Newsletter · · Ashutosh Agarwal

Is SaaS Broken? - Week of June 19, 2026: OpenAI's 39% Gross Margin Makes the SaaSpocalypse Real

Enterprise software newsletter for the week of June 19, 2026. OpenAI's leaked 39% gross margin gives the AI-COGS bears a number, the SaaSpocalypse becomes a named bull-vs-bear brawl, and Figma's 139% net retention is the cleanest counter-punch to seat erosion yet.

Is SaaS Broken?

Week of June 19, 2026: OpenAI's 39% Gross Margin Makes the SaaSpocalypse Real

Last week we said two numbers would settle this debate: an AI gross margin and a hard NRR print. We got proxies for both, pointing opposite ways. OpenAI's leaked financials show a 39% gross margin (inference looks structurally low-margin); Figma printed 139% net retention (seats aren't bleeding). And the bear thesis finally has a name: the SaaSpocalypse.

TL;DR

  • OpenAI's Q1 leaked at a 39% gross margin, the first hard look at AI-layer COGS, roughly half of legacy software. The "AI is structurally lower-margin" thesis now has a number.
  • The SaaSpocalypse is now a named bull-vs-bear brawl, but Ramp's data says seat contracts are still 60-75% of software spend and metered usage under 5%. The pronouncements run ahead of the data.
  • Figma printed 139% net retention, the cleanest counter-punch to seat erosion yet, and a direct read-through for Adobe.

What's New

1. OpenAI's margin is half of Big Tech's. On The Rundown, 6/17, Zaid Admani [Analyst] broke down OpenAI's leaked Q1: revenue $5.7B (triple YoY), a $9.3B operating loss, $3.7B burned, and "gross margins were only 39%… the cost of delivering that revenue is also enormous." Microsoft runs ~70%, Meta ~80%. It's the closest proxy yet for the "52% AI vs 75-85% legacy" gap, and the floor under every SaaS vendor's AI features.

2. The SaaSpocalypse got steel-manned, and the data favors incumbents. On Big Technology, 6/17, Ramp lead economist Ara Kharazian [Analyst] said "the death of SaaS has been greatly exaggerated": seat contracts are still "60 to 75%" of software spend, metered usage "less than 5%," Adobe credits "only like half a percent of their revenue." His sharper point cuts the other way, "80% of their revenue from businesses is token-based, not subscriptions" at OpenAI and Anthropic, so the labs, not the SaaS names, live on the deflating unit.

3. Token prices are down ~90%, and the labs know it. On RiskReversal, 6/17, Current's Stuart Sopp and Trevor Marshall [Operator] said token prices came down "like 90 percent" and "if I'm Anthropic or OpenAI, I'm starting to get a little bit worried." Their tell: Anthropic shipped Fable inside subscription plans with near-unlimited tokens, then promised a shift "to usage-based pricing shortly after, which is kind of a hidden tax," how the labs claw back a token deflation that should help SaaS COGS.

4. Anthropic's run-rate moonshot shows the money moved to usage. On The AI Daily Brief, 6/16, NLW [Analyst] cited SemiAnalysis: Anthropic's run rate jumped "from $30 billion to $47 billion by late May," with $1M-a-year enterprises going "from 500 to more than 1,000 in under two months," driven by "an insane amount of usage of cloud code," not seats. His framing, "seats have… ceased to be the core unit," is the bull case and the bear case at once.

5. The subsidy is enormous, which is why the price war is coming. On The Artificial Intelligence Show, 6/16, Paul Roetzer and Mike Kaput [Analyst] relayed SemiAnalysis: a $200/mo Claude plan delivers up to ~$8,000/mo of tokens at API pricing, a $200 ChatGPT plan ~$14,000, at full usage the labs earn "a negative 900% return." Scott Galloway echoed it on Prof G, 6/17: Claude Pro/Max "costs $200 a month… but… $5,000 a month for the compute," against a $14B projected 2026 loss.

The Debate

Bear: per-seat SaaS is structurally broken. Usage replaced seats, and usage routes to whatever's cheapest. On The Data Exchange, 6/13, Ben Lorica and Evangelos Simoudis [Analyst] named three vectors: frontier labs building enterprise apps, AI-native startups ("2 people and hundreds of agents"), and Fortune 500s using "headless AI" to hit SaaS databases with their own agents. Snowflake CEO Sridhar Ramaswamy [Operator] went further on In Good Company, 6/17, calling the coding agents "our biggest competition… the industrialization of software at a scale we have not seen before."

Bull: incumbents own the plumbing and the data. Same episode, the rebuttal: "the dirty secret in AI is not the model. It's the data integration and the plumbing," ripping out a 20-year CRM or HR install is slow. Swing factor unchanged: does consumption revenue outrun seat bleed? This week, for the first time, the data leans incumbent.

Stocks in Play

  • ADBE : Bull: ~11% creative/marketing growth, ~8x forward earnings. Bear: AI is "a drop in the bucket" at ~2% of ARR, price hikes deferred, the 20-year CFO leaving as the CEO moves to chairman mid-pivot (The Canadian Investor, 6/18) [Analyst]. Catalyst: new-CFO margin framing.
  • CRM : Bull: still ~80% of CRM share, revenue rising (Big Technology, 6/17). Bear: pundit Paul Morris [Analyst] flagged ~$158B of value erased over 16 months despite rising revenue, "the market… is saying the business model is broken" (Paul Morris Podcast, 6/15). Catalyst: next Agentforce consumption-ARR print.
  • DDOG : Bull: leaned into the cost backlash at Dash with bring-your-own-cloud, federated log search, agentic auto-remediation (Network Break, 6/15) [Analyst]. Bear: BYOC admits AI log volumes make pure-SaaS consumption too pricey; it protects share by shrinking the bill. Catalyst: whether BYOC dents NRR.
  • TEAM : Bull: Rovo rides the agent wave on a sticky cloud base. Bear: still seat-priced into seat erosion; even Atlassian brand exec Matt Scribner [Operator] conceded "tokens are expensive, sometimes more expensive than actual human beings" (Best Story Wins, 6/18). Catalyst: Rovo consumption metrics.
  • HUBS : Bull: SMB land-and-expand with outcome-focused agents, not "agent washing," per PMM Maitri Sheth [Operator] (Building With Buyers, 6/17). Bear: most seat-exposed in the group; cited down 28% / ~$4B in 30 days [Analyst]. Catalyst: Breeze attach and NRR.
  • ASAN : Bull: AI Studio agents as multiplayer teammates. Bear: smallest, most seat-dependent, squarely in the agent-threatened path. Catalyst: AI Studio monetization. (No coverage this week.)
  • MNDY : Bull: fast-growing Work OS with usage-based add-ons. Bear: SMB seat base in the erosion path. Catalyst: AI attach + NRR. (No coverage this week.)

Read-throughs

  • Seat-heavy SMB SaaS (HUBS, ASAN, MNDY) is most exposed, but Ramp says the rip-out is slower than the headlines. Watch NRR: Figma's 139% disproved the seat-cut thesis this quarter (Chip Stock Investor, 6/16) [Analyst]; a slowing NRR is the first crack.
  • Model/inference vendors are the actual margin victims. Perplexity's Aravind Srinivas [Operator] put it bluntly on 20VC, 6/15: "the model is no longer the product… if you're literally just a reseller of model tokens, you have no business." Falling model costs favor the application layer that owns the workflow.
  • Multiple de-rating is now bifurcated by proof, not vibes: consumption + durable-NRR names (DDOG, Figma, CRM's AI line) get the benefit of the doubt; pure seat-counters re-rate down until they print the same.

What Changed vs Last Week

Last week the price war was "reportedly prepping"; this week it's "declared," and we have a real number: OpenAI's leaked 39% margin, the AI-COGS proxy missing two weeks running. The bear case got a name (SaaSpocalypse) and its first data-driven rebuttal: Ramp's seat-contracts-still-dominate figures and Figma's 139% NRR. Coverage rotated: Datadog re-emerged; HubSpot, Adobe, and Salesforce got direct airtime. Still dark: Asana and Monday.com.