Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Friday, October 23, 2026

Back to today

From X

@convequity

Convequity on X

Why I think Nvidia buying Reflection AI helps Palantir more than it hurts $NVDA wants its models inside enterprises, not just hyperscalers, neo-clouds and AI labs. Buying Reflection AI would speed that up. The FT reported on 10 Oct that Nvidia is in early talks to buy Reflection, raise its stake, supply it more chips, or hire its team and license its technology. Bloomberg's source says no terms have been reached. Nvidia has already invested $800m, so a buyout or a bigger stake means pricing a company it already partly owns. The enterprise push Nvidia agreed in September to buy Hugging Face for $12.9bn. Jensen Huang says more than 200,000 companies use it to deploy AI. Reflection is validated on Dell's AI Factory with Nvidia, and $PLTR's sovereign AI stack runs on Nvidia's AI factory design. Reflection is built for enterprise code. Its Asimov agent reads a company's code, Slack messages, emails and documents. Its first model, Beam (501bn parameters, 23bn active), is trained for coding. Reflection says the weights ship this month under Apache 2.0, for on-prem or air-gapped use. The bear case: Nemotron is moving too slowly at Palantir One reading is that Nvidia wants Reflection because its Nemotron models are moving too slowly with Palantir. I think that is unlikely, given how Palantir has executed. At AIPCon on 10 Sep, Nvidia's supply-chain team said Nemotron 3.5 Lightning, post-trained in Palantir, beat Nemotron 3 Ultra by a large gap on Nvidia's own benchmark. It runs on-prem and retrains on one GPU in a couple of hours. Cisco will offer Nemotron post-trained with Palantir in its AI Factory with Nvidia. L3Harris said open models fine-tuned on its own data beat the frontier models it had used within 48 hours, at 95% lower cost. Palantir has kept shipping open-model deployments. In June it launched an engine to run Nemotron in sovereign environments, and customers own the retrained models. On 1 Oct Armada became its first certified modular data-centre partner. On 12 Oct $ONDS formed Regna AI on AIP and open-weight models. The likelier case: Nvidia is backing more than one route Palantir fits open models to a company's data, permissions and workflows. Reflection is built to read the company's code. Buying a team that has already built that is quicker than building it in-house. The limit is vendor API rules Reflection reaches a company's systems through the vendors' APIs, and vendors decide what passes through them. $SAP's April API policy bars AI agents that plan or execute API calls, and bulk data extraction, outside SAP-endorsed pathways. Since May 2025, Slack's API terms bar third-party apps from using Slack data to train a large language model or bulk-exporting messages. Those rules apply to Palantir too. In May, SAP made Palantir AIP for data migration an SAP Endorsed App. An independent Reflection would likely need similar approval from each vendor whose systems it reads. Who pays the compute rent Reflection trained Beam on Nvidia GB300s rented from $SPCX at Colossus 2 in Memphis, under a June deal worth up to $6.3bn: about $150m a month through 2029. $NBIS agreed in July to sell it $1bn more capacity. If Nvidia buys Reflection, it takes on those contracts and pays SpaceX rent on Nvidia chips that SpaceX bought. In an acqui-hire, the contracts stay with what remains of Reflection unless the deal says otherwise. The FT says an acqui-hire would likely avoid a long regulatory review. Nebius holders should watch which company ends up owing that $1bn contract. What it means for Palantir The risk is an independent Reflection selling its own forward-deployed team and turnkey stack, which is how Palantir sells. If Nvidia owns Reflection, I expect Beam to reach AIP customers the same way Nemotron does, with each customer fine-tuning it on its own data inside Palantir.
@StableBread

Fajasy on X

In November 2025, BWX Technologies $BWXT sold $1.25B of 0% convertible notes due November 2030 and used $830.8M of the proceeds to repay its old credit facility. The conversion price is $262.51/share, 80.2% above $145.70, and $131.9M of capped calls offset dilution up to $396.24/share. Dilution has been minimal so far. Diluted shares have held at ~92M since 2023, helped by small buybacks ($30M in 2025). But stock comp rose to $20.1M in H1 2026 from $13.7M a year earlier, although it’s still only 1.1% of revenue, up from 0.9%. Management also said on the Q2 call it doesn’t plan buybacks in 2026, despite $347.6M left on the authorization, so nothing will offset the new shares. Even after paying ~$200M for Precision Components Group (PCG) on July 1, $BWXT’s leverage is within its 2-3x target: → Debt: $2.05B ($400M of 4.125% notes due June 2028, $400M of 4.125% notes due April 2029, and the $1.25B convert). → Cash and liquidity: $400M of cash and $1.7B of total liquidity, including an undrawn $1.25B revolver running to November 2030. → Leverage: 2.7x net debt to TTM adjusted EBITDA, or $1.65B of net debt on $602.1M of EBITDA. The $750M from selling its medical isotope business would cut leverage to 1.5x. → Ratings: Fitch BBB (investment grade) since September 2026; S&P BB+ and Moody’s Ba1, one notch below. → Other obligations: $152.0M of underfunded pension and retiree benefit plans. Net debt has hovered between $1.0B and $1.5B since 2023. $BWXT pays a dividend of $0.27/quarter ($1.08/year, a 0.7% yield), raised from $0.25 in February 2026. The company expects $5.5B of capital to deploy through 2030 at 2-3x leverage, with $3.5B left after capex and dividends. That’s what CFO Mike Fitzgerald said at $BWXT’s Investor Day on September 29, 2026, and he also put organic growth first for the $3.5B. But he said $BWXT is “not anchored to the same size of transactions” it has done before, and the deck lists “Disciplined Acquisitions” among the revenue drivers behind the 2030 targets. In other words, $BWXT has the track record and, once the medical sale closes, the balance sheet to go after its 2030 targets.
@convequity

Convequity on X

We think Google's ($GOOGL) TPU volumes alone could add roughly $11 billion to $14 billion a year to Broadcom ($AVGO) revenue on our illustrative estimates, and Anthropic and OpenAI are a larger opportunity on top. April supply-chain estimates put Google's TPU demand at 4.2 million to 4.5 million units in 2026, rising above 6 million units in 2027. We allow for MediaTek's allocation to rise from around 600,000 units in 2026 to 1 million units in 2027 and assume Broadcom supplies the remainder. At the 6 million-unit level in 2027, that leaves Broadcom with roughly 1.1 million to 1.4 million additional chips compared with 2026. At an illustrative $10,000 of revenue per chip, that would add $11 billion to $14 billion in annual revenue from higher volumes alone. The actual increase will depend on which designs ship and their selling prices. Our working estimates put Anthropic's demand for Broadcom-supplied TPUs at around 1 gigawatt in 2026, rising to 5 gigawatts in 2027. OpenAI's separate custom-chip program has demand estimated at 1.3 gigawatts in 2027 and 5 gigawatts in 2028. Anthropic's TPU purchases overlap with the Google TPU opportunity, so we count those sales once. Together, the Anthropic and OpenAI estimates point to around 10 gigawatts of demand by 2028. Assuming Broadcom earns $10 billion to $20 billion for each gigawatt supplied, that represents a $100 billion to $200 billion revenue opportunity across those deployments.
@convequity

Convequity on X

We added to Tesla ($TSLA) as a long-term holding in physical AI through Optimus and Cybercab, and we keep the position small. Our view is that an affordable, capable Optimus could become an enormous product, extending Tesla's opportunity beyond its vehicle business. Tesla's batteries and its ability to coordinate difficult industrial supply chains add to that potential. The position stays small because reliability in real working environments, affordability and commercialization timing are unproven. Demonstrations alone do not establish that Optimus is ready for sustained use, and commercialization could take longer than investors expect. Reports that China was slowing humanoid-company IPO activity reinforce our concern that financial enthusiasm can advance faster than the underlying technology. A rumored Tesla facility in Ningbo, near Shanghai, could strengthen the industrial case if confirmed. Its proposed role is to coordinate Chinese suppliers and validate components for Optimus V3. The associated target of 50,000 Optimus units by year-end remains unconfirmed and ambitious, and we do not build our case around that volume. Our interest rests on the eventual product opportunity and Tesla's ability to solve production problems.

Investor Letters

All letters

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 11 clicks

Publications we follow

Every publication on this site, hand-picked.

129 publications

Newsletters 109

YouTube 4

Blogs 15

Seeking Alpha 56 of 248

Nominate a publication

X accounts we follow

Every account on this site, hand-picked.

90 accounts
Nominate an account

Investor letters we follow

190 firms

Sectors

Ordered by how often our editor reaches for them.