Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Thursday, October 29, 2026

Back to today

Another Haircut with Avy the Barber

SoFi’s stock case hinges on interest rates, tangible book value and private-credit loan sales after its $1.5 billion equity offering.

SOFI $22B
WHAT ARE WE DOING? “Contrarians At The Gate” 3 clicks

From X

@convequity 1 click

Convequity on X

We wouldn’t be surprised to see GPU rental revenue reach $100bn per gigawatt a year. Nebius ($NBIS) has already reported a short-term deal in the $40m–$50m per MW annualised range. That is $40bn–$50bn per GW. Likewise for Space $SPCX and others while $CRWV sold its long term capacity too early at low prices. Use 660,000 Blackwell GPUs as a simplified 1GW case. With every GPU-hour rented: • $50bn a year = $8.65 per GPU-hour. • $100bn a year = $17.30 per GPU-hour. DeepSeek reports $38 of API revenue per GPU-hour at off-peak prices for a V4.1 Flash cluster. Peak tariffs are double. Serving the same workload at those prices would imply $76 per GPU-hour. For an inference provider that can reproduce those economics on rented hardware, a move from $5–$6 rent to $9 can still be affordable. Even roughly $18 need not make the service uneconomic. Suppose another provider achieves 70% of that $38–$76 revenue benchmark across its rented GPU-hours. That gives $26.60–$53.20 of revenue per hour. After $18 of GPU rent, $8.60–$35.20 remains for other costs and profit. The condition is matching DeepSeek’s serving efficiency on the rented hardware and sustaining paid demand at those token prices. Releasing the inference code does not guarantee either. For operators that can do this, we think there is room to keep bidding up scarce GPU capacity, even at twice today’s roughly $9 hourly rate.

Investor Letters

All letters

CoStar: 3Q26 Business Recap

CoStar’s 3Q26 revenue growth slowed to 14% as bookings fell to $37 million, while cost cuts lifted EBIT margin to 8.8%.

CSGP $13B
Speedwell Memos 3 clicks

Deep Dive: Amazon (Updated)

Amazon’s AWS grew 28%, with a $364 billion backlog, supporting the case that up to $200 billion of annual AI capex can earn solid returns.

AMZN $2.9T
Rebound Capital

MercadoLibre: Valuation Model

MercadoLibre’s sum-of-the-parts model values commerce and payments on FCFF, credit on equity returns, producing a $1,980 base-case target…

MELI $86B
Hated Moats

Meta On the Offense!

Meta’s Muse Spark 1.1 targets agentic enterprise work, backed by employee workflow data and 3,000 engineers building reinforcement-learning…

META $1.6T
MBI Deep Dives 1 click

Publications we follow

Every publication on this site, hand-picked.

129 publications

Newsletters 109

YouTube 4

Blogs 15

Seeking Alpha 70 of 272

Nominate a publication

X accounts we follow

Every account on this site, hand-picked.

91 accounts
Nominate an account

Investor letters we follow

271 firms

Sectors

Ordered by how often our editor reaches for them.