exist.

My read is that the cliff is a demand signal, not a generosity signal. A vendor whose product sells itself to every employee does not need to make the 10,000th seat nearly free. A vendor whose product is stuck in pilots of a few thousand seats inside companies with 50,000 employees needs exactly this mechanism.

The lower threshold says the same thing at a smaller scale. A 30% band starting at 1,000 seats targets the many mid-sized enterprises that bought a few hundred seats for a pilot and stopped. The tiers are aimed at the gap between a trial and a rollout.

The attach-rate math behind the Copilot discount

Attach rate is the share of an existing customer base that also buys an add-on. For Copilot, the base is Microsoft 365 commercial seats and the add-on is the $30 Copilot licence. Microsoft discloses enough to estimate it.

Under 7% of the base pays

On its January 2026 earnings call, Microsoft said it had 15 million paid Microsoft 365 Copilot seats. CFO Amy Hood said on the same call that paid Microsoft 365 commercial seats had grown to over 450 million. In April, Satya Nadella reported over 20 million paid seats. In July, he reported over 30 million, with net seat adds more than doubling quarter over quarter.

Divide 30 million by 450 million and you get about 6.7%. The true figure is lower, because the 450 million base dates from January and Microsoft said commercial seats grow about 6% a year. More than 9 in 10 Microsoft 365 commercial seats do not carry a paid Copilot licence.

What 50% off has to buy back

Now combine the attach rate with the break-even column. Take a hypothetical customer with 40,000 Microsoft 365 seats and a 3,000-seat Copilot pilot at list price. That is $1,080,000 a year of Copilot revenue.

If the 50% tier moves that customer to 10,000 seats, the bill becomes $1,800,000. Microsoft's revenue from the account rises by two thirds and Copilot coverage goes from 7.5% of staff to 25%. If the customer only goes to 5,000 seats, it stays on the 30% tier and pays $1,260,000. Microsoft still comes out ahead.

The discount loses money only on customers who would have reached 10,000 seats anyway at a higher price. At 6.7% attach, there are few of those. Microsoft is pricing for the 93% who have not bought, and accepting a lower price from the few who would have. That is a rational trade for a product with a demand problem at the breadth level, not the depth level.

A Futurum Group analyst, Shay Boloor, summarised the move on X as roughly a 40% cut for large enterprises, according to Analytics India Magazine. The blended number is plausible. It depends entirely on how many customers land in each tier, which Microsoft does not disclose.

Copilot vs Google and Anthropic enterprise seat prices

The discount also looks different once you put it next to what competitors list. The tiers move Copilot from the top of the seat price range to the bottom.

Google's current pricing page could not be read during research for this post, so the Gemini rows are launch pricing and may have moved. Copilot is an add-on to a Microsoft 365 licence, while the Google and Anthropic seats in this table are standalone, so the seat price alone understates Copilot's total cost.

Gemini Enterprise

Google launched Gemini Enterprise in October 2025 with Standard and Plus editions from $30 per seat per month on annual plans, and a Business edition at $21. Google also bundles Gemini features into Workspace plans, which is a different competitive pressure. That bundling argument is covered in the comparison of what a $30 Copilot seat buys against Gemini in Workspace.

At the 30% tier, a Copilot seat costs the same $21 as Gemini Business. At the 50% tier, it undercuts every Gemini Enterprise edition in the table by at least $6 a month.

Claude Enterprise

Anthropic's pricing page lists Claude Enterprise at $20 per seat per month, billed annually, plus usage at API rates. The seat includes no token allowance. The logic of that structure is unpacked in the analysis of Anthropic's $20 seat and uncapped usage.

Claude's $20 seat and Copilot's $15 tier are converging on the same model: a low entry price and a meter behind it. I think that convergence is the real story. The seat is turning into the cheap part of an AI contract, and the meter is where the vendors expect to earn.

Seat plus usage: where the Copilot discount comes back

The discount does not come free. Every report ties the larger cuts to commitments to additional usage-based payments. That fits what Microsoft has been saying on its own earnings calls.

On the July 2026 call, Nadella said Microsoft was evolving its business model beyond per-seat to per-seat-plus-consumption. He also said usage-based billing had been added to Copilot Cowork earlier that month. In April he had said that any per-user business at Microsoft would become a per-user and usage business.

So the 50% seat discount is one half of a trade. The customer gets a cheaper seat. Microsoft gets a committed usage line that grows with adoption, and the seat discount makes adoption more likely. Which agent features meter, and which stay inside the seat, is mapped in the breakdown of what Copilot Credits meter.

This changes how you should read your own quote. A contract that halves the seat price but adds a usage commitment you cannot forecast may cost more by year two. The headline percentage tells you about the seat. It tells you nothing about the total.

The usage commitment also changes who carries risk. At a flat seat price, Microsoft carries the risk that heavy users cost more to serve. Under a committed usage pool, you carry the risk that adoption is lower than forecast and the commitment goes unspent. The concessions worth asking for in that structure are covered in the guide to buyer concessions on usage-based pricing.

Where this argument is weakest

Three things could make this analysis wrong, and you should weigh them before acting on it.

The whole thing rests on one anonymous source

The tiers come from a single person quoted by The Information and relayed second-hand. Microsoft has not confirmed them. The relays disagree on details too. Some describe the 30% band as "more than 1,000 seats" with no upper limit, and one describes it as 1,000 to 10,000. Whether the top tier starts at 10,000 or 10,001 also changes the exact cliff point.

The all-units assumption behind the dead zones is mine, not the reporters'. If Microsoft applies the discount only to marginal seats, there is no cliff, and the strongest claim in this post falls away. The attach-rate math and the competitor comparison survive either way.

Volume discounts are ordinary, and seat adds are accelerating

Every large Microsoft Enterprise Agreement already carries volume pricing. A sceptic could say a 50% discount at 10,000 seats is normal enterprise software practice, and reading demand weakness into it is over-interpretation.

The growth data supports the sceptic. Paid seats doubled from 15 million to over 30 million between January and July 2026, and net seat adds more than doubled in the most recent quarter. That is not what a stalled product looks like. On this reading, the discount is offence: Microsoft pressing an advantage against Google and Anthropic while its own momentum is strong.

Both readings can hold. Copilot can be growing fast in absolute terms and still be stuck at a low share of each customer. I think the share problem is the one the discount targets. But a reader who weighs the seat growth more heavily has a fair case, and the next two quarters of disclosed seat numbers will settle it.

Copilot seat price negotiation: what to ask for

If you are in a Copilot renewal or expansion conversation this quarter, the reported tiers give you specific questions to ask. Ask in writing, and keep the answers.

The ramp question matters most for companies in the dead zone. If you hold 8,000 seats and need 10,000 to reach the 50% tier, a ramp clause lets you commit to the tier without paying for 2,000 idle seats on day one. Without a ramp, the cliff saving is real but smaller than the headline.

The renewal question matters most for everyone else. Discounts given to win breadth are often taken back once breadth is achieved, because the switching cost has risen. How that plays out at renewal is the subject of the analysis of grandfathered pricing and renewal increases. If I had to choose, I would trade 10 points of discount depth for a written renewal cap on the same seat count.

Finally, bring the competitor table into the room. A $20 Claude Enterprise quote or a Gemini Enterprise quote is a credible alternative for a pilot group, even if a full migration is not. Your account team knows the tiers exist. A competing quote is what moves you from the published band to the bottom of it.

Frequently asked questions

Where to start

Pull your current Copilot seat count and your Microsoft 365 seat count, and divide one by the other. Then place yourself on the cliff chart above. If you hold more than 700 seats or more than about 7,143, the next conversation with your account team should open with the all-units question, before anyone mentions a renewal date.

Separately, ask your finance team which usage meters Copilot is already running in your tenant. If the answer is "none we track", fix that before you sign any tier that carries a usage commitment.

Related reading