Week of November 8, 2026
The best writing on how AI companies charge and make money, picked every week and linked to the relevant business models on this site.
- Why the Labs are Giving away API Credits
Anthropic and OpenAI now give top subscribers monthly API credits that match their plan price, such as $100 on a $100 plan, while cheaper tiers get none. The article argues this steers heavy users from low-margin chat toward the API, which SemiAnalysis roughly estimates at 92% gross margin versus 50% for subscriptions. It frames the consumer subscription as acquisition and pay-per-token APIs as expansion, borrowing the SaaS habit of bundling credit allowances into plans.
- ChatGPT ad visits earn shops $0.89 against $2.45 for Google search ads
A working paper tracking 1,898 online shops finds that a US visitor from a paid ChatGPT ad brought in $0.89, versus $2.45 from a Google search ad. The authors put a fair ChatGPT click price near $0.26, far below the $1.72 one large account paid. Ads earn roughly $1 per free user yearly against about $5 in inference cost. So ad-supported AI cannot yet fund a free consumer tier unless each click becomes far more valuable.
- Figma Gave Up to 2x AI Credits at Same Price; Shares Fell 7%
Since August 2026, Figma's paid AI credit packs give Professional customers twice the credits and Organization and Enterprise customers 60% more, with no price change. That cuts the effective cost per credit by 50% and 37.5%. The CFO described the move as giving up some revenue now to grow usage of its AI add-ons. Design agents that just left beta now draw on those credits, and shares dropped 7% after results.
- Figma Halved Pro AI Credit Price; Salesforce Charges $0.005
List prices for AI credits now run roughly $0.004 to $0.01 each at Salesforce, Atlassian, Notion and Adobe, so heavier use shows up on the customer's bill. Figma went the other way, doubling the Professional plan's credit allowance at an unchanged price, while its hosting and infrastructure spending climbed $52.8 million, about 45% of its revenue gain. The piece frames credit-based subscriptions and AI add-ons to existing SaaS as a choice over who absorbs inference cost.
- How to Make Credit-Based Pricing Work: Design, Sell, and Scale a Credit Model
Pricing adviser Ulrik Lehrskov-Schmidt argues that credits combine upfront cash like a license with flexible terms, and notes about 320 credit systems now appear on tracked pricing pages, roughly 8% of the market. For credit-based subscriptions, he advises keeping the dollar-to-credit rate simple and stable while adjusting what a credit buys inside the product. He also recommends full rollover tied to renewal, one shared pool for most products, and account-level spending controls, since 1% of users often drive 75% of AI cost.
- ChatGPT Searches Near 1/5 of Google’s; Ads Top $1B Run Rate
OpenAI says ChatGPT's ad business passed $1 billion in annualized revenue within about 200 days, while weekly users grew to roughly 1.2 billion. That ad money sits alongside subscriptions and credits, showing ad-supported AI can scale fast next to a consumer subscription. Still, ChatGPT handles about a fifth of Google's search-like query volume, yet its ad run rate equals only around 0.4% of Google Search's annualized revenue, so money earned per query lags far behind.
- [Figma Q3 2026 Earnings Call] Figma Revenue Jumps 42% to $391M as AI Credits Ignite 131% Net Dollar Retention
Figma's Q3 2026 revenue rose 42% to $391 million, and AI credits helped push net dollar retention to 131%. Gross margin slipped from 85% to 82% because Figma absorbed inference costs for its agent before launch. After an August repricing gave each AI add-on more usage credits, add-on use rose about 40%, drawdown deals made up 80% of eligible purchases, and some enterprises lifted credit commitments six to eight times.
- How AI Applications Earn Their Premium
a16z argues that cheaper, stronger models make AI apps easier to replace, so a premium must be earned through better workflow execution, context customers would rather rent than rebuild, and proven value per unit of work. It urges pricing on recognizable completed tasks or credits mapped to tasks, not tokens. Clay's example splits Data Credits from Actions and passes reasoning-model token costs through with no markup, a path toward per-task pricing.
- Anthropic Revenue: $11.6B in Q2 2026 and the Run-Rate History
Anthropic's S-1 shows Q2 2026 revenue of $11.6 billion, up from $787 million a year earlier, after $4.6 billion for all of 2025. Pay-per-token API consumption brought in 80% of first-half 2026 revenue, or $13.1 billion, while Claude subscription plans added $3.2 billion. The filing defines run-rate as 28 days of usage fees times 13, plus monthly subscriptions times 12. By that measure, run-rate passed $47 billion in May.
- ChatGPT Credits: What They Cost and How They Work
OpenAI sells ChatGPT credits as a top-up once Plus or Pro usage runs out, and this guide works out the hidden price at $0.04 each. That rate matches API list prices exactly, so credits are the same token meter in a different unit. It is a clear case of subscription with usage credits layered on a consumer AI subscription, with credits expiring after 12 months and an $80 monthly top-up marking the point where upgrading to Pro makes more sense.