According to Zylo's 2026 SaaS Management Index, 43% of enterprise software licences go unused. Apply that rate to an AI assistant rollout and the number on your invoice stops being the number that matters. This page sets out what the major enterprise AI platforms actually charge in 2026, what a 300-person Hong Kong company should expect to commit, and when a seat licence is the wrong purchase entirely.
This is a buying-decision page. If you are choosing between per-seat AI assistants for your organisation this quarter, the figures below are the ones to take into the CFO conversation.
How much do enterprise AI seat licences cost in 2026?
Published 2026 list prices sit between roughly US$18 and US$60 per user per month for the main enterprise AI assistants, with Microsoft 365 Copilot at US$30, Gemini Enterprise from US$21 to US$60 depending on tier, Claude Enterprise from around US$20 plus usage, and ChatGPT Enterprise quote-only.
Published per-seat pricing, mid-2026
--- Microsoft 365 Copilot: US$30 per user per month for enterprise, which is US$360 per user per year on annual billing. A small-business standalone tier is US$18 per user per month at a promotional rate (standard US$21) with no seat minimum, available through 31 December 2026. Critically, this is an add-on price. It sits on top of a base Microsoft 365 plan, so a real E3 plus Copilot user costs about US$66 per month and an E5 plus Copilot user about US$87.
--- Gemini Enterprise: Business edition starts at US$21 per seat per month. Standard is US$30 per user per month on annual commitment, or US$35 on flexible monthly billing. Plus is from US$50 per user per month on annual commitment, or US$60 without one. Consumption charges apply beyond quota.
--- Claude Enterprise: A hybrid model. The base seat fee starts around US$20 per seat per month plus usage billed at API rates, with reported effective ranges of roughly US$20 to US$60 per user per month depending on contract terms and committed volume. Minimum 50 seats, custom contract negotiated with Anthropic.
--- ChatGPT Enterprise: No published price. Quote-only, with 2026 procurement reports converging on roughly US$45 to US$75 per seat per month, averaging around US$60, a reported 150-seat minimum and annual prepay.
Prices are vendor-published or drawn from 2026 procurement reporting and change frequently. Verify current terms with the vendor before signing. Figures are in US dollars; at roughly 7.8 HKD to the US dollar, US$30 per seat per month is about HK$234.
What is the real minimum annual spend for a 300-person Hong Kong company?
For 300 seats, published rates produce an annual range of roughly US$75,600 to US$216,000, or about HK$590,000 to HK$1.69 million, before any base-licence upgrades, usage charges, integration work or change management. The floor is set by seat minimums, not by your appetite.
Annual cost at 300 seats, list rates only
--- Gemini Enterprise Business at US$21: US$75,600 per year, about HK$590,000
--- Microsoft 365 Copilot add-on at US$30: US$108,000 per year, about HK$842,000, plus your existing Microsoft 365 base licences
--- Gemini Enterprise Plus at US$50 on annual commit: US$180,000 per year, about HK$1.40 million
--- ChatGPT Enterprise at a mid-range US$60: US$216,000 per year, about HK$1.69 million, on annual prepay
Two structural facts change the shape of this decision. ChatGPT Enterprise carries a reported 150-seat minimum, which puts the floor for any Enterprise deployment at roughly US$108,000 per year even for a smaller pilot population. Claude Enterprise requires a minimum of 50 seats. If you wanted to run a genuine 30-person pilot before committing, neither Enterprise tier accommodates that directly.
What is not included in the per-seat price?
The seat fee buys access, not outcomes. Four cost categories sit outside every quoted per-seat number, and in mid-market deployments they frequently exceed the licence cost in year one.
--- Base platform licences. Microsoft 365 Copilot is an add-on. If your organisation is on a lower Microsoft 365 tier, the upgrade is a separate line item.
--- Usage beyond quota. Claude Enterprise bills usage at API rates on top of the seat fee, and Gemini Enterprise applies consumption charges beyond quota. Heavy users cost more than light users on the same seat.
--- Integration and data connection. An assistant that cannot reach your document management system, ERP or CRM produces generic answers. Connecting it is project work, and it is where legacy systems create the real timeline.
--- Adoption and change management. This is the category most often left out of the business case and the one that determines whether the licence returns anything.
Which platform fits which buyer?
The right platform depends on where your work already lives, not on benchmark scores. For most Hong Kong mid-market organisations, the existing productivity suite and the location of your documents decide this more than model capability does.
Verdict by buyer type
--- Already standardised on Microsoft 365 with documents in SharePoint and Teams: Microsoft 365 Copilot at US$30 is the path of least integration resistance. Confirm your base tier first, because the effective cost is US$66 to US$87 per user per month, not US$30.
--- Running on Google Workspace: Gemini Enterprise Business at US$21 is the lowest published entry point among the four, and the integration work is minimal.
--- Document-heavy professional services work, long analysis, drafting and review: Claude Enterprise suits this profile, but the hybrid seat-plus-usage model means you need a usage estimate before you can forecast spend. Model the heavy users, not the average.
--- Large deployment, more than 150 seats, strong preference for the broadest third-party ecosystem: ChatGPT Enterprise fits, provided you are comfortable with quote-only pricing and annual prepay.
--- Under 150 employees, or wanting to prove value on one workflow before any organisation-wide commitment: None of the Enterprise tiers is a good first purchase. See the next section.
Why does the 43% unused-licence rate matter to your cost per active user?
Because cost per active user, not cost per seat, is the number that survives board scrutiny. At Zylo's reported 43% unused rate, a US$30 seat becomes an effective US$52.60 per person who actually uses it, and the ROI case in your business plan quietly loses 43% of its denominator.
The underlying waste problem is well documented. Zylo's 2026 SaaS Management Index puts unused enterprise licences at 43%, costing large companies an average of US$80.6 million annually, and one 2026 analysis reports 51% of purchased SaaS licences going unused, the highest waste rate recorded. For mid-market organisations above 200 employees, average software waste runs to roughly US$235,000 per year, close to 48% of total software spend.
An idle AI licence costs exactly what an intensively used one costs. The difference is adoption, not the price on the invoice. Which is why the most useful question in a vendor negotiation is not "what is your best per-seat price" but "how do we measure and lift active usage in the first 90 days".
When is a per-seat assistant the wrong purchase?
A per-seat assistant is the wrong purchase when the work you want automated belongs to a role rather than to a person. Seat licences give everyone a chat window and leave the workflow design to them. Role-based AI workers take a defined function, such as first-line customer response or invoice processing, and run it end to end.
The distinction is a budget distinction. A per-seat rollout is a productivity bet: you are paying for 300 people to each work somewhat faster, and your return depends entirely on whether they change how they work. A role-based deployment is an operations bet: you are paying for a defined process to run at a defined quality and volume, and the return is measurable in that process.
This is the model behind UD's AI Staff Solution, which deploys AI workers against specific business functions including marketing, HR, administration, accounting, customer service and IT, rather than issuing a licence to every desk. For organisations under 150 employees, or organisations that need one workflow to demonstrate return before an organisation-wide commitment, it avoids the seat-minimum floor entirely.
Being straightforward about it: this is not a replacement for a productivity assistant. If your objective is to give every knowledge worker a general-purpose drafting and summarising tool, buy the seat licence that matches your suite. The two approaches solve different problems, and plenty of organisations end up running both.
What are the limitations of this comparison?
Four limitations you should hold in mind before treating these figures as a quote. This page is a planning aid, not a proposal, and anyone presenting AI pricing as settled is overstating what is knowable.
--- Enterprise pricing is negotiated. ChatGPT Enterprise has no list price at all, and the US$45 to US$75 range comes from procurement reporting rather than the vendor. Your quote will depend on volume, term and timing.
--- Usage-based components are not forecastable from a seat count. Claude Enterprise and Gemini Enterprise both add consumption charges. Two organisations with identical headcount can see materially different bills.
--- Prices moved several times in 2026. Microsoft adjusted Microsoft 365 pricing from 1 July 2026, and OpenAI moved Workspace Agents to credit-based billing on 6 July 2026. Verify before you sign.
--- UD is a deployment and integration partner, not a licence reseller for all of these platforms. Our commercial interest is in deployment, adoption and role-based AI, so read the previous section with that in mind. If your best answer is a seat licence, the honest recommendation is to buy the seat licence.
What should you do next?
The correct next step is not a vendor demo. It is establishing which workflows in your organisation would actually change if an AI assistant were available, because that determines whether you are making a productivity bet or an operations bet, and those have different price tags.
A workable 30-day sequence:
--- Confirm your current base licence position, because the effective Copilot cost depends on it entirely
--- Identify three workflows with measurable current cycle times, so a before-and-after exists
--- Decide whether those workflows belong to roles or to individuals, which resolves the seat-versus-role question
--- Set an active-usage target before signing, not after, and make it a contractual review point
--- Only then request quotes, with your seat count and usage profile already defined
Enterprise AI pricing is one of the few technology decisions where the cheapest published rate and the lowest total cost are rarely the same option. Twenty-eight years of Hong Kong enterprise deployments have taught us that the deciding factor is almost never the per-seat figure. We understand the cold edges of AI and the hard parts of your work, and UD has walked with Hong Kong enterprises for twenty-eight years, making technology a partnership with warmth.
Further reading: what happens to output quality after a broad AI rollout.
Reviewed by the UD enterprise AI team. Pricing verified against vendor-published rates and 2026 procurement reporting as at 29 July 2026.
Get a costed recommendation
Before you commit to a per-seat contract, it is worth knowing whether a seat licence or a role-based AI worker fits your workflows and your headcount. We'll walk you through every step, from readiness assessment and workflow selection to cost modelling, deployment and active-usage tracking.
