Stripe's AI Index Shows 103% Revenue Growth, But the Real Story Is the Cool-Down
6 min read · 15 sources
- Stripe's AI Index shows 103% aggregate revenue growth across 20,000 AI companies, cooling to 80% annualized from 140%.
- Lovable hit $600 million ARR and 1 billion monthly visits to built apps in two years.
- OpenAI's ChatGPT app store gives developers a new distribution channel with Spotify, Canva, and DoorDash as launch partners.
- KeyBanc finds 43% of B2B vendors have customers using AI coding tools to rebuild their products.
- 84% of AI answers to the same 210 questions changed when asked two days in a row.
Stripe finally put a number on the AI boom, and it’s a big one: 103% aggregate revenue growth across nearly 20,000 AI-dependent businesses over the past year. But the number that matters more is the one underneath - recent annualized growth has cooled to 80%, down from 140% just a few months ago. The AI gold rush isn’t over, but the easy money phase is.
The Stripe AI Index is a weekly measure of same-business payment activity, modeled on the company’s existing SaaS Index. It’s the first time founders have had a hard, recurring benchmark for what AI companies actually earn, not what they raise. And the median tells a different story than the aggregate: median growth was 41%, meaning the big players are pulling the average up hard. US companies grew 104%, versus 83% in the UK and 67% in the EU. If you’re benchmarking your AI business against “the market,” pick your geography carefully - the spread is wider than most people assume.
Aggregate AI revenue grew 103% over the past year, but recent annualized growth has cooled to 80% from 140% just a few months prior.
The new App Store nobody's optimizing for
Source: operatorsnotebook.beehiiv.com ↗
OpenAI’s DevDay launched apps that run inside ChatGPT conversations, and the distribution play is real: Spotify, Canva, and DoorDash are launch partners. The apps use MCP servers and small interfaces to complete actions without leaving the chat window. For developers, this is a new storefront where your tool’s description in ChatGPT is the listing, and ranking is whatever OpenAI decides it is.
The Operators Notebook post makes the warning explicit: OpenAI controls ranking, may take a cut of transactions, and owns the user relationship entirely. Build one if your core action finishes in one or two chat messages and users already ask ChatGPT for it. Treat it as cheap distribution that never fully belongs to you - a lead generation channel, not a moat.
Lovable: $600M ARR with no growth hack
Lovable turned two this week with over $600 million in annual recurring revenue, 1.2 million new projects weekly, and 1 billion monthly visits to apps built on the platform. The company’s growth lead, writing in Elena Verna’s newsletter, attributes this to the “Whitewater Framework” - growth comes from riding a fast-moving market, not executing a playbook. AI is the river; Lovable built a raft.
The operational details matter more than the metaphor. The lean team ships big updates nearly every week to keep word of mouth moving, and new hires are expected to automate the repetitive parts of their own jobs with AI from day one. That’s not a growth tactic; it’s a hiring filter. For engineers, the lesson is that in an exploding category, standard growth tactics that failed at older software companies now work simply because the tide lifts everything. The risk is mistaking the tide for your own strength.
Your next GTM hire: a creative marketer, not another engineer
Cold email volume has exploded now that anyone can launch a massive campaign with Claude Code in minutes. About 91% of cold emails get no reply, and everyone can automate the sending now. The argument from Alex Vacca is that the next hire after a GTM engineer should be a creative marketer who owns the angle and the copy - the things AI can’t judge.
The data backs it: reply rates rose from 3.4% to 15% when sending on fresh signals like funding rounds or new sales hires, and narrative skills appear in 29% of senior content marketing roles, up from 8% in 2023. Score that hire on replies and message tests, not on production volume. The production is commoditized; the positioning is not.
Employee-led content: 21.1 million views in 30 days
Thirty Vercel employees generated 21.1 million views on X in 30 days. The playbook is straightforward: pick 5-10 writers, use AI for ideation and drafting, appoint a content champion, and build a daily leaderboard with tools like Opus 5.5 or a Grok bot.
What kills these programs is predictable: copy-paste captions, forced posting quotas, and posts that only promote the company. People can smell a mandate. The leaderboard works because it gamifies authenticity without scripting it - but it only works if the writers actually have something to say.
Microsoft Advertising's biggest bidding upgrade in a decade
Source: about.ads.microsoft.com ↗
Microsoft Advertising rolled out its largest bidding model upgrade in ten years, doubling optimization speed and pacing budgets more evenly. The results for advertisers: 10% more conversion value per click on Maximize Conversion Value or Target ROAS, 8% greater conversion value switching from Target CPA to Target ROAS, and 18% more conversions moving from eCPC to Maximize Conversions with a CPA target.
The catch is conversion tracking. The new models only work if you feed them real transaction values, not proxy signals. If your tracking is broken, you’re not getting the 10% - you’re getting noise.
Vercel Billing CLI: billing without leaving the terminal
Vercel grouped every billing command under vercel billing in CLI 65.17.0. You can check your plan, download invoices, and preview upcoming charges with commands like vercel billing overview and vercel billing invoices inspect. The --json flag makes it scriptable for agents and cron jobs, and the old vercel usage and vercel budgets commands remain as aliases.
It’s a small thing, but for teams running cost monitoring on Vercel spend, it removes a dashboard trip from every billing check. Automation-friendly billing is the kind of unglamorous feature that saves real time across a team.
AI answers: 84% change overnight
Supercite asked the same 210 questions to four AI engines two days in a row. 84% of the answers changed - 75% after correcting for formatting variance. Twenty-seven percent of brands disappeared entirely from answers, and the #1 pick retained its spot only 63% of the time. The instability traces to the pages AI reads: only 42% of cited pages were the same day-to-day, so swapping pages swaps answers.
For anyone doing SEO or brand monitoring, this is the warning: don’t rely on a single AI recommendation check. The volatility is structural, not a bug in one model. If you’re tracking your brand’s AI presence, you need a baseline across multiple days and engines, not a single screenshot.
KeyBanc's 2026 survey: 43% of vendors watching customers rebuild their products
KeyBanc’s survey of 81 private B2B companies (median $22M ARR) found 24% median ARR growth in 2025, with plans for 23% in 2026 and 40% in 2027. Only 6% met the Rule of 40 in 2025. But the scariest number is 43%: that’s the share of vendors saying customers use AI coding tools to rebuild their products - 55% at sub-100-employee companies. Only 41% charge for AI, typically bundled with a 20% median premium.
Heavy AI spenders (over 5% of revenue) run 66% gross margins versus 78% for light spenders. The AI boom is real, but the economics are bifurcating: companies that charge for AI are growing, and companies that don’t are watching their customers vibe-code the product themselves.
The rest
The AI social loop that generated 25 million views in seven weeks from 919 posts is worth reading for the mechanics: post three times daily without human approval, score core ideas, rest winners for a month. Views are cheap - only 7,000 followers gained - so treat it as brand awareness, not lead gen. The M&A process piece reminds founders that bankers leak, deals die early, and you tell the fewest people possible until the definitive agreement is signed. And the cold-email-Sam-Altman advice is simple: email the CEO asking who the right person is, cc the CEO on the follow-up, and the target is almost forced to take the meeting.
You May Also Like
Stripe's Index Says Non-AI SaaS Grew 30% Through the $2T Sell-Off
Software stocks lost an estimated $2 trillion this year on AI fears, but Stripe's new weekly index of 72,000 non-AI B2B businesses shows revenue up over 30% …
OpenAI’s “future of the firm” vision and Databricks’ 7‑billion‑run‑rate shake up enterprise AI
OpenAI’s new essay maps AI agents onto every corporate role, turning every employee into a manager of bots without shrinking headcount. Databricks hits a $7 …
The Robots Aren't on the Line Yet — and Other Hard Truths
Robotics costs are falling 40% a year and a leading maker's bots have logged 65,000 hours, but factory visits show the machines aren't actually on production …




