What is AI credit pricing?

AI credit pricing is the SaaS pricing model — dominant in 2026 — where AI features are sold via consumption-based credits rather than included in a flat per-seat subscription. Customers buy a credit pack, each AI action (prompt, generation, agent run) deducts credits, and overage above the included tier triggers automatic top-up charges. Figma Make, Notion Custom Agents, Microsoft Copilot Studio, ChatGPT Workspace Agents, and Zoom AI Companion 3.0 all use AI credit pricing.

Why are SaaS vendors moving to AI credit pricing in 2026?

Three reasons. AI inference cost is variable and high, and AI credit pricing shifts that volatility from vendor margin to customer P&L. Credit packs unlock a second budget pocket — departments buy them outside per-seat IT contracts. And paid credit balances create a sunk-cost switching tax, increasing customer lock-in. Per Zylo's 2026 SaaS Management Index, US enterprise SaaS spend rose 8% YoY in 2026 even as app counts went flat, almost entirely via AI credit pricing add-ons.

How do I forecast AI credit pricing costs accurately?

You cannot forecast AI credit pricing accurately, which is the core problem CFOs flagged in 2026. Credit consumption depends on prompt patterns, agent chain length, and tool-call counts that change weekly. Best practice: run a 30-day baseline measurement with a credit cap, calculate cost-per-outcome (not cost-per-credit), apply a 50% safety margin, and renegotiate annual contracts to include hard caps with rollover credits.

What's the difference between AI credit pricing and per-seat pricing?

Per-seat pricing is fixed: $X per user per month, AI features included up to fair use. AI credit pricing is variable: $X per credit pack, every AI action consumes credits, overage charges apply automatically. Per-seat is forecastable; AI credit pricing is not. Per-seat caps vendor revenue at headcount; AI credit pricing has no upper bound. In 2026, most workspace AI tools shipped with AI credit pricing as the default and per-seat as a legacy enterprise option.

Which AI tools still use predictable per-seat pricing in 2026?

Per-seat AI inclusion is now the differentiator, not the default. Tools like Coommit (canvas + video + contextual AI bundled in the seat), some open-source self-hosted stacks, and select enterprise contracts at Microsoft, Google, and Salesforce still offer flat per-seat AI. Most workspace AI tools require active negotiation to convert AI credit pricing into a flat ceiling. Audit your stack quarterly to catch credit-meter pivots in renewals.

The AI Credit Pricing Trap: How SaaS Doubled Your 2026 Bill

# The AI Credit Pricing Trap: How SaaS Doubled Your 2026 Bill

In April 2026, the Figma community forum lit up with designers reporting AI credits "drained every single day even when not making changes." Some were burning $120 to $240 per month per seat on credit packs because real product flows in Figma Make consume 21,500 to 30,000 credits versus the 3,000 included. Other threads documented 2x overage charges when users blew past their balance with no warning.

That was one vendor in one month. AI credit pricing is now the dominant SaaS pricing model of 2026, and almost every workspace tool your team uses has quietly moved to it: Figma, Notion, Microsoft Copilot Studio, ChatGPT Workspace Agents, Loom, Zoom, Cursor, Lovable. The result: US enterprise SaaS spend rose 8% in 2026 even as app counts went flat, per Zylo's 2026 SaaS Management Index. The same report flagged ChatGPT as the #1 most-expensed app in US enterprises.

This deep-dive maps the AI credit pricing trap across the 2026 SaaS stack, explains why vendors pivoted, breaks down the three hidden costs nobody mentioned in the launch keynote, and ends with the five-step audit you can run on your own AI credit pricing exposure this quarter.

The 2026 Data Stack: Six Receipts That Prove the AI Credit Pricing Trap Is Real

AI credit pricing is not one vendor's quirk. It is the 2026 default. The receipts are stacking up across every category of workspace tool.

Figma Make's credit cliff. Real product flows hit 21,500 to 30,000 credits per design iteration; included credits = 3,000. Designers paying $120 to $240 per month per seat in overage packs. Coverage in Medium's April analysis called the iteration tax "scandalous." Cancellations are public on the forum.

Notion Custom Agents' May 4 metering cutover. Notion shifted Custom Agents from a flat $10/seat add-on to $10 per 1,000 credits effective this week. A typical RevOps team burning agents on weekly forecast roll-ups will exhaust the included tier in days, not weeks. AI credit pricing turns a fixed line item into a variable one mid-quarter.

Microsoft Copilot Studio's $200/25,000-credit ledger. Microsoft's Copilot Studio billing model packages "Copilot Credits" on top of a $30/seat Copilot license. With Microsoft 365 E7 launching May 1 at $99/user, the "all-in-one" AI bundle still ships with metered credit overage on top.

ChatGPT Workspace Agents' May 6 cutover. OpenAI moves Workspace Agents from free preview to credit-metered paid tier on May 6 — three days from this article going live. Buyers who built workflows during the free window are now scrambling to estimate credit consumption per agent run.

Loom's Atlassian-era 100x cost story. Loom's post-acquisition repricing hit some teams with effective per-recording costs 100x what they paid in 2024. Many migrated to alternatives covered in our Loom alternatives breakdown.

Zoom AI Companion 3.0's $10 standalone tier. Zoom now sells AI Companion as a $10 standalone add-on on top of Workplace seats — formalizing AI credit pricing as a permanent line item rather than a feature.

The macro number: only 6% of organizations qualify as AI "high performers" attributing more than 5% of EBIT to AI, per McKinsey's State of AI 2025. Forrester predicts enterprises will defer 25% of planned 2026 AI spend to 2027. The buyers paying AI credit pricing premiums are not seeing the EBIT to justify them.

Why Vendors Pivoted to AI Credit Pricing in 2026

AI credit pricing did not appear by accident. Three vendor incentives converged in late 2025 and made it the new default for 2026.

Inference cost is real and uneven. Unlike storage or seats, AI cost-per-action varies by model, prompt length, and tool calls. Per-seat pricing assumes a stable cost per user. AI credit pricing pushes that variability onto the buyer instead of the vendor margin. From the CFO seat at Figma or Notion, AI credit pricing is risk-shifting — the customer absorbs the inference volatility.

AI credit pricing opens a second budget pocket. Per-seat budgets are owned by IT or RevOps and capped at renewal. Credit packs are purchased by individual teams using their departmental P&L, often without procurement review. Vendors pivoting to AI credit pricing are mathematically expanding the surface they can sell into. This is land-and-expand 2.0 — and it works in 2026 because Zylo data shows ChatGPT alone surpassed Salesforce as the #1 most-expensed app. Once that pattern lands, every other vendor copies the playbook.

AI credit pricing creates lock-in via sunk credit balance. A team that has paid for 100K Notion credits is unlikely to switch tools mid-quarter even if a competitor lands a better feature. Credit balance becomes a switching tax that did not exist under flat per-seat pricing. Harvard Business Review documented this dynamic in February 2026: AI rollouts intensified work because saved time was reabsorbed into more output expectations. AI credit pricing extracts the value of that intensification by tying every additional output to a metered token.

The combined effect: enterprise SaaS spend rose 8% YoY in 2026 even though app counts are flat. The growth is happening inside existing contracts via AI credit pricing add-ons, not new logos. We unpacked the broader cost dynamic in our deep-dive on AI tool sprawl — AI credit pricing is the financial expression of that sprawl.

The Three Hidden Costs of AI Credit Pricing

The sticker on AI credit pricing is straightforward: $X per 1,000 credits. The hidden costs are not. Three categories of hidden cost rarely appear in the procurement conversation.

The iteration tax

AI tools fail. Designers retry Figma Make prompts 3+ times when the first generation misses. Each retry costs credits at full rate, even when the failure is on the model. The Figma forum threads explicitly call this out — credit packs drain on AI errors users had to fix manually anyway. Under per-seat pricing, retries were free. Under AI credit pricing, they are revenue. The iteration tax has no upper bound, which is why some Figma seats in April 2026 burned through a $240/month overage pack in 18 days.

The forecasting failure

CFOs cannot budget AI credit pricing accurately. Credit consumption depends on prompt patterns, agent chain length, and tool calls — variables that change weekly as teams adopt new patterns. A single Notion Custom Agent rolled out to 50 reps could consume 0 credits or 250K credits in a month. The variance kills procurement's ability to negotiate annual contracts and creates the surprise-bill pattern that defines AI credit pricing in 2026.

Sunk-credit lock-in

Once a team has 100K credits on the Notion ledger, switching to a competitor with better AI agents writes off that balance. AI credit pricing is the most effective lock-in mechanism the SaaS industry has invented in a decade. It works precisely because credits are unfungible — you cannot port a Notion credit balance to ChatGPT Workspace Agents or vice versa. The switching cost is no longer "data export and retraining users"; it is "data export and retraining users plus losing six figures of paid credits."

These three hidden costs explain why Forrester is predicting 25% of 2026 AI spend gets deferred to 2027. CFOs who could not forecast AI credit pricing accurately in their 2026 budget are now slow-rolling new commitments until they understand consumption patterns.

The Five-Step AI Credit Pricing Audit Every CFO Should Run This Quarter

The 2026 AI credit pricing trap is fixable, but only with deliberate procurement work. Five steps, in order of impact-per-hour.

1. Inventory every credit-metered tool

Pull every AI add-on across the stack. Categories to check: design (Figma, Adobe Firefly), productivity docs (Notion, Coda, ClickUp), code (Cursor, Lovable, Replit, GitHub Copilot), video (Loom, Zoom AI Companion, Granola, Otter, Fireflies), workspace agents (ChatGPT Workspace Agents, Microsoft Copilot Studio, Slack AI), and content (Jasper, Writer, Anthropic Claude API for Workflows). Most enterprises will surface 8 to 14 distinct AI credit pricing line items. If you cannot list them in 30 minutes, your shadow AI exposure is bigger than the AI credit pricing exposure — see our shadow AI risks playbook for that audit.

2. Map credits to actual user value

For each credit-metered tool, calculate cost per outcome — not cost per credit. Example: Notion Custom Agent at $10/1k credits, average forecast roll-up burns 200 credits = $2 per roll-up. Multiply by frequency. The unit economics expose which tools are paying for themselves and which are pure margin extraction.

3. Flag the 80/20 credit hoarders

In every credit-metered tool, 20% of users will consume 80% of the credit pool. Identify them. They are either your highest-ROI power users or your least-trained credit-burners. Both need attention — the first to negotiate volume pricing, the second for usage retraining. AI credit pricing rewards ignoring this analysis with a renewal surprise.

4. Run the switch-cost math

For each AI credit pricing line item over $20K annual run-rate, calculate full switch cost: data export, user retraining time, lost credit balance, parallel-run period, integration rework. Compare to a 12-month forward AI credit pricing forecast at current consumption growth. In 2026, switching cost is increasingly lower than running cost — especially for tools whose AI is non-differentiated.

5. Renegotiate or replace

Walk into renewals with the audit numbers. Vendors will trade flat seats for credit hoarder caps if the alternative is churn. Tools where they will not negotiate are tools you should replace. The renewal-cycle leverage is highest in Q2 and Q4 — the AI credit pricing buyer's market is real if you bring data.

What to Demand in 2026 AI Tool Pricing

Going forward, the procurement bar for any new AI tool should rise. Three minimums.

Predictable inclusion. AI capability priced at the seat level, with a generous floor of included usage that covers 95% of normal workflows. Overage tier exists for power users but is not the default revenue mechanism.

Open data export. No proprietary data jail. The Salesforce-Slack lockdown of April 2026 made every IT team newly skeptical of vendors that hoard conversational data behind API restrictions.

BYO-model optionality. The right to point a tool at your own LLM via OpenRouter, Anthropic API, or self-hosted deployment. AI credit pricing without a BYO-model escape hatch is permanent margin extraction.

This is the bar Coommit was built for. Coommit packages contextual AI inside the seat — meeting, canvas, and workspace AI in one surface — with no per-prompt credit meter, no overage panic, and no AI add-on SKU to negotiate. For teams burning $120 to $240 per seat per month on Figma Make overage, the math against an inclusive-AI alternative is no longer close. We compared the broader AI workspace agent stack in our 2026 AI workspace agents comparison — the pricing model is now the primary differentiator.

The AI credit pricing trap is not going away on its own. McKinsey's 6% high-performer floor and Forrester's 25% deferral both point to a market that is starting to reject the model. The vendors that move first to predictable inclusion will take share. The ones that double down on AI credit pricing will hit the renewal-cycle ceiling in Q3 and Q4 2026.