ChatGPT is now the most-expensed app inside the average US enterprise, ahead of AWS and ahead of Microsoft 365, according to Zylo's 2026 SaaS Management Index. The same report puts the average company on 305 SaaS apps, $55.7M of annual SaaS spend, and roughly $19.8M wasted every year on licenses nobody actually uses. That is the backdrop for the most important SaaS AI pricing story of the year, and almost nobody is naming it: in April 2026, four of the biggest workplace vendors quietly unbundled AI from their base subscriptions and put it behind a separate meter, a higher tier, or a brand-new SKU.
Call it the Great Unbundling. Microsoft pulled in-app Copilot from Word, Excel, PowerPoint, and OneNote for non-licensed tenants on April 15. Figma Make moved AI consumption to metered credits managed by admins. Notion locked AI inside the $18 Business plan. Zoom shipped AI Companion 3.0 as a standalone $10/user SKU. None of these were marketed as price increases. All of them are. If you build budgets, sign renewals, or manage a SaaS stack, your 2026 line items just got rewritten.
This data-report walks through every vendor move, the math on what it actually costs your team, why the SaaS AI pricing model is shifting from per-seat to metered, and a practical procurement playbook to keep the bill sane. Steal the table, run the numbers, send it to your CFO.
What Just Happened to SaaS AI Pricing in April 2026
SaaS AI pricing has been quietly mutating for two years, but April 2026 is when the dam broke. Four moves in 17 days reshaped how the biggest workplace tools charge for AI.
Microsoft, April 15: Tenants without a paid Copilot license lost in-app Copilot from Word, Excel, PowerPoint, and OneNote. Full removal applies to organizations with more than 2,000 users; smaller orgs got throttled. Microsoft also locked Copilot Business at $18/user/month only if purchased before June 30, after which the price jumps to $21/user/month. For a 500-person company, that is a $216,000 to $252,000 annual line item where there used to be free in-app AI.
Figma, April: Figma Make AI Administration shipped to Org and Enterprise plans, with the Pro tier following in May. Designers no longer get unlimited AI generations bundled in their seat — admins must purchase metered AI credits, the same per-use SaaS AI pricing model that triggered the Cursor and Replit billing backlash in 2025.
Notion, April 17: The April 17 release refactored Mail and Calendar, and Notion AI is now bundled only in Business ($18/user) or Enterprise. Free, Plus, and standalone AI add-on customers were quietly migrated. For a five-person startup on Notion Plus, getting back into Notion AI means an 80% increase in the per-seat bill.
Zoom, April: Zoom AI Companion 3.0 went GA and decoupled from Zoom Workplace as a $10/user/month standalone SKU, with an agentic web companion at ai.zoom.us. AI is no longer "included with the meeting" — it is a separate purchase, with separate governance, and a separate bill.
The pattern is unmistakable. Every major workplace vendor is rebuilding its SaaS AI pricing on a single thesis: AI is too expensive to give away inside the seat, and bundling AI into the base price was leaving margin on the table. That decision, taken vendor by vendor, becomes a structural shift in how every distributed team in the US has to budget AI.
SaaS AI Pricing in 2026 — The Vendor Move Table
Here is the consolidated picture, vendor by vendor. This is the SaaS AI pricing table almost no one has put in front of buyers yet, and it is the single most useful artifact in this article.
| Vendor | Pre-2026 AI access | 2026 AI pricing | Per-user delta |
|---|---|---|---|
| Microsoft 365 + Copilot | In-app Copilot in Word/Excel for many tenants | Removed for non-licensed; Copilot Business $18 → $21 after June 30 | +$216-$252/year |
| Notion | AI add-on $10, available on all plans | Bundled into Business $18+ only; no Plus add-on | +$96-$216/year per user |
| Figma Make | Unlimited AI generations in seat | Metered AI credits, admin-purchased | Variable; $0.02-$0.10 per generation |
| Zoom Workplace | AI Companion bundled in paid plans | $10/user/month standalone SKU | +$120/year |
| Loom (Atlassian) | AI summaries included | AI editing and translation gated to higher tiers | +$60-$120/year |
| Granola | Solo plan with AI included | New Business tier with AI Quotas | Variable |
| Otter.ai | Pro AI summaries unlimited | Pro AI minute caps; over-cap = pay-as-you-go | Variable |
| Atlassian (Rovo) | Rovo agents in trial | Rovo per-credit billing in 2026 GA | Variable |
A few notes on how to read the table. The "per-user delta" column is the realistic annual increase a previously-bundled customer faces in 2026. For metered SaaS AI pricing, the delta is variable by definition — that is the whole point and the whole risk. For per-seat increases, the delta is a hard line item your CFO can see immediately.
The aggregate effect, for a hypothetical 100-person team using Microsoft 365, Notion, Figma, Zoom, and Loom, is roughly $54,000 to $80,000 of new annual cost in 2026 versus 2025 — assuming the team keeps the same tools, the same tier shape, and the same level of AI usage. That is before any new agent SKU, any pilot of agentic AI, or any new vendor entering the stack.
Why SaaS AI Pricing Is Moving From Per-Seat to Metered
To understand where SaaS AI pricing goes from here, look at three forces that all point in the same direction.
COGS Reality
LLM inference is a real variable cost. Unlike storage or API calls, where the marginal cost approaches zero at scale, every AI generation has a measurable token cost on the vendor's side. When a vendor bundles "unlimited AI" into a $20/user seat, a single power user running 200 long-context generations a day can wipe out the gross margin on that seat. A metered SaaS AI pricing model passes that variability back to the buyer. From the vendor's P&L, it is the only model that survives the agentic era.
Per-Seat Math Broke
The traditional SaaS seat assumed one human consuming one application's worth of work. AI agents broke that assumption. According to Gartner, 40% of enterprise apps will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. One human plus three task-specific agents now consumes the work of five seats. Per-seat pricing becomes a leak. Usage-based AI pricing is the patch, and it is now the dominant new SaaS AI pricing model — Metronome's 2026 survey found 64% of SaaS companies now bill at least partially by usage, up from 39% in 2024.
Investor Pressure for AI Revenue
Public-market and late-stage SaaS investors want to see "AI ARR" as a separate line in the financials. Bundling AI into the base subscription hides that signal. Unbundling makes AI a discrete revenue stream that can be reported, modeled, and rewarded with a higher multiple. That, more than any feature decision, is why every major vendor is breaking out an AI SKU in 2026.
The result is a SaaS AI pricing landscape where the buyer carries more variability, the vendor protects margin, and the budget conversation moves from "how many seats?" to "how many tokens, credits, or minutes?" Buyers who do not adapt their procurement playbook to that shift will find their AI line items growing 30-80% year-over-year with very little fanfare.
What "Cheap" SaaS AI Add-Ons Actually Cost at Scale
The dangerous part of the new SaaS AI pricing reality is that no single add-on looks expensive. A $10 Zoom AI SKU. A $10 Notion AI delta. A $3 Microsoft Copilot uplift. A few hundred metered credits in Figma. Each line item is rounded to a coffee. Stacked across a 100-person company, it is a new mid-six-figure budget item.
Here is the rough math for a 100-person distributed team with a typical 2026 stack:
- Microsoft 365 + Copilot Business at $21 → $25,200 net new
- Notion Business uplift ($18 vs $10 Plus) → $9,600 net new
- Figma Make AI credits at average $30/user/month for design-heavy roles (20 users) → $7,200 net new
- Zoom AI Companion 3.0 at $10/user/month → $12,000 net new
- Loom AI tier uplift at $5/user/month (creators only, 30 users) → $1,800 net new
That is $55,800 of net new annual spend in 2026 for a 100-person team — and the team has not added a single new tool. Stretch the same math to 500 people and the number crosses $250,000.
Now layer in Zylo's finding that 54% of SaaS licenses go unused. A meaningful fraction of that $55,800 is being paid for accounts that do not log in this quarter. The combination of the new SaaS AI pricing structure plus existing shelfware compounds badly. We covered the underlying mechanics in our deep-dive on duplicate SaaS subscriptions and the broader SaaS sprawl problem.
A 2026 Procurement Playbook for SaaS AI Pricing
The Great Unbundling does not have to mean uncontrolled bill creep. Buyers who treat SaaS AI pricing as its own discipline — separate from "regular" SaaS procurement — recover most of the new spend. Four moves, in priority order.
Audit AI Shelfware Before It Auto-Renews
Pull a 90-day usage report on every AI add-on already in the stack. Anything below 30% utilization should be downgraded or cut at renewal. Apply the same rigor to AI seats as to any other seat — the Atlassian State of Teams 2025 report shows knowledge workers already lose 25% of their time to search and tool-switching, so an unused AI seat is not just wasted budget, it is also added cognitive load.
Demand AI Entitlements in Renewal Language
In 2026, the contract clause that matters most is "AI entitlements." Every renewal should explicitly list which AI features are included, which are metered, what the per-unit cost is, what the cap is, and what happens at the cap. Vendors that resist this are signaling that their SaaS AI pricing model will keep mutating, and your bill will mutate with it. Walk away or shorten the term.
Cap Metered Usage With Hard Guardrails
For any metered AI line item, set a hard monthly cap at the admin level. Most vendors offer this; few advertise it. Without a cap, a single power user or a runaway agent loop can spike a single month's bill by 4-10x. With a cap, the worst case is degraded service, not a surprise invoice. Treat metered SaaS AI pricing the same way a finance team treats AWS — with budgets, alerts, and CloudWatch-equivalent monitoring.
Consolidate Where AI-Included Pricing Still Exists
Some categories still have credible vendors with AI included in a flat seat price. Meeting and collaboration is one of them. Coommit, for example, includes contextual AI in the base seat — the assistant that sees both the canvas and the conversation is not a metered add-on, it is the product. When a category has a flat-AI option that consolidates two or three line items (video + canvas + AI summarizer), the math almost always beats stacking unbundled SKUs from three vendors. Our SaaS consolidation playbook walks through the diligence step-by-step.
The procurement teams that combine these four moves consistently absorb the Great Unbundling without growing their net SaaS spend. The teams that do nothing watch their 2026 AI line item land 25-40% over budget by Q3.
The Great Unbundling Is Permanent
The shift in SaaS AI pricing during April 2026 is not a temporary repricing. It is a structural rebuild driven by inference economics, the rise of agents, and investor demand for cleanly reported AI revenue. Every workplace vendor that has not yet unbundled AI is now planning to. Buyers who price 2026 budgets on 2025 assumptions will end the year over plan, every time.
The good news is the playbook is straightforward: audit shelfware, write entitlements into renewals, cap metered usage, and consolidate where flat-AI pricing still exists. The teams that do this in Q2 2026 will end the year with the same productivity and the same AI access at meaningfully lower cost than the teams who let the unbundling happen to them. If you are evaluating where to consolidate first, video and canvas collaboration is the highest-leverage starting point — it is the category where AI-included pricing still beats stacked add-ons by the widest margin, and it is the category most directly hit by the new SaaS AI pricing model.