New Ask Claude or ChatGPT about any stock, using Bullpen. Bullpen in Claude or ChatGPT. 7 days free

Tuesday, November 10, 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.

Seed Corn

Chiron’s failed $49 million Beaumont sale and $178 million Pinnacle acquisition leave XRN with $70 million of liquidity and medical-office…

STRL: 80 new electricians a week

Sterling’s $4.1 billion backlog faces Texas permitting risk after a governor proposed a moratorium, while Connexion, Capital Asset Planning…

Roblox: Capture This Moment

Roblox’s Q3 FY26 DAUs fell year over year for the first time, while bookings declined 12% to $1.7 billion.

DoorDash and Google

DoorDash and Google.

Diamond Estates: Research Note

Diamond Estates converts $4.7 million of Lassonde debentures at a 69% premium, cutting debt to $10.9 million and raising Lassonde ownership…

AVO · long

Mission Produce INC

Mission Produce (AVO) is pitched long as a consumer-staples investment, countering a September 2024 short case.

Vicor Update, Part Four

Vicor’s Federal Circuit hearing put Foxconn’s disputed patent license under pressure, reinforcing its AI server royalty model and licensing…

Floor & Decor 3Q26: Business Update

Floor & Decor’s 3Q26 sales rose 3% while comparable sales fell 1.4%; it cut 2027 openings to 10-15 as high mortgage rates pressure…

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

360 firms

Sectors

Ordered by how often our editor reaches for them.