Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Saturday, October 24, 2026

Back to today

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 rate and customer wafer prepayments through 2027.

TSEM $19B
CV's Substack 7 clicks

From X

@StableBread 5 clicks

Fajasy on X

I went through every Boost Run $BRUN contract with disclosed terms to see how much of its $2.6B+ of total contract value (TCV) they cover. Cohere, the Canadian developer of AI models for businesses and governments, agreed on September 30, 2026, to pay $BRUN $525.6M over the agreement’s term for $NVDA’s GB300 NVL72 racks, each with 72 Blackwell Ultra GPUs. Part of the $525.6M is a prepayment $BRUN hasn’t disclosed. But if we follow the 22% average $BRUN says its prepayments run, the prepayment would be ~$116M. Each rack’s term runs 60 months, and the first racks should start in Q2 2027. Cohere can terminate and get back everything it prepaid if it hasn’t accepted a minimum number of racks by July 15, 2027. $BRUN has disclosed terms for seven contracts, and the two largest are with Cohere and Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati: → Fluidstack (November 6, 2025): $63.6M over 24 months, or $31.8M/year. → 1,280 B300s, Charlotte (March 15, 2026): $116.1M over 36 months, or $38.7M/year. → 256 H200s, Raleigh (March 16, 2026): $3.7M over 12 months, or $3.7M/year. → Thinking Machines Lab (May 21, 2026): $471.7M over 36 months, or $157.2M/year. → June order, 240 GPU servers (June 8, 2026): $207.6M over 36 months, or $69.2M/year. → July order, 1,536 B300s (July 9, 2026): $222.5M over 48 months, or $55.6M/year. → Cohere (September 30, 2026): $525.6M over 60 months, or $105.1M/year. --- → Total: $1.61B, or $461.3M/year. Unnamed customers are labeled by what they rent, including $NVDA’s current B300 (Blackwell Ultra) and older H200 GPUs. $BRUN’s TCV is over $2.6B, so the other $989M+ is in deals without disclosed terms. The October 6 release calls the $2.6B “total revenue customers have committed to pay,” but the August 14 Q2 deck also counted renewal options and LOIs in TCV, and $BRUN hasn’t said whether the definition changed. So the conservative read counts only the seven contracts with disclosed terms, worth $1.61B, not the $2.6B headline.
@BrokenMoats

Broken Moats on X

Specifically on $DOCU which is approaching levels not witnessed since 2021. Not only is there the rates sensitivity (usually software long duration, higher rates should hurt (been opposite this rate move for now) and the exposure to AI threats (how hard is it to rewire an esignature) even if you want to put those two aside, the business has a significant exposure to the real estate (mortgage, refi, etc) industry and that by all accounts is not performing well (see starts data today, number of existing closings, etc)... seems like this lack of pricing fundamental developments of the business is the tell tale that the move is not based in anything more than a squeeze it would seem. Would love to know from any buyers, why buy here, and what causes you to sell (purely momentum?) It would seem a deep rerating (the opposite from last earnings set up) might be coming for the software trade (especially if AI trade comes back to close out the year). Upsetting they are so inversely correlated, because AI rate of change spending levels does not matter, current AI is enough to present problems for these software providers, at some point that linked should break as AI spend less still won't save software names from their problems
@convequity 7 clicks

Convequity on X

Networking, not compute, now decides custom AI chip wins, and in our view that is Broadcom's ($AVGO) edge. As GPUs become more like specialized AI processors, matrix compute looks much the same everywhere. At 224G high-speed chip connections (SerDes), only Nvidia ($NVDA) and Broadcom can do it. If you want a custom chip, you go to Broadcom. That is why we avoid Marvell ($MRVL). A cheaper design partner is not necessarily cheaper overall: weaker connections and networking compound costs after deployment, resulting in higher total system costs. The bottleneck is moving from the switch to the optics. Lasers and optical signal-processing chips are short, and Broadcom can bundle priority optical supply with its custom chips, even for customers running Nvidia GPUs. Nvidia's lack of optical know-how is becoming its biggest disadvantage, and we expect Broadcom to take more share from it for years to come.
@convequity 3 clicks

Convequity on X

$CRWV 5-year CDS hit a record ~1,195bp on 15 Oct and the whole neocloud group sold off with it. Our ABBX tracker suggests the market is pricing contagion into the wrong names. The balance sheet first. At 30 June CoreWeave had $35.6B of debt against $5.5B of cash, and its ~$33B market cap now sits below its debt. $4.4B falls due in the rest of 2026 and $6.2B in 2027, mostly loan amortisation; no bond matures before 2030. Its 9.625% notes sold at par in June now yield ~14.9%. Add $16.3B of lease liabilities and $35.5B of leases not yet started. CoreWeave's Q2 operating cash flow: Net loss: −$0.63B + Depreciation & amortisation: +$1.39B + Stock-based comp: +$0.17B + Other non-cash items: +$0.28B − Working capital (payables, receivables, prepaids): −$1.32B + Customer prepayments: +$0.79B = Operating cash flow: $0.68B The core business makes cash, but working capital ate it. Without the prepayments OCF was −$0.11B. That OCF is already after ~$0.44B of cash interest, equal to 65% of it, and capex ran 9.5x OCF. Funding is where the group splits. CoreWeave builds with borrowed money: GPU-backed loans, 8.5–9.75% bonds and vendor credit, with ~$9.8B of H1 equipment supplier-financed outside the cash-flow statement. Prepayments covered 12% of its Q2 capex. $NBIS builds with convertibles (~$9B this year) and prepayments that covered 21% of Q2 capex; at 30 June it held $8.5B of debt against $8.0B of cash. $IREN builds with equity, a $3B 1% convertible and prepayments that covered 87% of its latest quarter's capex; net debt is ~$1.7B and cash interest equalled ~1% of OCF. The exposure that matters is its landlords. CoreWeave is ~77% of $CORZ revenue, the tenant at Galaxy's Helios campus and leases 400MW from $APLD. Yet since 2 Oct Core Scientific is down 17%, while Microsoft-anchored IREN is down 34% and Nebius 30%. Our view: the risk is CoreWeave's debt-funded build, not the neocloud model. If CoreWeave stumbles, the hit lands on the landlords that depend on its rent. Nebius and IREN share the model, not the funding.

Investor Letters

All letters

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 7 clicks

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” 11 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 8 clicks
TCEHY · long

Tencent Holdings LTD

Tencent, valued near $500 billion, is a long idea built on its Chinese gaming, social-network and advertising franchises.

TCEHY $449B
Value Investors Club

Putting bear cases to rest

Medpace authorized $1 billion of buybacks through 2028 after Q3 revenue rose 8.5% to $716 million; service revenue grew roughly 12% as…

MEDP $18B
Best Anchor Stocks 12 clicks

Publications we follow

Every publication on this site, hand-picked.

129 publications

Newsletters 109

YouTube 4

Blogs 15

Seeking Alpha 62 of 252

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.