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.

From X

@convequity 3 clicks

Convequity on X

$META bulls say its ad cash flow lets it undercut the AI labs on price. The Muse Wonder launch shows Meta matching their prices instead. Meta released Muse Wonder and Muse Wonder Full on Monday and says they score level with Claude Fable 5.1 and GPT-6 Astra on its own tests. Three details from the launch. 1. Wonder costs $4 per million input tokens and $20 per million output tokens on Meta's API. TechCrunch puts that in the same price band as GPT-6 Sol and Opus 5.5. Meta is pricing with the labs, not under them. 2. The stronger model, Wonder Full, is not on the API at all. It is only available to Muse Code subscribers. A developer comparing API models is not buying the model Meta uses to claim the frontier. 3. Both models are distributed only through Meta's own API and Muse Code subscriptions, and the launch says nothing about releasing the weights. I read this as Meta adopting the labs' business model rather than attacking it. The frontier model is a subscription product, and the API is priced at the labs' level rather than below it. That weakens one part of the bull case. If Meta is not undercutting, the price pressure on the labs that bulls expected has not arrived with this launch. The question for the results on 21 October is what Muse subscriptions, Muse Code and the API are earning against the spend on the Prometheus data centre that trained these models, not whether Meta can take the labs' customers on price.
@ParadisLabs 1 click

Paradis on X

Memory fundamentals are still very strong (stronger than Q1). $MU CEO: - "We expect memory and storage supply demand conditions to be much tighter in fiscal 2027 and 2028 than they were in 2026." - "We do not have line of sight to when supply and demand will return to balance." Back in July, $NVDA supply commitments rocketed from $119B -> $279B which was "primarily related to the procurement of memory", with ~$175B of commitments for 2028-29... $MU also said that >75% of FY27 output is already committed + "a majority of discussions with our customers today are already around 2028." So as a reminder....ultra long-term demand visibility in the context of the memory market. Samsung’s Q3 prelim last week was 107.4T Won O.P. on 195T Won revenue: - So that's ~55% margins company-wide. - Micron did 87% GM in their most recent quarter - $SNDK 84.6% GM. All accelerating/stabilizing margins as of their most recent fiscal Q's. Mass production from new capacity (from Micron, Samsung and SKHY) is also not expected to come online for quite some time. In other words, the shortage will intensify next year at least. Then long-term contracts solidify long-term earnings further via price floors + committed volumes: - $MU with 26 SCAs worth >35% of revenue through to 2030 with some now running into 2031. - $MU CFO: "even at floor prices, we expect margins meaningfully above any prior cycle peak margins." - $SKHY has ~10 LTAs of ~5 yrs w/ pricing structures "designed to respond to price fluctuations." - Sammy (as of recent reporting) holds ~70% of memory capacity under LTAs until 2031 with $MSFT, $NVDA and $GOOGL, with ~25% of contract value prepaid. - $SNDK also has their "New Business Model" agreements with 8 customers that lock in ~66% of fy28. Then on buybacks, $MU will step up returns from December - "over time, we expect to return 100% of our excess cash." Just crazy how much cash these companies are bathing in rn. Obviously there are risks, mainly with positioning and sentiment towards the AI trade currently - thanks to macro/ARR fears/pacing talks/other stuff I'm probably forgetting rn. China is ofc another risk with CXMT but I don't see how they'll be a massive problem for HBM / US hyperscaler RDIMM where they'll continue to face US restrictions + massive qualification walls imo. But ultimately, memory fundamentals are in-tact and very strong. Stronger than Q1.

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…

Margin Call

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

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.

TCEHY · long

Tencent Holdings LTD

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

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…

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…

Publications we follow

Every publication on this site, hand-picked.

129 publications

Newsletters 109

X accounts we follow

Every account on this site, hand-picked.

91 accounts

Investor letters we follow

190 firms

Sectors

Ordered by how often our editor reaches for them.