Research · Posts

Evenrock Signal — Week of April 20, 2026

2026-05-02·podcast digest·draft

Top insight of the week

Agentic AI isn't disrupting SaaS at the edges — it's breaking the seat-based business model at the core. When a Workday customer who needed 50 seats now needs two because agents are the interface, every per-seat valuation model in the public software stack is wrong. Thoma Bravo just handed Medallia's keys to creditors and wrote off $5.1B of equity; Salesforce trades under 10x free cash flow; ServiceNow grew $5M+ ACV customers from 434 to 630 with 97% retention and the stock dropped 18%. The market is pricing a regime change, not a cycle. Worth being wrong about: maybe the AI tax on legacy SaaS revenues is overstated and 2027 numbers come in fine. But if Friedberg is right — "He needs two seats because the agents act as the way to go in and out of Workday… multiply that by a million customers" — the de-rating has barely started.

The hyperscaler capex super-cycle is real, and it's compressing free cash flow

The MAG-4 prints validated AI demand and simultaneously raised the bill. Capex is now structurally larger than free cash flow at multiple names, and the financing model is changing.

  • Combined 2026 capex guidance is approaching $580B across Amazon, Microsoft, Google, and Meta, with Amazon alone at $200B — the first company ever to hit that mark. (Jason Calacanis, All-In)
  • Amazon's free cash flow fell 97% YoY; Microsoft and Meta down 12% and 8%. "These companies will now get levered… they're going to look like this bulky industrial business in five years." (Chamath Palihapitiya, All-In)
  • Microsoft's RPO hit $625B, up 110% YoY, with ~45% tied to OpenAI — the single most eye-popping forward-revenue print in enterprise AI. (TBPN, Tech Earnings Quadkill)
  • Google Cloud grew 63% to $20B with backlog near $460B; AWS re-accelerated to 28%, fastest in 15 quarters at a $150B run rate. The MAG-4 trade at 17–25x forward earnings versus Cisco at 200x+ at the dot-com peak. (TBPN, AI Profits Surge)

Compute, not weights, is the moat

The constraint on AI revenue is power and silicon, not demand. That changes who captures the value.

  • Less than half of announced gigawatt capacity is actually built — the rest is stuck in red tape. This hurts Anthropic and OpenAI most; benefits Oracle, Meta, Microsoft, and Google. (Chamath Palihapitiya, All-In)
  • OpenAI raised $122B and expects to burn through it in three years; CFO has privately warned the company may not be able to pay its data center contracts if growth stalls. (Michael Batnick, Animal Spirits)
  • The SpaceX–Cursor deal: $60B acquisition option or $10B non-dilutive payment — effectively a breakup fee that covers Cursor's GPU rental bill. Cursor's $10B run rate is contracted, but gross margins are negative 23%. The deal monetizes XAI's idle Colossus capacity rather than its model. (TBPN; Chamath, All-In)
  • Nomer Brown: "Model weights become less important as inference becomes more important… inference capacity becomes a competitive advantage." (TBPN, China Blocks Manus Deal)

The SaaS PE bomb

Floating-rate debt stacked on COVID-era multiples is meeting AI-driven seat compression. The fuse is already lit.

  • Medallia: $6.4B Toma Bravo LBO in 2021, $5.1B equity wipeout in 2026. Debt service was set to triple from $100M to $300M/year; sales reps were hitting ~20% of quota. Apollo marked the debt at 74¢. (All-In; TBPN)
  • Blackstone's head of private credit blamed execution, not AI — "Medallia had been struggling not because of anything to do with AI but due to what we believe to be execution-driven issues." Take that at face value or don't. (TBPN, Thoma Bravo Loses Medallia)
  • Public SaaS comps are still fine on the surface: Snowflake +30%, DataDog +29%, Salesforce reaccelerating from 8.7% to 10.8%. The disconnect between fundamentals and multiples is the trade. (TBPN, Marc Benioff vs. Schulman)
  • Adam Parker's framework: software companies will "miss first on margins and then eventually miss on sales" as AI tooling spend hits gross margins that analysts have modeled flat through 2029. (The Compound and Friends)

The analog barbell

While capital chases compute, several operators are betting the next decade rewards the things AI can't replicate.

  • Josh Kushner launched Thrive Eternal, a permanent-capital holdco; first asset is a stake in the San Francisco Giants. "Concentrated in a small number of assets that we can own and steward many decades." (TBPN, Intel Rips)
  • Gary Vaynerchuk: "Extreme AI… extreme analog. It's a barbell. In 10 years it's going to feel like 1950." China is on track for $1T in live-commerce GMV this year. Vinyl outsold CDs for the first time in a century. (TBPN, Gary Vee)
  • GSP's buy-and-build playbook: acquire bolt-ons at 5–8x cash flow in industries where strategic platforms trade at 12–16x; never use leverage upfront; only buy businesses 17+ years old. Authentic Brands started as a $20M check four years ago and is now several billion. (Capital Allocators)
  • Matt Ankrim's 100-bagger discipline: 84% of the top 50 wealth-creating stocks were high-quality fundamentals; Buffett made 98–99% of his wealth after age 65; the average holding period has collapsed from 8 years in the 1960s to 5.5 months today. (The Compound and Friends)

Governance, geopolitics, and the things that bite later

Three stories worth keeping on a watchlist — none of them moved markets this week, all of them set precedent.

  • China ordered Meta to unwind its $2B Manus AI acquisition, threatening fines, business restrictions, and criminal charges for individuals. Delian Asparouhov's read: "Why would any founder start an AI company in China if they had a choice?" (TBPN, China Blocks Meta's $2B Manus Deal)
  • OpenAI–Musk trial: the jury verdict is advisory only — Judge Rogers can disregard it and sets damages alone. She's also restricted Musk's doomerism narrative as a "sideshow distraction." Musk admitted on the stand that distilling rivals' outputs is "an open secret" across the industry. (Mike Isaac, TBPN)
  • Tim Cook hands the keys to John Ternus on September 1. Cook shrank Apple's share count by 44% and added $684M of market cap per day for 15 years. Ternus inherits a China-concentrated supply chain and an AI strategy Ben Thompson likened to "the alcoholic who admits they have a drinking problem but plans to limit their intake to social drinking." (TBPN)

Episodes worth your full attention

  • All-In, "OpenAI Misses Targets, Codex vs Claude, Big Hyperscaler Beats" — Chamath's compute-scarcity framework and Friedberg's seat-based-pricing thesis are the two most useful mental models this week.
  • The Compound and Friends, "How to Find 100 Bagger Stocks" (Matt Ankrim) — A clean, evidence-backed argument for why quality-first long-duration holding still works, with a Fastenal case study that will make you rethink any sale you've made on a single earnings miss.
  • Capital Allocators, "GSP Buy-and-Build Playbook" (Sloane & Perelman) — The clearest articulation of why disciplined roll-ups in the boring core economy keep compounding while levered SaaS burns. The 17-year minimum age rule alone is worth the listen.