Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Saturday, November 7, 2026

Back to the front page

From X

@convequity

Convequity on X

WSJ’s “$750B Anthropic compute bill”, which includes deals still being negotiated, is overblown as a panic. It isn't $750B a year. It's spread over about 10 years. By year 5, the fixed compute bill is about $104B a year. The math, stress-tested: assume compute eats a deliberately punishing 56% of revenue (far above where a scaled AI lab should settle), still leaving a 20% operating margin after other costs of ~24%. $104B / 56% ≈ $186B revenue in year 5 $186B / ~$40B revenue per GW ≈ 4.6 GW required $186B is also only ~2.8x today's $67B run rate. That's ~23% growth a year for five years, far slower than Anthropic is growing now. Demand isn't the problem. On paper, 4.6 GW is well inside the ~13.8 GW Anthropic has already lined up according to our ABBX. And it mostly pays as capacity comes online. The catch: what if the GW roadmap doesn't fully materialise? Only ~1.9 GW is live today. Getting even 4.6 GW running on Earth means beating power shortages, slow permits and growing local opposition to data centres. So the real risk isn't demand, it's delivery. That's why $SPCX matters. If Starship makes orbital compute work, it sidesteps the grid, the permits and the politics. The tide lifts all boats, and Anthropic is one of the biggest.
@convequity 1 click

Convequity on X

Our AI Bubble Barometer (ABBX) says the hyperscaler complex is not yet in bubble territory. Across the hyperscalers and neo-clouds we track, enterprise value relative to the capital invested in the business has fallen from 7.67x at the 15 May 2026 series start to 6.12x at the 15 Oct 2026 close. Two-year forward returns on that capital were 3.07x the cost of capital in Q2'26, down from Q1'26. Still comfortably value-creating, but the direction is lower, which fits this stage of the cycle: data centres and chips go on the books immediately while the revenue they earn arrives later. Q3 2026 reporting starts with Meta ($META) on Wed 21 Oct after the close, with Alphabet, Microsoft and Amazon the following week. We will watch whether forward returns keep falling as the capital base grows ahead of revenue. Falling returns alongside rising valuation multiples would be our classic bubble signal.
@convequity 2 clicks

Convequity on X

$META's Muse download numbers measure Meta's distribution, which was never in doubt. What decides whether Muse is a business is what the agent completes for a user, and nothing reported this week measures that. Sensor Tower puts Muse among the fastest US app launches on record, with early retention above the average for rival AI apps, and says a large and rising share of Meta's own ad spend now goes to it. That ad spend likely explains much of the install curve, and Meta can run the same ads for any product it owns. 1. Tasks completed. Nothing Meta or Sensor Tower has reported says how many sessions end in a booking, a purchase or a message sent for the user. Those are what a user would pay for, and in our view they are also where a merchant fee could come from. 2. Where the revenue shows up first. In our view Muse's first financial effect is on the ad business Meta already runs. When shopping happens inside Muse, Meta sees the whole path from what a user wants to what they buy, which sharpens its ad targeting. If users start shopping in Muse, Meta can also sell sponsored placement in its answers. Further out, ads could become part of the conversation itself, something we suggested earlier for Meta AI and which applies even more to an agent like Muse. A user preparing for their first marathon with Muse could be offered trainers, energy gels and hydration products in a conversational manner as the plan takes shape, rather than seeing an ad placed beside it. We expect Muse to show up in ad pricing before it appears as a subscription or a take rate. 3. The sharing suspension. Meta paused conversation sharing after security researchers found shared chats could expose personal and business information. That suggests people are already giving Muse real work with real data and sharing it with colleagues. It also means workplace use will meet the data-handling limits that slowed enterprise use of consumer chatbots. A real catalyst for $META will be Meta disclosing how many tasks and transactions Muse completes, especially if the numbers are impressive.

Investor Letters

All letters

Axon 3Q26 Business Update

Axon cut full-year guidance after delayed federal Dedrone contracts slowed 3Q revenue growth to 28%, while software and services grew 43%.

AXON $31B
Speedwell Memos 3 clicks

AMD Q3 FY26, Annotated

AMD's Q3 revenue rose 45% to $13.4 billion as data center reached $8.5 billion, 64% of sales, driven by EPYC and Instinct ramps.

AMD $660B
Chipstrat 2 clicks

ARM Q2 FY3/27 Earnings Review

Arm reported Q2 revenue of $1.443 billion, up 27%, as licensing rose 50% while royalties grew 8%; Q3 guidance calls for $1.54 billion.

ARM $181B
Cestrian Capital Research

AppLovin 3Q26: Business Update

AppLovin's 3Q26 revenue rose 55% year over year and margins reached 78.2%, driven by AXON model gains, while publisher bidding pressure…

APP $118B
Speedwell Memos 2 clicks

Publications we follow

Every publication on this site, hand-picked.

129 publications

Newsletters 109

YouTube 4

Blogs 15

Seeking Alpha 74 of 277

Nominate a publication

X accounts we follow

Every account on this site, hand-picked.

91 accounts
Nominate an account

Investor letters we follow

359 firms

Sectors

Ordered by how often our editor reaches for them.