Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Sunday, October 25, 2026

Back to today

From X

@StableBread 1 click

Fajasy on X

Here’s what Boost Run $BRUN has already committed to pay, and what it still has to finance in H2 2026. $BRUN’s GPUs need space and power, which it rents in other companies’ data centers (colocation). The four leases signed since April already cover the space and power for the year-end deployments, and CEO Andrew Karos said its partnerships now reach 253 MW of power. $BRUN owes the rent on those sites whether or not the GPUs get financed. In 2027, the commitments $BRUN has already signed call for ~$221M of payments, including $103.5M on its GPU leases, $28.9M of rent on the sites in use as of June 30, ~$69M on the four new leases, and $20M for software. $BRUN’s $191.8M of pro forma cash and $93.6M of contracted prepayments, $285.4M together, cover the ~$221M of 2027 payments before counting any monthly billings from customers. But the same cash also pays $BRUN’s share of the H2 2026 build, so its monthly billings will have to cover part of the 2027 payments. To reach the $1.0-1.2B of GPU spending management expects for 2026, $BRUN has to deploy $663-863M of GPU servers in H2 2026, 2.3-3.0x its cash and contracted prepayments. If lessors fund 79.5% of the H2 2026 build, as they did in H1 2026, $BRUN needs $527-686M of new leases, 2.2-2.9x the $238.1M the company owed on finance leases as of June 30. The other $136-176M would be paid in cash. The $527-686M works out to $264-343M/quarter on average across Q3 and Q4, 1.3-1.6x the $208.2M of leases $BRUN took on in Q2. $BRUN’s lease debt would rise by less than the $527-686M, since the company keeps paying down its leases while adding new ones. During H1 2026, its finance lease debt rose $207.7M on $337.3M of new GPU equipment, or 61.6%, which for H2 2026 would imply ~$410-530M more debt. New leases will likely cost more than ~7% after the rise in rates. For context, at the 9.0% $IREN agreed to pay on its August GPU financing, the extra interest on $527-686M would be ~$6.5-8.5M/year, 1.6-2.1% of the $400M ARR target. And $BRUN isn’t slowing down. By June 30, the company had already placed orders well above the $578M first-year minimum in its April 2026 agreement with Dell $DELL, which also requires at least $215.5M/year of Dell equipment from April 2027 to April 2031 or a payment for the shortfall.
@convequity 1 click

Convequity on X

Banks are right, for now, to block third-party AI agents from taking action inside consumer bank accounts. An agent paying with a capped card, as $META's Muse does, is a different and narrower permission. Axios reported this week that major U.S. banks are finding unapproved third-party agents inside customer accounts, most let in with the customer's own login. Some tried actions that looked independent of what the customer asked. $MS is blocking such agents, per Axios sources. $JPM, $BAC and $C are restricting them. Banks are slow to change how a deposit account works for a reason: the duty runs to every customer and to the bank's own systems, not only to the early adopter who opted in. An agent logged in with real customer credentials puts fraud, liability and security risk on the bank, and that risk reaches customers who never opted in. The bank is right to keep that action out of the account for now, even if some customers want it. An agent holding the login can see the whole account and send money out. An accepted wire is generally final and sits outside Regulation E, the consumer rule for electronic transfers. An ACH transfer the account sends has no chargeback. Getting it back depends on the receiving bank agreeing. Tell an agent to pay, and an ACH payment inside that instruction can count as authorized under that rule. Money lost inside that grant is the customer's. Jamie Dimon called this a liability transfer: "If people are using them, they should understand the risk." That settles the customer's loss, not the bank's exposure. Muse pays through Stripe Link or a single-use virtual card capped at the total the buyer approves. The customer can dispute a wrong charge. The issuer can cancel that card so the next attempt fails. The agent never touches the deposit account. Agentic commerce has to prove itself on the capped card before that access reaches the deposit account.
@convequity

Convequity on X

Neoclouds Now Trade as a Hedge Against AI Demand Growth Convequity’s latest ABBX review shows $CRWV, $NBIS and $IREN are inversely correlated with AI demand growth. Stronger demand widens the working-capital hole and keeps investors risk-off, even as order books expand. Insatiable demand is usually the reason to own the supplier. Here it does the opposite. Jensen Huang told CNBC at GTC Berlin that neocloud demand is “off the charts” and that working capital is the missing piece. Working capital is current assets minus current liabilities. A cash-conversion mismatch squeezes it even when demand is huge. Liabilities hit first. A neocloud pays $NVDA and networking suppliers in roughly 30–60 days. Asset-backed hardware loans add heavy near-term debt service. Assets lag. A signed customer contract sits off the balance sheet as backlog. It only becomes receivables or cash once the cluster is deployed, energized and billed — months later. Cash leaves before revenue arrives. A faster pipeline simply widens the short-term deficit. Latest ABBX figures NBIS: $8.5B gross debt vs $8.0B cash; guidance for >$9B customer prepayments this year. IREN: $7.6B debt vs $5.9B cash; $13.8B hardware and site commitments; 20% prepay on the Microsoft contract. CRWV: $35.6B debt vs $5.5B cash, plus $16.3B operating leases and $35.5B rent not yet commenced. Build-out relies far more on debt than customer prepayments. Even with those prepayments, NBIS and IREN keep raising large extra funding. CRWV faces the same hole with a far larger debt stock. Handing contracts back to resell at higher spot prices would ease CRWV’s cash hole only if customers and lenders agreed. Customers pay even if they do not use the capacity, CRWV must deliver it, and we found no public right to change the price or exit early. Break a customer contract and the hardware lenders can lock up CRWV’s cash. If that lockup holds, they take the cash to repay the loan. Borrowing later would cost more, and the switch only pays while spot stays high. If spot fell, CRWV would lose both the committed revenue and the contract value backing those loans. For investors to turn risk-on again, AI demand growth needs to slow enough for infrastructure supply to catch up.

Investor Letters

All letters

Tower Semiconductor Deep Dive

Tower Semiconductor’s low-loss photonics chips halve lasers in 1.6T optical modules, supporting a $680 million annualized photonics run…

TSEM $19B
CV's Substack 9 clicks

Deep Dive: $ASML

ASML’s €230 million EUV machines underpin advanced AI chip production, with demand booked through 2028 after nearly $9 billion and two…

ASML $616B
Rebound Capital 8 clicks

Margin Call

Petrobras’s Morpho-1 well revived hopes that Brazil’s Equatorial Margin can become a major new oil frontier after hydrocarbons were…

PBR $96B
Doomberg 9 clicks
JBFCY · long

Jollibee Foods CORP

Jollibee Foods trades at PHP 144, about 56% below its January 2019 high, framing a long thesis on the Philippine restaurant operator.

JBFCY $2.5B
Value Investors Club 1 click

Our Views on CoreWeave

CoreWeave's financial statements point to a business model familiar to the authors after a closer review of the AI infrastructure company.

CRWV $34B
WHAT ARE WE DOING? “Contrarians At The Gate” 12 clicks

Capital Wars

CoreWeave’s $35 billion to $39 billion 2026 spending plan relies on debt as unsecured yields reach 15%, making its GPU cloud expansion a…

CRWV $34B
Net Interest 11 clicks

Publications we follow

Every publication on this site, hand-picked.

128 publications

Newsletters 108

YouTube 4

Blogs 15

Seeking Alpha 62 of 253

Nominate a publication

X accounts we follow

Every account on this site, hand-picked.

91 accounts
Nominate an account

Investor letters we follow

190 firms

Sectors

Ordered by how often our editor reaches for them.