Credit Convergence: Q1-Q2 2026's Universal Currency Rush

Every major SaaS company just invented its own currency. And none of them convert.

Between January and May 2026, at least eight companies launched, enforced, or restructured credit-based AI metering. Credits went from an exotic pricing mechanic to the default way incumbents monetize AI. The PricingSaaS 500 Index counts 79 companies (15.8%) using credits now — up 126% YoY from 35 at end of 2024 [PricingSaaS Q2 2026 Report — no public URL].

This isn't coordination. It's convergence. Every company hit the same wall independently: per-seat pricing breaks when AI agents do the work of humans. Credits are the patch.

The wave

Eight companies. Six months. Same conclusion.

Seven currencies, zero interoperability

Here's what procurement teams are dealing with:

None of these convert. A Copilot Credit is not a Figma credit is not an SAP AI Unit. A procurement team evaluating Microsoft vs. Salesforce vs. ServiceNow can't compare credit-per-action economics without a spreadsheet and three vendor calls.

This is by design. Credits obscure unit costs, make competitive comparison nearly impossible, and create switching costs through sheer incompatibility. AI credits in 2026 are where cloud compute pricing was in 2008 — every player inventing its own unit because nobody has established the standard yet.

The counter-trend: Windsurf drops credits

Not everyone is joining the rush.

On March 19, 2026, windsurf replaced its monthly credit pool with daily/weekly usage quotas. Users complained that credits created anxiety — "how many credits do I have left?" became a recurring friction point. Daily quotas give users a fresh allowance every morning instead of a dwindling monthly balance.

Windsurf simultaneously raised Pro from $15 to $20/mo (+33%) and added a $200/mo Max tier. Pricing went up, metering got simpler.

This matters because Windsurf competes directly with github Copilot and cursor, both of which use credit/request-based systems. If Windsurf's retention holds, it challenges the assumption that credits are inevitable.

Credits and the fee retreat are the same story

The credit convergence and the fee-retreat are happening simultaneously. They're two sides of the same coin.

The fee retreat shows companies cutting existing credit allocations: relay slashed AI credits from 5,000 to 2,000/mo (-60%). replit trimmed Core credits from $25 to $20/mo (-20%). elevenlabs cut included minutes across plans.

The credit convergence shows companies adopting credits for the first time: Microsoft, GitHub, Figma, Atlassian, SAP. These are companies that had flat per-seat pricing and are adding a consumption layer.

Combined pattern: new entrants launch generous credit pools to drive adoption, then ratchet them down as costs become clear. The fee retreat is what credit convergence looks like 12-18 months later.

The scorecard

• 79/500 (15.8%) of PricingSaaS 500 companies now use a credit model
• 126% YoY growth from 35 companies at end of 2024
• 50% of ServiceNow's net new business is non-seat-based — highest reported by any legacy vendor
• $750M Now Assist ACV, targeting $1.5B for 2026
• 60% of Microsoft Dynamics 365 customer service customers purchase usage-based credits
• 8 companies in this dataset launched or enforced credits in a 6-month window

What happens next

Short-term (2026-2027): more fragmentation. Each vendor makes their credit system more complex — tiered credits (SAP's Basic/Standard/Advanced agents), credit pools with different refresh rates (daily vs. monthly), bundled vs. add-on credits, rollover vs. no-rollover. Complexity becomes the moat. Switching costs go up. This favors incumbents.

Medium-term (2028+): standardization pressure from buyers. A third-party benchmark emerges — "AI action units" or equivalent — that lets procurement compare credit costs across vendors. Analogous to how TPC benchmarks standardized database pricing. This will come from buyers, not sellers.

The wildcard: if AI inference costs drop another 10x (as they did from 2023 to 2025), credits may fold back into flat pricing. The companies building credit infrastructure today may be metering a cost structure that won't exist in three years. But nobody wants to be the one who gave away unlimited AI and discovered the costs were real. See the fee-retreat for what that looks like.

May 2026 update

The credit wave continues accelerating:

New signal: API toll gates. ServiceNow's Action Fabric and SAP's API lockdown represent the enterprise response to the credit wave — not just metering their own AI, but gating external agents' access to their data. If credits are the new currency, Action Fabric is the customs checkpoint. Companies wanting to run AI agents against ServiceNow or SAP data will now pay per operation.

Metronome taxonomy. The Metronome "50+ AI Pricing Models" report identifies three credit functions: compute proxies, abstracted bundles, and access gating. Our dataset confirms all three are in production: compute proxies (Anthropic API tokens), abstracted bundles (GitHub AI Credits, Figma credits), and access gating (SAP AI Units, ServiceNow Assist Credits).

Anthropic billing split (June 15). Anthropic is splitting subscription billing into first-party vs. third-party agent/SDK usage. New Agent SDK credit pools: Pro $20/mo, Max 5x $100/mo, Max 20x $200/mo. This is the first major AI provider to separate credits for agents that others build from credits for their own product. It creates a two-sided credit market: your Claude Pro subscription fuels your own usage, but agents built by third parties draw from a separate pool.

Intercom at $100M+ ARR on per-resolution. Poyar confirms Intercom's Fin AI agent has crossed $100M ARR on pure outcome-based pricing ($0.99/resolution). This is the highest-profile evidence that outcome-based can scale — but it's still the exception, not the rule. Intercom can do it because resolution is measurable, attributable, and predictable (the CAMP framework holds). Most AI features fail at least one of those criteria.

See also: ai-credits, fee-retreat, free-tier-economics, ai-bundling