Digest: Anthropic opens its books, Meta sells $35B of stock, and the Fed week begins

Quick pulse for Tuesday, October 27. Anthropic filed its S-1 last night, and the numbers are wilder than the rumors. Meta is selling stock to pay for AI. The Fed meets today, and Microsoft, Alphabet and Amazon report right after it. Bitcoin is still stuck around $80K.

Anthropic’s S-1: 30x revenue in five quarters, and a loss that’s mostly on paper#

Anthropic made its IPO filing public on Monday. It plans to list on the Nasdaq as ANT. The S-1 leaves the price and share count blank. Reuters and IFR report it wants to raise up to $100B at around $2T, with the roadshow starting Nov 9 and pricing the week after.

The revenue line is the story. Quarterly revenue was $376M in Q1 2025. It hit $11.6B in Q2 2026. That’s about 30x in five quarters. IFR reports Q3 came in around $16.8B. That number isn’t in the S-1 yet, and it means growth slowed from +144% quarter over quarter to about +45%.

Bar chart of Anthropic quarterly revenue from $376M in Q1 2025 to $11.6B in Q2 2026, with a press-reported ~$16.8B for Q3 2026

Source: Anthropic Form S-1, quarterly statements of operations; Q3 2026 per IFR.

Then the headline loss. Anthropic lost $80.7B in the first half of 2026. Don’t read that as burn. $72.4B of it is a fair-value markup on convertible notes, mainly Amazon’s, which convert at the IPO. The actual operating loss was $693M. In Q2 alone, operating margin turned positive at 8%.

The customer numbers are just as extreme. The S-1 puts net revenue retention at about 900% as of Sept 30. It counts only qualified business customers, not Pro or Max subscribers. Roughly, a customer who paid $1 a year ago now pays about $9. Evan Armstrong at Every compares that with Snowflake’s record 189% and a public software median around 110%. More than 11,900 customers pay $100K+ a year, up from about 950 a year earlier. About 250 pay $10M+. Claude Code is the reason. Enterprise users who don’t use it spend under $75 a month on average.

The risks are in the same filing. The top five customers were about 25% of revenue over nine months. Their share fell from roughly 36% in Q1 to 21% in Q3. Compute is the big bill. Anthropic committed up to $120.5B for SpaceXAI’s Nvidia capacity into 2029, about $3.1B a month once it’s all delivered. It also committed at least $114.5B to Google through 2033, whether it uses the capacity or not. It warns its growth “may not be sustainable.” Armstrong’s read is that a $2T price tag needs that compute to stay sold out. Some investors already value it nearer $1.5T, per Reuters.

Meta launches a $35B stock sale, with $60B more planned#

Meta started a roughly $35B equity raise on Monday to fund AI infrastructure. It has two parts. $22.5B of Class A stock is sold through forward agreements with Morgan Stanley and Goldman Sachs. Meta gets the cash when they settle, by Nov 19, 2027. The other $12.5B is mandatory convertible preferred that turns into common stock around Aug 15, 2029.

Two details stand out. Nvidia, AMD and Broadcom indicated interest in buying up to $5B of the shares. That’s not binding. The companies selling Meta its chips would be helping pay for them. Meta also plans a $60B at-the-market program once its new Class C shares start trading. That’s about $95B of possible stock in total. Last quarter capex was $34.9B against $32.8B of operating cash flow, per Milk Road Stocks. The money has to come from somewhere, and now it’s coming from shareholders.

Fed day, then Microsoft, Alphabet and Amazon#

The Fed’s two-day meeting starts today. The decision lands Wednesday at 2pm ET (20:00 Helsinki). The Fed hiked to 3.75-4.00% in September. Markets put a hold this week at about 90-95% across futures, Kalshi and Polymarket. The fight is over December, where futures price a hike at roughly 75-80%. Thursday brings PCE inflation.

Earnings stack up around it. Robinhood and Seagate report tonight. Microsoft and Alphabet report Wednesday after the close, and Amazon on Thursday. Here’s what cloud growth looked like last quarter:

Bar chart of Q2 2026 year-over-year cloud growth: Google Cloud +82%, Microsoft Azure +43% in constant currency, Amazon AWS +37%

Source: Alphabet, Microsoft and Amazon Q2 2026 results.

Google Cloud grew 82%, Azure 43% and AWS 37%. Anthropic’s filing shows at least $114.5B committed to Google Cloud through 2033. That’s the kind of backlog to listen for on Wednesday.

Milk Road’s macro score goes deeper into cash#

The Milk Road Macro Index fell from -0.93 to -1.51. That keeps it in “cash” mode, with zero risk exposure. Market momentum did all of the damage, dropping from -0.68 to -1.26. The VIX jumped from 14.84 to 19.99 as stocks caught up with the bond market’s fear. Sept core inflation was +0.3% month over month, the second hot reading in a row.

Bitcoin is at about $79.6K this morning (CoinGecko). More than $815M of crypto longs were liquidated over the weekend. Milk Road’s levels: $80K support, then a liquidity pocket near $77K. The author stays bullish unless $75.5K breaks. Traders at cookbook mostly agree. They want 81K reclaimed and are watching the mid-70s.

An open model is 3 points from the top#

Moonshot released Kimi K3.1 Pro on Oct 23. Artificial Analysis scores it 55 on its Intelligence Index, behind only Claude Opus 5.5 (58), Claude Sonnet 5.5 (56) and Meta’s Muse Wonder (56). That’s 11 points better than Kimi K3. It scored 67% on Terminal-Bench 4.0 in AA’s testing, the top score of any model, up from 13% for K3.

Weights are due by Nov 4. When they land, the gap between the best open model and the best closed one shrinks from 12 points to 3, per AA. The API costs $2 per million input tokens and $10 per million output, or about $1.98 per task on AA’s index. It’s a 2.8T-parameter model with 104B active. Moonshot’s own benchmark claims haven’t been independently checked.

InvestAnswers: the Tesla-SpaceX math, Optimus RAM and a pause at Strategy#

James at InvestAnswers had a busy Monday on Patreon. Five posts stood out.

Barclays prices in the merger. Barclays analyst Dan Levy upgraded Tesla to Overweight and raised his target to $450 from $370. That’s about 21% above Friday’s close. Levy treats an all-stock SpaceX deal as close to settled. At $450, SpaceX would need to offer about 3.1 of its shares per Tesla share, and Musk would end up with about 53% of the votes. James gets the same ratio: $450 over SpaceX’s $147.49 close is about 3.05. The catch is the vote. It needs more than 50% of all outstanding voting power at each company, and abstentions count as no. Levy calls that a “high hurdle.” Tesla rose about 6% on Monday.

Robotaxi’s compounding bet. Per Barclays, Tesla’s robotaxi weekly miles went from about 70,000 in mid-July to about 560,000 by Oct 18, with 3.1M unsupervised miles across six cities. James says miles are growing 17-20% a week. If that held for eight more weeks, he thinks a $200M+ annual run rate by Christmas is possible. That’s his extrapolation, not company guidance.

Optimus will need a lot of RAM. Musk says robot memory demand will “exceed everything else on Earth.” Tesla’s AI5 chip carries 72GB of LPDDR5. James’s math: 10M robots a year at 72GB is about 720PB of RAM. That’s $2.2B a year at $3/GB and $10.8B at $15/GB. The 10M robots is a scenario, not a plan. James likes Micron and SK Hynix for it.

Strategy stopped buying bitcoin again. Strategy bought zero BTC last week. It spent $204.9M buying back its STRC preferred stock instead and still holds 848,517 BTC. James calls it “throwing good money after bad.” Meanwhile Charlie Bilello’s ranking puts bitcoin (-7.3%) and gold (-7.5%) as 2026’s two worst major assets year to date. He says that has never happened in a calendar year.

ASML still can’t make enough machines. ASML raised its 2026 sales outlook to €44-45B and says it won’t fully meet AI-driven demand in 2027 or 2028. James calls lithography the chokepoint for the whole buildout.

James covered the Strategy pause and Strive’s buying in his livestream.

Video: InvestAnswers, “$68M Bet on BTC, ETF Flows, Strive Closes In, MSTR Stops + Bagging 5% of ETH”

-Q

Sources#