Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Friday, October 2, 2026

Back to today

From X

@convequity

Convequity on X

The $PLTR and $NVDA work on Nemotron, NVIDIA's open model family, is still under the radar, unlike partnerships that promise more than they deliver. In our view it could be industry-transforming, because each side needs the other: Nemotron is the Western open model that fits, and Palantir is the only vendor that can put a model into an enterprise securely, safely and privacy-first, and limit it to what each person may see and change, which is why AIP, Palantir's AI platform, took hold and why NVIDIA needs Palantir. Palantir has no proprietary model and leans on closed models from OpenAI and Anthropic. Chinese open models are almost certainly off the table for a U.S. government contractor. On August 26 Jensen Huang said every major company needs its own proprietary AI, and open models make that possible. On June 29 Palantir launched an engine that runs custom Nemotron models for U.S. agencies and critical industries on networks cut off from the outside. The customer can train this American open model further on its own data and keep the weights, the trained model. On September 10 they announced a supply-chain stack, first on NVIDIA's own supply chain. On July 14 NVIDIA said other companies already specialize Nemotron for their own domains. The test is whether that stack moves from NVIDIA to named customers in manufacturing, energy, healthcare, automotive or aerospace.
@convequity 1 click

Convequity on X

An optical circuit switch never converts the light. A conventional switch takes light in, turns it into electricity, makes a decision, turns it back into light. An optical circuit switch steers the beam. The principle in the note: rather than optimizing a difficult conversion step, remove it. Three constraints follow. 1. Power. A 100-terabit-class silicon photonics OCS consumes only 200–300 W, against roughly 6,000 W for an equivalent electrical switch. More than a 20× advantage. The 200–300 W figure is measured with the amplifiers already in place. 2. Rate. The switch is optically transparent, so the same module supports 800G, 1.6T, 2.4T and 3.2T. It does not have to be replaced each time the transceivers get faster. 3. Insertion loss. Silicon photonics currently loses more light in transit than the MEMS alternative, so semiconductor optical amplifiers are added to restore optical power. That adds system complexity. Google buys these switches from optics makers, not from the usual switch-chip vendors. Calient, Lumentum ($LITE) and Coherent ($COHR) supply Google ($GOOGL).
@convequity 5 clicks

Convequity on X

Natural gas itself has not been the scarce input. The turbines have. Once AI data-centre load outran what the grid, renewables, and near-term nuclear could cover on a GW scale, gas became the compromise — and turbine demand surged from 2024–2025. Lead times still run roughly three to five years, with manufacturers already sold out into 2029–2031. That puts the first large batch of those ordered turbines on a delivery horizon closer to the next twelve months. As they arrive, and as behind-the-meter projects sit at or near gas fields, the incremental gas draw is likely to show up in a concentrated window, not a smooth ramp. Storage has grown only modestly while demand has risen, and drilling and pipeline buildout have stayed slow. New customers do not want to wait another three to four years for later turbine slots, so they are also exploring workarounds: reciprocating engines — internal combustion units closer to older, smaller airplane engines than to jets. The tech is not high-end blade metallurgy, so these generators can be mass-produced. They still prefer natural gas (more energy at the same density, and cleaner). Slightly less efficient than a turbine, but available sooner — and still burning gas. Either path concentrates the fuel draw. $AR, $EXE and $CRK are the producers sitting on that timing gap.

Investor Letters

All letters

Nike, Part Two: The Deep Dive

Nike’s second Latticework program examines the company and the global athletic footwear and apparel industry for professional investors.

NKE $53B
Latticework by MOI Global

Nike, Part One: The Story

Nike grew from a $1,000 shoe-distribution venture in 1964 into the largest sports brand, using athlete endorsements, product innovation and…

NKE $53B
Latticework by MOI Global 1 click

Toll roads in the sky

Asia's elevator boom favors maintenance and modernization providers, with Otis deriving nearly 90% of operating profit from after-sales…

OTIS $24B
Asian Century Stocks 1 click

Publications we follow

Every publication on this site, hand-picked.

110 publications

Newsletters 93

YouTube 4

Blogs 12

Seeking Alpha 41 of 198

Nominate a publication

X accounts we follow

Every account on this site, hand-picked.

72 accounts
Nominate an account

Investor letters we follow

189 firms

Sectors

Ordered by how often our editor reaches for them.