Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

Tuesday, November 3, 2026

Back to today

AMD Earnings Review

AMD reported $13.4 billion in Q3 revenue, with Data Center sales up 97% to $8.5 billion, and guided Q4 revenue to $17 billion.

AMD $660B
Cestrian Capital Research 3 clicks

From X

@StableBread 1 click

Fajasy on X

Even if the AI model Nubeva Technologies $NBVAF (TSXV: $NBVA) is training to read malware binaries (compiled programs) works, six risks could keep it from paying off: → Labs closing the gap: OpenAI’s GPT-6 Astra nearly doubled GPT-5.6-sol’s score on SRE-Bench, a Columbia University-led test of AI models on reading binaries, within a month of SRE-Bench’s release, and Anthropic says its Claude Mythos Preview model can already reconstruct plausible source code from stripped binaries. The better the labs’ own models get at reading hard binaries, the less they need to pay for outside labels. → Big tech building in-house: Microsoft’s $MSFT Project Ire reverse engineers software files on its own to decide whether they’re malicious. Ire caught only about a quarter of the malware among nearly 4,000 hard-to-classify files in a test Microsoft described in August 2025. But in June 2026, Microsoft said Ire flagged a new malware sample that most major security products, including CrowdStrike’s $CRWD and SentinelOne’s $S, still missed. Google $GOOGL already owns VirusTotal and Mandiant, so the likelier customers are labs that don’t own a malware-scanning site or an incident-response firm already, like OpenAI and Anthropic. → Paid like a contractor: Data labeling companies (firms that turn experts’ knowledge into training data for AI labs) pay 60-70% of what they bill to their contractors, and Scale AI is already recruiting its own malware analysts. So a data labeling company could pay Nubeva by the hour or task, or hire analysts itself, instead of licensing the model or buying the company. → Selling labels teaches the buyer: Every binary Nubeva labels helps the buyer’s model learn the same skill, the reason CEO Randy Chou gave for not patenting how Nubeva reads binaries. So each sale could shrink the edge Nubeva is selling. → Cash and compute: Chou gave no cost for running “hundreds and thousands” of AI agents on a private model. For reference, Nubeva spent $714,772 on R&D in FY2026, equal to 28.7% of its $2.49M in cash as of April 30, 2026. → Missed targets: Nubeva’s AI decryptor builder missed four targets starting in August 2023, so “days and weeks” (Chou’s timeline for the model) is a target, not a result just yet. So even with a working model, Nubeva still has to (1) sign a contract before the labs can read the hardest binaries themselves, (2) get paid for its model and engineers instead of by the hour or task, and (3) fund the training without draining its cash.
@convequity 3 clicks

Convequity on X

The market still prices $NVDA's moat on one idea: the GPU is special because it can do anything. But Nvidia has spent every generation trading that flexibility for raw AI horsepower. Its chips now look less like general-purpose GPUs and more like the AI-specific chips Google and Huawei build. That's great for performance, but it chips away at the very thing that set Nvidia apart. The closer its GPU gets to a TPU, the easier it becomes to replace. Deeper dive available at Convequity.
@StableBread 1 click

Fajasy on X

AI labs and data labeling companies (firms that turn experts’ knowledge into training data for the labs) could make my $12-19M sale range for Nubeva Technologies $NBVAF (TSXV: $NBVA) look conservative. My previous write-up’s sale comps were five cybersecurity companies buying small startups that detect, stop, or recover from ransomware, and I valued a Nubeva sale at $12-19M, the low end of those deals. AI labs and data labeling companies would widen the set of buyers beyond cybersecurity companies, but I found none that has bought a ransomware recovery company like Nubeva. The closest deals were for a stake in a ransomware recovery company, malware data, and specialist teams: → Microsoft $MSFT: Made an equity investment in Veeam, a data protection and ransomware recovery company, in February 2025, as part of a partnership to build AI-based threat detection and other Microsoft AI services into Veeam’s products. → Google $GOOGL: Owns VirusTotal, a malware-scanning site it bought in 2012, and Mandiant, the incident-response firm it bought for $5.4B in 2022. Its experimental Sec-Gemini model, introduced in April 2025, draws on Mandiant’s threat intelligence data. → OpenAI: Agreed to buy Promptfoo in March 2026 for an undisclosed price. Promptfoo, which tests AI systems for security holes, was valued at $86M after its July 2025 funding round, per PitchBook. → Mercor (a data labeling company): Agreed to buy Deeptune, which builds practice environments for AI agents, in July 2026, months after Andreessen Horowitz led Deeptune’s $43M Series A. The two 2026 deals didn’t disclose prices, and Mandiant and Veeam are far larger businesses than Nubeva, so I can’t put a new number on a sale. Still, my $12-19M range is probably conservative now, since it only counted cybersecurity companies as buyers, and an AI lab or data labeling company would be paying for Nubeva’s engineers and model rather than its ~$500K of FY2026 revenue. But an AI buyer wouldn’t get patents on what it’s paying for. Nubeva’s patents would cover how it recovers files and how it detects hidden encryption, but it won’t patent its binary analysis (figuring out what malware programs do) at all, and CEO Randy Chou explained why: “Anything you patent, that means you have to show how you did it. If you show how you did it, well, somebody can train an LLM to go do it.” So the model is only worth what Nubeva can prove it does, through a signed contract, a named partner, or test results.
@convequity 1 click

Convequity on X

3D chips could create more room for specialist EDA startups. We are not convinced that more difficult chip design automatically translates into a large revenue windfall for $SNPS, $CDNS and Siemens. Fine-pitch stacking changes decisions that used to be made within one die. Put a circuit upstairs and its wires may get shorter, but its heat affects the circuit below. Placement, signal timing and power delivery have to be optimized together across the stack. That creates a reason for $AVGO and $NVDA to build custom tools around their architectures. Broadcom already describes a proprietary design flow built on industry-standard tools. Nvidia publishes its own research on assigning connections across tiers in 3D chips. Our inference is that the chip designers will keep some of the most differentiating optimization work in-house. A startup that solves a costly gap in cross-die planning or optimization could also win business without replacing an entire EDA suite. The incumbents still supply implementation, verification and manufacturing sign-off tools, and they are developing 3D capabilities themselves. Internal tooling is evidence that customers need more than the standard flow; it does not establish that incumbent spending falls. We see a more interesting startup opportunity where a tool can shorten repeated design iterations or improve the resulting chip enough to earn a separate budget. That is the part of the 3D transition we would investigate before assuming the largest EDA vendors capture most of the upside.

Investor Letters

All letters
BSP · short

Bending Spoons Spa

Bending Spoons is a software and digital-assets holding company that has completed more than 50 acquisitions since 2013.

BSP $29B
Value Investors Club 1 click

FTAI Aviation Update, Part Two

FTAI Aviation’s $234 million Jereh JV deposit funds Power inventory, while customer concentration, lower cash guidance and pending U.S.

FTAI $20B
CV's Substack 5 clicks

Publications we follow

Every publication on this site, hand-picked.

129 publications

Newsletters 109

YouTube 4

Blogs 15

Seeking Alpha 73 of 273

Nominate a publication

X accounts we follow

Every account on this site, hand-picked.

91 accounts
Nominate an account

Investor letters we follow

338 firms

Sectors

Ordered by how often our editor reaches for them.