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 5 clicks

From X

@convequity 16 clicks

Convequity on X

Cerebras ($CBRS) is a taxi built for very fast inference while Nvidia ($NVDA) runs the bus network. Most of the confusion around CBRS comes from reviewing the wrong vehicle. In our view, Cerebras' wafer-scale design and abundant on-chip SRAM let it reach high utilization at batches of roughly 3–4. That makes it exceptionally fast for low-batch inference. GPUs need much larger batches and tightly connected clusters to get there, but their flexibility, scalability and throughput make them the better mass-market token factories. The bus wins on economics when serving many users at once. We think Cerebras is mostly attractive over the next two to three years, provided demand grows enough for premium-priced, high-speed inference rather than mostly low-cost, high-volume tokens. How big that premium segment gets is the main variable in our thesis.
@convequity 7 clicks

Convequity on X

Nvidia's DSX POD container shortens the gap to system-level competitors like Google and Huawei, but it does not eliminate that gap. At GTC Berlin on 21 October, $NVDA introduced DSX POD, a partner-built container for Vera Rubin AI factories. Nvidia says it cuts deployment at a prepared site from six months to as little as 14 days, with worldwide delivery targeted for early 2027. Our latest Nvidia report, published a few days ago, flagged one drawback: Nvidia is not expanding into the full system solution quickly enough. Huawei, Google and Broadcom are leveraging system-level optimization, where the chip, the racks, the network between racks, power, cooling and software are designed together for one owner's workload. That is where they are finding further efficiency gains in the competition against Nvidia. Nvidia has stayed within rack-level optimization: the GPU, the NVLink switch and the rack they sit in. That is understandable. Nvidia serves many customers with many different designs and criteria, so it has to find the most common denominator those customers can accept. It also means this handicaps Nvidia's competitiveness against vertically integrated competitors like Google and Huawei. A container sits between a rack and a full system. Partners integrate the Vera Rubin racks, the networking, the power and the cooling inside the container and validate them together before delivery, so the customer receives a tested unit rather than racks to assemble. In our view that is why the gap shortens. It is still a gap, because Nvidia is not a vertical player and still does not provide the system-level solution. What DSX POD does show is that Nvidia is adaptive and very eager to stay in the game. We expect it to try whatever it can to alleviate this drawback.
@convequity 2 clicks

Convequity on X

$META's Muse progress confirms our view that Meta owns the essential assets for frontier AI, and it is progressing better than most people believe. Zuckerberg called Muse Wonder Meta's first frontier model on the 21 October Q3 2026 call. More than 10 million people set up Muse in about six weeks, and shopping is the largest use case. We expected Meta to come back, and we are delighted to see it back among the frontier labs. Two things concern us. 1. Meta appears to have recognized it will not catch up on frontier enterprise use cases, coding above all. So it is differentiating on consumer. In our view that is a double-edged sword. Consumer is where Meta's assets are strongest. But we think the next big shift is silicon-based consumption squeezing out carbon-based consumption: spending by AI agents replacing spending by humans. Humans have finite attention and a limited addressable market, so human demand grows roughly linearly. AI agents have a near-unlimited spending demand, and we think it is about to go exponential. A consumer-first AI is built for the side that grows linearly. That is a longer-term concern for Meta. 2. We are disappointed that Muse is only available in the US, not in the rest of the world. That is cautious big-company behavior, closer to Google than to SpaceX. Our guess is that the legal warnings and threats Meta keeps receiving have made it extra careful on the legal side. That caution has a cost: it handicaps Muse against the other leading frontier apps.

Investor Letters

All letters

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 6 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 5 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” 10 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

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

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 61 of 249

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.