Anthropic Files to Go Public | Cognition Raises $1BN at $26BN Valuation | The 996 Work Ethic

Key points

• **The private-stays-cool era is over — the IPO queue is now a stampede.** O'Driscoll: *'we are done with the ooh, I don't want to do the public market. Staying private is cool. We are done with that.'* He tallies **Google's $80BN equity raise, SpaceX 'formalised its price at $1.75 trillion for early June,' and Anthropic + OpenAI both pointing at October — '300 to $400 billion of equity issuance' across four names, 'all of which is really AI related.'** This is the supply side of the same wave [Thomas Laffont frames as the '$4T AI IPO wave' and Sarah Friar's OpenAI S1 confirm from the issuer's chair this week](/issues/2026-06-07) — three independent vantage points on one capital event.
• **The bear note under the euphoria: capex-light → capex-heavy.** O'Driscoll: *'all these businesses have gone from capex-light cash-flow machines to capex-heavy cash-consumptive machines.'* And the historical prior: *'across history, things that have high cash flow spinning out are really good investments, and things that eat money tend to be bad investments.'* **The cleanest one-line bear case on the entire mega-cap AI cohort** — Google issuing $80BN at an all-time premium is the smart-money tell that the capital intensity is structural, not cyclical, and the same shift [the OpenAI-CFO episode dramatises at $100B+ of compute spend](/issues/2026-06-07).
• **The single number that prices the whole cohort: tokens as a share of the engineering wage bill.** O'Driscoll, naming it explicitly: *'this is the number that will determine: is $1 trillion a fully priced company that could slow down a little bit for a year… or is it, oh my God, no one's even going to pause for breath… it's going to eat one third of engineering salaries and that's going to get you to 4 trillion two years from now.'* **This is the load-bearing signal of the episode** — the bull/bear gap on Anthropic and OpenAI collapses to a single ratio (10% vs 33%) that no one yet knows, and it is the quantitative version of [last week's 'measurability gate'](/issues/2026-05-31).
• **Lemkin's call: tokens over humans by year-end.** *'I really do think by the end of the year we're going to choose tokens over humans for engineering and product.'* His mechanism is budget substitution — give a department head a fixed dollar budget and *'I'll take tokens, I'll take tokens over a B… we've already made that choice at SaaStr, we got rid of all our B's.'* **The 'AI-backfill' thesis stated as an operating decision, not a forecast** — and the granular read-through to the layoff data [the prior issue tracked as the AI-job-loss vibe shift](/issues/2026-05-31).
• **The token bill came due all at once — accruals 10x over budget.** O'Driscoll's set-piece: *'in early 2026 they changed the pricing model so you pay as you go… everyone cranked in Q1… by mid-May they suddenly realised we're 10x wrong on our accrual, and the penny dropped simultaneously across the entire corporate US.'* Uber's answer: cap *'everyone… 1,500 bucks a month.'* **The specific, dated anatomy of the ROI panic** [Issue 07 reported from the enterprise side](/issues/2026-05-31) — confirming the bill is real, even as the panel splits on whether it slows spend or just disciplines it.
• **The deflation argument that keeps the spend alive: the frontier ratchet.** O'Driscoll: *'the frontier models aren't getting cheaper… they're getting slightly more expensive. But the model that is frontier today will in a year be 5 to 10x cheaper because it won't be the frontier model anymore.'* Lemkin's countervailing fact: *'we are all in on the best. We are all in on Opus… so we're not benefiting as much from the deflationary benefits of AI and we're paying into the inflationary side.'* **The crux of the model-commoditisation debate** — deflation is real but enterprises refuse to ride it down, which is exactly why the model-provider TAM holds.
• **SaaS apocalypse over-corrected — but the survivors are AI-attached, not re-rated.** Lemkin: *'I never understood the total panic of the SaaS apocalypse… but the issues haven't gone away, it's just we over-panicked on the timing.'* The winners attached to agents — **Twilio and Okta both 'up 57 this year,' Datadog 'up 100 this year'** — while *'classic human per-seat software really is dying… they're cutting it because they've got to come up with money for all these goddamn tokens.'* With **AI software spend 'up 60% this year' (Gartner)**, *'you can't grow 60% AI software without cutting some material amount of the rest.'* The token bill is being paid out of the seat-licence budget — a direct CRE-relevant read on which software vendors survive the substitution.
• **Cognition's $26BN re-rates the whole coding cohort — and reframes Cursor.** Harry: *'Cognition raised a billion dollars at a $26 billion valuation. Devin… hit $492 million in ARR.'* Lemkin's frame: autonomous agents beat copilots — *'it's much more compelling to just tell the agent what to build… have these autonomous engineers that do it'* — but *'every one of the companies worth a trillion bucks is going to want to eat your lunch.'* **Application-layer defensibility is the open question** [Mercor's CEO presses in this week's other 20VC episode — 'application-layer companies have no defensibility'](/issues/2026-06-07): a $26BN harness sitting on rentable frontier models is exactly the thesis under stress.
• **Kirkland & Ellis's own-model build is dismissed as theatre — the build-vs-buy line for legal.** Harry frames **'Kirkland spending $500 million on building their own… $100 million over five years.'** Lemkin: *'it's a law firm that does 11 billion in revenue growing 20%… committing 100 million a year… a reallocation of less than 1% of revenue. They won't even notice it.'* O'Driscoll's deeper cut: firms buy horizontal tools with no differentiation, but pause when the AI might *'encapsulate your secret sauce'* — *'wait till you see what Claude does with your IP.'* **Directly load-bearing for a CRE/transactional lawyer**: the durable moat is *'the ruthlessness… and the willingness to flog their associates,'* and *'$10,000 an hour is nothing on a 10, 20, $100 billion transaction.'*
• **The distributions wave will reshape the VC industry itself.** Harry: *'Menlo will make $10 billion in carry, Spark will too, plus Founders Fund will make more than that from SpaceX.'* Lemkin's read on why some partners walk: *'when you make eight figures, just leave… if you don't love the game, retire.'* O'Driscoll's structural point: large distributions make the *next* fund's marginal carry look small, so the rational fix is *'be the LP as well'* — Peter Thiel's move. **The downstream consequence of the IPO wave nobody prices**: the Anthropic/SpaceX exits don't just return capital, they restructure who manages it.
• **The contradiction at the heart of the build-out.** O'Driscoll: *'we're all here in the Valley with a plan to automate white-collar work such that there's going to be mass unemployment… and yet you talk to every single person and they're like I've never worked this hard, I'm working 24/7.'* His GDP frame is the bear anchor: *'the average over the last 200 years has been 2%… even when one part of the org speeds up, it doesn't matter if you can make a gazillion pieces of software if you can't package it, price it, sell it'* — the weakest link sets the convoy's speed. **The structural counterweight** to [Satya Nadella's 'full-stack builder' / hyper-leveraged-generalist thesis on No Priors this week](/issues/2026-06-07): productivity bottlenecks may cap the realised lift, and therefore the realised token budget.

Notable quotes

Themes

• The AI IPO supply wave
• Token-spend ROI reckoning
• Tokens-over-humans / AI backfills
• Model commoditisation vs Opus lock-in
• Build-vs-buy and application-layer defensibility