Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Saturday, November 7, 2026

Back to the front page

From X

@convequity

Convequity on X

I think some bears underestimate $PLTR's model expertise. We have argued that Palantir had more model expertise than it had shown publicly, and that lacking its own foundation model did not put it at the competitive disadvantage those bears assumed. The earlier NVIDIA partnership, which included NeMo Retriever and Nemotron, was my first clue. I see the public Nemotron results as evidence of that expertise. These results fit our argument and were what I expected. Palantir's slide on McCarthy's construction requests for information shows a tuned Nemotron model outperforming Sonnet and Astra on classification. FourWeekMBA reads the chart, which has no data labels, as approximately 73% balanced accuracy for tuned Nemotron, approximately 66% for Sonnet and approximately 62% for Astra. On a cost basis the slide leaves unexplained, FourWeekMBA places Nemotron at approximately $0.10 per million tokens. The slide leaves open whether that includes hosting, tuning and Palantir's software fees. I would want the full customer cost comparison to judge the savings. I still regard Nemotron as a weak starting model. If Western developers make state-of-the-art open-source models available, I expect Palantir to use customer data, task-specific benchmarks and tuning to deliver further improvements in performance and cost. That could make Palantir economical for smaller enterprises as well as very large ones. Savings on model usage could cover Palantir's software premium if customers can move enough work to cheaper tuned models while maintaining the quality they need. As customers spend more on tokens relative to software, the potential savings grow. I expect Palantir's model expertise to become more valuable because it could reduce a larger part of the customer's total bill.
@aleabitoreddit 1 click

Serenity on X

I’m just calling out what I see with a failure in corporate communication, when so much is at stake. I’m really not sure why SK Hynix/Samsung representing SK’s economy let narratives of memory oversupply be dominated by their own customers on X. Or why $CBRS, which lost billions in valuation after the short seller report, is waiting for an analyst question in earnings to fix the damage Or why many of the “guppies” that would face a death in the Neocloud market aren’t actively fighting back against that claim here. A lot of narratives and claims originated and were distributed via X. Even if you don’t like Elon or X, even the Democratic Party combats narratives here as seen with the ratio on the VP’s New York post I don’t see why this can’t be done with AI companies calling BS on $4 B300 price gluts.

Investor Letters

All letters

Axon 3Q26 Business Update

Axon cut full-year guidance after delayed federal Dedrone contracts slowed 3Q revenue growth to 28%, while software and services grew 43%.

AXON $31B
Speedwell Memos 2 clicks

ARM Q2 FY3/27 Earnings Review

Arm reported Q2 revenue of $1.443 billion, up 27%, as licensing rose 50% while royalties grew 8%; Q3 guidance calls for $1.54 billion.

ARM $181B
Cestrian Capital Research

AppLovin 3Q26: Business Update

AppLovin's 3Q26 revenue rose 55% year over year and margins reached 78.2%, driven by AXON model gains, while publisher bidding pressure…

APP $118B
Speedwell Memos 2 clicks
AAP · long

Advance Auto Parts INC

Advance Auto Parts, AAP, is presented as a long idea following prior write-ups at $48.81 per share in January 2025.

AAP $2.9B
Value Investors Club 1 click

Publications we follow

Every publication on this site, hand-picked.

129 publications

Newsletters 109

YouTube 4

Blogs 15

Seeking Alpha 74 of 276

Nominate a publication

X accounts we follow

Every account on this site, hand-picked.

91 accounts
Nominate an account

Investor letters we follow

359 firms

Sectors

Ordered by how often our editor reaches for them.