Mag7 Earnings: Google & Amazon Up, Meta & Microsoft Down. Anthropic's Raise & What It Means for Potential IPO?

Key points

• **'Most aggressive quarter in American capitalism'** (Evan Armstrong, quoted by Rory). Mag 7: ~$540B combined revenue + ~$700B AI capex commitments. Top 5-6 companies growing 20-40% revenue and 50-60% capex — capex now consumes most of their free cash flow. Rory frames it as 'top of the distribution pulling away' — uniquely American and sobering for inequality, but also a reminder that 'these are 5 of the 7 largest market cap companies saying we're not going to get pushed around.' Sub-text: they're working for two privately-held companies (Anthropic, OpenAI) and providing distribution.
• **Alphabet runaway winner.** Cloud backlog nearly doubled to $462B = nearly all of Google's 2025 revenue. GCP +63% YoY (fastest of the hyperscalers). Search hasn't died — Lemkin's own Saster SEO is up 60% YoY (the highest ever). 'Google has no trade-offs' — every dollar of capex feeds clear, attributable revenue. Token production +60% Q4→Q1 (10B/min → 16B/min) but **Anthropic 10x'd tokens in the same period** — Gemini is materially behind in coding share despite the eval near-parity claim.
• **Microsoft's disclosure of the week: ex-AI, revenue is flat YoY.** AI ARR $37B. Capex $190B. Without the AI initiative, Microsoft trades like a SaaS company at 3x revenues. Lemkin: 'welcome to our world.' All the Mag-5 growth is from AI. If the AI bet is wrong, all valuations re-rate.
• **Meta got crushed for the same reason.** Capex raised from $125B → $145B. Wall Street can't model attribution — they're not running Mark's spreadsheets, but he isn't either. 'He didn't make $200B by running a spreadsheet at Harvard.' But the math: $150B capex for a 10-15% lift on a $200B business = probably overspend. The $150B is a 'bet on relevance' if humans switch from talking to other humans to talking to chatbots. The market puts a higher discount rate on that.
• **Token intensity reality check.** Lemkin's Saster runs an **autonomous AI VP of Marketing + AI VP of Customer Success for $254/month TOTAL in tokens** — both full-time, both replacing multiple humans. 'My partner Amelia thought that was per day.' He'd assumed it was a $3-6k/month bill. Rory's portfolio data: AI software cos at sub-10% LLM cost-as-percent-of-cogs; engineering token spend at 2-15% of salary (vs Rory's hypothesis of 20%). **Implication: Anthropic's $44B ARR requires concentrated heavy spenders on coding, not broad-based knowledge work.** Coding remains the 'tip of the spear / canary' for all other token use cases — the marginal returns elsewhere are much lower than expected.
• **Palantir blew out everything else** — RPO +134% to $4.45B, Rule of 40 at 145%. The mechanism: corporate CEOs whose top 2 mandate is AI transformation need a partner who can take **$10M chunks** at minimum to make the initiative meaningful. The 25-year-old AI-startup founder selling $200k/year point solutions can't call high enough. Alex Karp on the call: 'every stakeholder shows up to the meeting now — entire buying cycle compressed to less than a year' (vs Adobe historical of 5-year deployments). Lemkin: 'Palantir is general catalyst for AI transformation' — the consultant + integrator at the heart of the wave. **Implication: there's now a clear corporate-procurement winner-take-most dynamic in the AI services layer.**
• **SaaS apocalypse partial walk-back.** Atlassian +29%, Twilio +20%, NS8 +23%. Lemkin's two-pronged test: (a) monetise AI to existing customers + (b) attract net-new customers. **Atlassian** passes (a) — Rovo AI monetised heavily — but net-new customers slowing. **Twilio** passes both — net-new customer growth ~40% YoY because every AI-native startup (Replit, Lovable, 11 Labs, Sierra) defaults to Twilio. **NS8** the laggard at 9%. **HubSpot's coming release announces 'agents on par with humans' and full platform openness to agents** — the next big test of the headless-software thesis. If HubSpot reaccelerates, the bucket-2 'system of record' framework gets revalidated.
• **Anthropic raises $50B at $900B valuation in 48 hours.** Lemkin admits he was wrong two weeks ago when he said they should skip the raise and go straight to IPO — too valuable an option to forgo. Capex math: every $1 of Anthropic revenue requires **$3-4 of capex from someone** (Anthropic or partner). At 10x growth and 1-year forecasting, you're committing $30B+ in capex for every $1B of current ARR. 'Riskiest game of financial guesswork ever seen.' **The raise pushes the IPO timeline out — both Anthropic and OpenAI could now realistically slip to 2027** because both have ~$200B in raised cash combined and don't *have* to go public.
• **Sierra at $15.8B / ~105x AR.** $150M ARR, $950M raise. Brett Taylor's first business since chairing OpenAI. Lemkin's gut concern: customer-support software market is $20-30B, customer-support labor market is $400B — the price assumes massive labor-replacement TAM expansion, not just incumbent share take. **The deeper meaning: OpenAI's own chairman is betting massive money on building application-layer software on top of LLMs — direct counter to 'LLMs will eat everything' framing.**
• **Memory chips as stealth inflation.** Apple raised Mac Mini $599 → $799 because of memory cost. 'A significant slug of the capex raise this quarter is for the same physical amount of capex at higher prices because the memory portion has exploded in cost.' Direct cross-reference to **Brad Gerstner's 25% memory portfolio bet** from this week's All-In, and to **Reiner Pope's 50%-of-capex-on-memory thesis** from Issue 03. Three independent signals converging on memory as the highest-conviction sectoral trade.
• **Coinbase 14% layoffs — Armstrong's 'no managers of managers' edict.** 'If you can't ship and manage, if you can't deliver a campaign AND be the head of marketing, I don't want you at Coinbase.' Lemkin: 'Lead from the front with AI. Anyone on LinkedIn that talks about their team should be fired.' Direct echo of Foroughi at AppLovin (Issue 03) — eliminating layered management is now an operating pattern across martech (AppLovin), prediction-market exchange (Kalshi), inference cloud (Baseten), and now crypto exchange (Coinbase). **The 'CMO who can run their own campaigns via agent' is the new bar for the role.** This is the lean-ops thesis from Issue 03 with a 5th independent proof.
• **Musk v Altman trial — week 1 takeaways.** Musk admitted xAI distilled OpenAI models 'partly'. Greg Brockman stake $30B. Statute-of-limitations risk: if Musk's contemporaneous tweets showed he knew about OpenAI structural changes 5+ years ago, judge could throw out on procedural grounds. Donor-Advised-Fund technicality: Musk's actual contributions came via a DAF (separate legal entity); standing may be at the DAF, not Musk personally. **Bench trial with advisory jury → it all comes down to Judge Gonzalez Rogers's read on the technical issues, not the optics.**

Notable quotes

Themes

• 'Most aggressive quarter in American capitalism' — Mag 7 capex eating free cash flow
• Microsoft's ex-AI flatness is the disclosure of the week
• Token intensity reality check — coding is the canary, other roles use far fewer tokens
• Palantir as the dominant AI-transformation procurement winner ($10M-chunk minimum)
• Memory chips as the highest-conviction sectoral trade (Apple Mac Mini stealth inflation)