Cosmo Li @CosmoThinks

$518B. About 80% of it can't be canceled. Pay whether you use it or not. Anthropic's 2025 revenue: $4.59B. Per the draft of Anthropic's IPO prospectus seen by Reuters, here's who gets it: Broadcom $161.2B Google $111.1B Amazon $110B xAI/SpaceX up to $84.5B Microsoft $31.4B Plus AMD The Google and Amazon contracts say Anthropic pays whether it uses the capacity or not, and covers any shortfall. The main exception is xAI. Most of that deal can be canceled with 90 days' notice. Nearly a quarter of that revenue came from two unnamed customers. I think it's a bet they have to make. Compute is everything in AI. The cost is that the risk stays with Anthropic, and in the end it lands on the shareholders who come in after the IPO. My guess: it goes public before October 2027, with a market cap above $2T. The public prospectus isn't out yet. Final numbers may differ.

$79B of profit in one quarter? Samsung reports in a few hours. Consensus operating profit: ₩106.1 trillion, about $79B Same quarter last year: ₩12.17 trillion If it lands, it's the first time a Korean company has topped ₩100 trillion in a single quarter. I think AI has turned memory into a money printer. The memory business is expected to run at a 76% margin. But price hikes are slowing. Up about 60% in Q2, only 10% to 15% expected in Q4. Profit estimates have been cut 7.7% since late August. Profit may already be close to the peak. My guess: operating profit roughly meets expectations, somewhere between ₩100 and ₩106 trillion. A year from now, memory prices are about flat.

Valuation doubled in two months. AI chip startup Etched raised at a $21B valuation in August. Now it's getting offers at $40B+. Etched makes chips built only for AI. Per TechCrunch, it has about $1B in orders, and its test chip only came out of TSMC this summer. Lambda runs a GPU cloud. Per the WSJ, it's raising up to $4B at a $14.5B pre-money valuation. It's the last round before an IPO, planned for 2027. The shovel sellers are getting repriced too. I think it's overheated. Etched's $40B offer sits on $1B of orders, and orders aren't revenue yet. What holds a valuation up is long-term contracts and orders. My guess: Lambda goes public before October 2027, at a higher valuation than this round. What we still don't know: Lambda's revenue and contract backlog. The reports don't say.

@TheBlockCo July reports had $75M at a $1.5B valuation. It closed at $90M, same valuation. Bigger round, not a higher price. But Hermes is MIT-licensed and Nous's own note shares no revenue. 24M clones is reach. Nobody pays for a clone. The $1.5B leans on Hermes for Businesses, not out yet.

@monokern The code is free. Running it isn't. A dot comes with its own cloud computer and GPT-6 Astra, all inside the plan. Self-host and you pay for an always-on box plus the model, per call or in hardware. Cheaper for some people, sure. Free, no.

@SophiaCai99 @WHOSTP47 @mkratsios47 @Katie_Roof $100M is the face value of the credits. Credits usually get counted at list price, so it costs National Compute less than the headline. The number that matters is GPU-hours, and how long researchers have to use them. Nothing reported so far says either.

@TechCrunch If you read this as a $300M raise, that's the valuation. Tab won't say how much it raised. The bigger unknown is how it makes money. It buys things for you with its own wallet. Subscription, or a cut of every purchase? The article doesn't say.

@theinformation Workday's own Q2 number: 5,500+ of its 11,500+ customers use at least one of its agents. That's a low bar. AI was 25%+ of new ACV last quarter, so the money is already booked. Renewal is the real test: do customers pay again for AI they barely use?

OpenAI's valuation is up more than 60% in six months. So how did it get cheaper? March: $852B valuation, ~$24B annualized revenue, ~35x Now: ~$1.4T valuation, ~$70B annualized revenue, ~20x The valuation rose 60%+. Revenue nearly tripled. Priced on revenue, it got cheaper. This round has no lead investor. OpenAI set the price itself. UAE funds could put in up to $10B, and BlackRock is in talks. OpenAI holds the pricing power. Investors are lining up. I think the price is fair. AI is doing more and more of the work of training the next AI. Its reach will snowball into more industries. A lot of what traditional companies earn today will end up as AI companies' growth. My guess: by October 2027, OpenAI's annualized revenue will top $150B. Data from OpenAI, Axios and Bloomberg. The round is still being negotiated.

@StockSavvyShay Those long-term deals cut both ways. Per Reuters, they're a growing share of memory supply and come with price ceilings, part of why Q4 contract hikes are expected at 10–15% after ~60% in Q2.

@StockSavvyShay Same playbook, opposite direction. AMD pays customers in its own stock (up to 320M shares for OpenAI and Meta) to lock in GPU orders. Nvidia invests in customers whose compute deals end up as Nvidia GPU orders. Figure's $3.5B Nscale contract is exactly that.

@illscience Money runs both ways. Per Reuters, Anthropic's draft IPO filing commits up to $84.5B to xAI/SpaceX compute, while Grok will route requests to Claude. But Anthropic can mostly walk away with 90 days' notice. The house wins only while this customer stays.

@Techmeme @jglasner Crunchbase's own read: the 35% drop is mostly OpenAI and Anthropic skipping mega-rounds this quarter, not a broad pullback. So it's less "funding fell" and more "two companies didn't raise."

1 cent a share. Up to 320 million shares. That's AMD's price for OpenAI and Meta. Its stock already sits above the final $600 price target. If it all vests, the two would own about 16% of an enlarged AMD. At $649 a share, that's roughly $200B. It comes with real strings: · The first tranche vests only after the first 1GW is delivered · They get all of it only after buying the full 6GW · Every tranche also needs the stock to hit a price target. The last one is $600 $AMD closed at $649 on Oct 6. What's left is mostly delivery. I think it's smart: future stock for big orders now. My biggest worry is customers using AMD to squeeze Nvidia on price. They sign for 6GW and end up buying much less. Investors look at the dilution. Founders should see that more and more suppliers are using equity to win customers. My guess: both first tranches vest within the next 12 months. Smart deal, or a discount in disguise?