How’d I do? What a cool experience! Just the start for us here at @AvoryCo

Ed Zitron now calling World Series games?

Solid Robinhood results, the knock could be the deposit decel from 28% annualized to a multi-year low of 17% annualized… Last year we saw a step up from from Q2-Q3… 2026 a 11% step down. $HOOD

Visa, strongest growth in 15 years. Yet I keep hearing consumer must be in bad shape, or AI is only thing driving economy.

Snowflake and Anthropic models are not to different. From lens that these are consumption based. They have to earn multi year credit consumption purchases. Point is you can go from little to no visibility in +1 out year revenue, which is where Anthropic sits today, to multi year commits. What’s the product for Anthropic tho??? $SNOW $ANT

Bloomberg now pointing to what we expressed here recently. Expecting serious software M&A. $IGV

Hiring remains steady 4-week avg: +16,750 per week 8-week trend: +19.6K.

Nvidia taking $3B stake in $META

Another acquisition in software as Guidewire acquires hyperexponential for $1.7B. Expect more of this in software. Salesforce buying Braze opened the gates. $IGV

This is what full stack looks like. Own the customer. Own the compute. Own the model. $META

So this is ultimately what model provider business model looks like. Cowork should help create deeper integrations with customers overtime. BUT clearly hyperscalers found their way into another position of strength. $ANT

Been reading this S-1 all morning and having hard time with this one. So chip in. So the leading AI lab that is at frontier. Since 2023 has spent $19B of cumulative compute/infrastructure expense + ~$630M of direct capex. Yes $19B… according to S-1 this includes training, inference and everything. 2026 is closer to $16-18B annually. But still. Where we are at this point in intelligence, which is pretty powerful, that’s all they spent. Now two things. 1. The industry is contemplating trillions of dollars of incremental infrastructure. Like square the capabilities today at$16-18B with trillions. 2. Then you have in S-1 their future commitments. $161B Broadcom equipment leases $114.5B+ Google $110B+ AWS $44.9B+ Microsoft up to $120.5B SpaceXAI $44.6B Nscale ~$41.3B Lambda Something here doesn’t add up. THE what has been achieved at the cost, vs the future spend, means the value extraction has to be like enormous. Like inconceivable enormous. Am I missing something? $ANT

One thing that jumped out to me in Anthropic’s S-1 is the marketplace revenue. Roughly 48% of revenue is coming through cloud marketplaces like AWS, Google Cloud, etc. Anthropic recognizes that revenue gross. Meaning if a customer spends $100 through one of those marketplaces, Anthropic can recognize the full $100 as revenue even though the cloud provider keeps a fee before sending Anthropic the cash. Looks like in 2025, the effective marketplace fee looks like it was around 16%. On S-1 Anthropic says channel partner fees increased by roughly $2.7B in H1 2026. When you compare that with the growth in marketplace revenue, the incremental economics suggest the channel take could be meaningfully higher today. Potentially somewhere in the 25%–35% range, although the disclosure isn’t clean enough to say every dollar of that expense is a marketplace fee. So instead of looking at Anthropic’s $16.3B of H1 reported revenue at face value, I think it’s useful to also look at an economic “net revenue” number. Depending on the marketplace fee assumption, I get something closer to roughly $13.5B–$15B. I’d probably use ~$14B as a working number. The revenue isn’t fake. The customer is spending the money. But if almost half of your revenue runs through a distribution layer that takes a meaningful percentage of every dollar, I want to normalize for that when thinking about valuation and unit economics. Also the math on it was 1. Calculate marketplace revenue H1 2026 reported revenue = $16.286B We knowmMarketplace share = 48% $16.286B × 48% = $7.82B of marketplace revenue Cool that then leaves: $16.286B − $7.82B = $8.47B of non-marketplace revenue. 2. Perfect now apply different marketplace take-rate assumptions Using the ~16% historical fee Reuters calculated for 2025: $7.82B × 16% = $1.25B of marketplace/channel fees Then: $16.286B − $1.25B = ~$15.0B economic net revenue At a 25% marketplace fee: $7.82B × 25% = $1.95B $16.286B − $1.95B = $14.33B At a 35% marketplace fee: $7.82B × 35% = $2.74B $16.286B − $2.74B = $13.55B Reuters calculated approximately $351M of distribution fees on $2.16B of 2025 marketplace sales, which is where the ~16% historical benchmark comes from. The reason I started entertaining a much higher number for 2026 is this calculation: H1 2025 marketplace revenue: $1.163B × 44% = ~$512M H1 2026 marketplace revenue: $16.286B × 48% = ~$7.82B Increase in marketplace revenue: $7.82B − $0.51B = ~$7.31B Anthropic then says channelpartner fees increased by $2.7B year over year. So: $2.7B ÷ $7.31B = ~37% That’s why I think the incremental economics look potentially closer to a ~37% channel cost. So maybe higher but that makes revenue multiple 20% more expensive at mid point. $ANT

to add, like some of the best analogies for moats go back to history. A moat really is just a layer of defense. It can be inherent to the business or industry, or purposely created through something in a company’s operations. The original concept was more territorial. Castles had walls, water moats, guards, cannons, etc. None of those made the castle impossible to conquer. They simply made it harder, more expensive, and more time consuming to attack. That is how I think about business moats. So to @jack point … every moat can eventually succumb to competitive pressure. The point of the moat is not permanence. It is creating a layer of defense that gives the business some duration before competition can break through.

He’s right in the absolute sense that there are no moats… if a moat is being defined as PERMANENT impenetrable edge against competition. Now, if the definition of a moat is a competitive edge that has some duration to it, then I think yes, there are absolutely moats.

😱

I’m so long Anderson Paak right now after this new album.

The water debate around data centers: Here’s some stats from Amazon 1. Golf courses and almond farming consume significantly more water than data centers. 2. Golf courses alone consume roughly 200x more water than all of their data centers combined. 3. These industries can consume more water in two days than their data centers use in an entire year. 🤯

Uhhh. 😬 $META

Wild. I remember demo of their voice product for sellers back in November. This is 10x $XYZ

New Investing with Data is out. People are less inclined to step in front of AI equities here, and more inclined to look at the debt of them... $ORCL $CRWV $META $TSLA $BX investingwithdata.substack.c…