SAnthropic just filed its S-1 with the SEC. The ticker is A-N-T.
ESixteen billion dollars in revenue, first half of the year.
SAnd a net loss big enough to make you blink.
EThat's the AI business right now. Revenue growing fast, losses even faster.
SBut here's what got me. Most of that revenue—around eighty percent—is usage-based.
EThat's good. Customers use it, they pay. Real product-market fit.
SExcept they've also promised Amazon, Google, and Microsoft they'll pay for compute whether they use it or not.
EThat's how you reserve capacity. You don't build frontier models on spot pricing.
SNo, wait—if usage drops, revenue drops, but the cloud bills don't. It's upside down.
ECome on, usage isn't dropping. They're one of two companies developers actually build on.
SThe S-1 doesn't list annual payment amounts. You can't see which year's bill could blow through their cash.
EAnthropic says cash plus revenue covers at least the next twelve months. That's confidence.
SOr it's what you say right before public investors take the risk! That's exactly my point.
EOkay, that's fair. The compute obligations are the part I'd read twice.
SAnd remember, this is a filing, not a listing. The stock doesn't trade yet.
ERight. But it tells you the AI race is becoming a public market story, not just a private valuation story.
SEl Mundi's daily AI briefing on X had four items. Two are confirmed, two are just claims.
EAnthropic's filing is confirmed. And OpenAI did release a cheaper model.
SGPT-6.1 Terra, for computer tasks. One dollar input, five dollars output per million tokens.
EThat's a real price cut for builders automating browser work.
SThe briefing also says Meta plans a stock sale up to ninety-five billion dollars. No official word.
EAnd OpenAI claiming big math and physics results. Again, unverified.
SMy take: the IPO makes the math public. Quarterly earnings meet fixed compute contracts. That's a collision.
EMy take: usage-based revenue is exactly why this scales. You capture the upside as usage grows.
SUntil it doesn't.
EYou are determined to find the downside.
SSomebody has to read the footnotes.
ESpeaking of people opening their wallets for AI, a ChatGPT Plus subscriber is practically demanding to pay more.
SThat's not how customers usually talk.
EGlenn Twiddle posted at Sam Altman on X. He's a ChatGPT Plus subscriber who says two advocates convinced him he'd pay another eighty dollars for Dots.
SEighty dollars on top of his current plan? For an always-on assistant?
ERight. Most people ask for a discount. This guy's requesting an upsell.
SAnd he credits two creators, Emma Rainville and Mitch Barham, for demoing it so well.
EThat's organic advocacy. OpenAI doesn't have to sell Dots if power users do it for free.
SBut one happy subscriber is a testimonial, not data. It doesn't mean Dots is worth that for everyone.
ETrue. But it tells you where the heat is: always-on assistants, not chat windows.
SI'll wait for the renewal bill before I call it a trend.
EFair. But if I'm OpenAI, I'm watching whether that enthusiasm spreads.
SThe dangerous part? When advocacy blurs into marketing. We've seen that movie.
EThis one reads like a fan, not a paid ad. But we can't see the direct messages.
SThat we know of.
ENow who's reading footnotes?
SFair point.
EIf Dots delivers, eighty dollars may look cheap. Always-on agents are the next subscription battle.
SAnd the next lock-in. You build your life around one assistant, switching costs get real.
EThat's the business model.
SDoesn't mean I have to like it.