Stripe's Index Says Non-AI SaaS Grew 30% Through the $2T Sell-Off
7 min read · 16 sources
- Stripe's new index of 72,000 non-AI B2B firms shows revenue up 30%+ YoY, 4% above its pre-sell-off trend, despite the $2 trillion software stock crash.
- YC Series A rounds have fallen off a cliff since 2023, hitting all batches regardless of founder quality.
- AWS now lets Claude Code, Codex, and Cursor create accounts and deploy without manual sign-up, returning auto-refreshing short-lived credentials.
- Anthropic's linked billing lets third-party platforms charge Claude usage to a customer's own account without sharing API keys.
- Earthly shut down its managed build runner business in 2025 because teams got the speed themselves and never bought Satellites.
Software stocks lost an estimated $2 trillion early this year on the theory that AI would eat business apps. Stripe’s new weekly index of the companies that bill through it says the obituary was written too soon: non-AI B2B revenue is up more than 30% year over year, about 4% above its pre-sell-off trend. That’s not a rounding error, and it’s not the enterprise names that cratered.
The index covers roughly 72,000 non-AI B2B businesses paying by card, mostly self-serve and product-led. Those are the long tail, not the seat-based enterprise vendors whose net revenue retention everyone watches. If you want to know whether the SaaSpocalypse actually killed demand, this is the earliest signal you’ll get - and it says the long tail never stopped growing.
A company that shuts down also drops out of the index instead of counting as a loss, so the 30% figure is a floor, not a ceiling.
Stripe's Index Contradicts the $2 Trillion SaaSpocalypse
Stripe’s new SaaS Index is built from weekly payment flows across ~72,000 non-AI B2B businesses. Year-over-year revenue growth sits above 30%, about 4% above the pre-sell-off trend. Growth is strongest for companies under a year old, and mature companies have stopped slowing - the two facts together suggest new demand is still being created, not just existing contracts renewing.
Read the methodology before you cite it, though. The index covers mostly self-serve, card-paid, product-led businesses, not the large public enterprise companies that drove the sell-off. A company that shuts down drops out of the index instead of counting as a loss, so the 30% is survivorship-biased by construction. It’s an early indicator of long-tail health, not a signal about enterprise seat pricing power - for that, watch public companies’ net revenue retention.
Why YC Series A Rounds Fell Off a Cliff
Source: jaredheyman.medium.com ↗
Series A rounds for Y Combinator startups have declined drastically since 2023, and the cause isn’t founder quality. Rising interest rates cut global early-stage funding across every YC batch, regardless of admission year or how good the team was. The downturn was indiscriminate: good companies and bad companies alike found the Series A door shut.
Recent signs point to a recovery, but the essay makes a sharper claim: raising Series A may become less necessary. AI has cut the cost of building, and capital efficiency is now a feature, not a bug. If you’re a YC founder planning a raise, the takeaway isn’t “the market is bad” - it’s that the market may never come back in the same shape, and your business model should not assume it will.
Technical Founders: Win the Technical Battle First
Dakota McKenzie of Dynamic Growth Partners argues that even the best technology fails if its value isn’t legible to both technologists and buyers. His firm takes a hands-on operating model, and he assesses founders by observing decision quality, learning speed, and ownership - not by pitch meetings. The “technical win” should come before the commercial one.
For engineers considering fundraising, this is a different bar than the usual growth-at-all-costs pitch. Investors like Dynamic prioritize building enduring products and adoption over immediate commercial success. If you’re raising, expect to be evaluated on how you think, not how you present.
Brand as Infrastructure: Ramp's Slack Agent Builds the Decks
Source: links.tldrnewsletter.com ↗
Ramp’s brand team got tired of remaking the same decks and one-pagers for other teams, so they handed that work to a Slack agent. The agent builds collateral from approved examples the designers make in Figma. Automated tests catch broken layouts and missing fonts before anything goes back to the person who asked.
The numbers are the story: one contractor automated about 90% of the company’s whitepaper work in her second week. Designers now spend their time on work nobody has made yet, and each new hand-made deck goes back into the system as an example. This is the pattern for in-house creative work going forward - the humans make the first instance, the agent makes the next hundred.
AWS Lets Coding Agents Create Accounts for You
AWS’s new agentic account creation lets coding harnesses like Claude Code, Codex, and Cursor open a sign-up experience directly. Developers without an AWS account can deploy workloads without manually navigating to aws.amazon.com. The browser automatically completes sign-up and returns short-lived credentials that refresh automatically, with the first project pre-configured with a home Region and IAM role manager.
Try it by configuring the AWS Agent Toolkit via the MCP server setup guide, or run aws signup or aws login from the CLI. The friction of account creation was a real adoption barrier for agentic workflows - removing it means the agent can go from zero to deployed without a human touching a browser.
Anthropic's Linked Billing Kills API Key Sharing
Anthropic’s linked billing is now in beta on the Claude Platform, and it solves a real operational problem. Third-party platforms like coding agents can offer Claude API usage to end customers without sharing API keys. Customers link their Claude Platform organization, set monthly spend limits per platform, and can unlink anytime. Platforms see per-customer usage via Console or the Usage and Cost Analytics API.
The security win is obvious: no API keys change hands. The business-model win is subtler - platforms can decide per-request whether they or the customer pays, which means no building metering or invoicing from scratch. For anyone running a tool that resells API access, this is the pattern to copy.
ChatGPT for Sales: OpenAI Goes After the CRM Workflow
ChatGPT for Sales helps sales teams prepare for calls, tailor proposals, and follow up faster using information from the tools they already rely on. The new Sales plugin from OpenAI provides ready-to-use workflows for everyday sales work. Availability of partner plugins will vary by plan, region, and workplace settings.
This is OpenAI moving up the stack from chat to workflow automation. The pitch is less time on busywork, more informed decisions. For founders, the question is whether this displaces the point solutions you’re already paying for - and whether your sales team will actually use it.
Earthly's Post-Mortem: Developers Adopted, Business Didn't Follow
Earthly’s commercial failure is the canonical dev-tools cautionary tale. Developers adopted the free Earthfiles build tool, but few bought Satellites, the managed build runners - because teams could run persistent build environments themselves and get similar speed. Broader adoption stalled because migration work was hard to prioritize, and interviews revealed a company-wide problem where even simple fixes sat unmerged for quarters.
The pivot to Lunar, an engineering guardrails product, addresses a problem engineering leaders already own with a budget they control, rather than relying on organic adoption. The lesson for anyone building dev infrastructure: check whether your free version already delivers the value you plan to charge for.
The Death of the Backlog
The traditional product backlog is dying, argues this essay from experience at Apple (Radar), Vouch (Jira), and AngelList (no formal system, work picked from Slack). At AngelList, engineers resisted backlogs as “where tickets go to die,” but the author defends that as a feature - filtering requests is part of an engineering manager’s job to remove bottlenecks.
The sharper point: small software changes can cost about as much to try with an agent as to describe in a ticket. When that’s true, the backlog stops being a planning tool and becomes a delay mechanism. How a company decides what to build reflects its culture, and the backlog’s role is shifting as planning cadences shorten.
Quick Hits: Sales, Round Size, and IRL Events
Source: telemetrymachine.com ↗
Getting off your engineer ass and doing sales took the Telemetry Machine founder nearly two years to get the first customer. His advice: treat sales calls like Mom Test interviews, let the customer talk more than you, and nurture relationships rather than pushing a pitch.
Choosing a round size has shifted from a rational framework to a “vibey” guessing game, where the amount is a signal of ambition. A larger ask shrinks the pool of lead investors and requires more internal political capital. Pick a specific number based on business needs and aim for regret minimization.
On IRL events, Brex’s former events lead says ROI isn’t just direct revenue - it’s brand awareness and customer retention. Treat events as a long-term relationship-building motion, not a direct sales channel.
The first order is curiosity, the second is truth, says Marc Randolph. A first purchase proves only curiosity, not product-market fit. Free users are often “polite liars” who give positive feedback but never pay again. Treat repeat purchases as the only meaningful metric.
RevenueCat’s paywall piece settles the placement debate by distinguishing the curiosity gap model from the freemium model. Once you define what “value” means for your product, the placement question answers itself.
You May Also Like
GPT-6 Luna Cuts AI Prices in Half as Agent Economics Turn Ugly
OpenAI and Anthropic both shipped flagship models within an hour of each other, and OpenAI's GPT-6 Luna undercuts its predecessor by 50% per token - good news …
Frontier AI Prices Just Blinked: Anthropic Cuts Costs 20% as Databricks Publishes Real-World Usage Data
Anthropic cut frontier model prices 20% to $8/$40 per million tokens, the first time the top tier has gotten cheaper, and the cuts roughly match a 30% speedup. …
Nvidia's Board Hunts for Huang's Replacement as Agents Take Over AWS
Nvidia's board has hired an outside firm to identify CEO candidates as it weighs ousting Jensen Huang following reports of extramarital affairs, including one …




