Thursday, October 8, 2026

@convequity Oct 7 2 clicks

Convequity on X

More $NVDA chips could end up going to SpaceX and other buyers that can borrow at investment-grade rates, leaving neoclouds with fewer chips than they need. SpaceX's planned $40bn chip borrowing shows why. The deal: $SPCX is in early talks for about $10bn of bank loans and $30bn of investment-grade debt, led by Apollo, with Pimco also in talks. It would close in 2027, and the talks could still fall apart. The comparison: in December, Apollo lent about $3.5bn to fund xAI's Colossus 2 chips through a separate company set up just for that deal. The interest rate was about 10% a year. xAI is now part of SpaceX. Investment-grade debt usually costs much less. US investment-grade corporate bonds yield around 6% on average today. The rate on the new $30bn has not been reported, so we don't know the exact saving yet. Why it could be cheaper: lenders would be relying on SpaceX's launch and Starlink businesses to repay, not only on the chips. That is the real financial payoff of putting xAI inside SpaceX. Most GPU buyers can't borrow like that. $CRWV is rated junk (Moody's Ba3, Fitch BB-) and pays 9.25% on its 2030 bonds. Standalone AI labs have no public credit ratings that we could find. Not every neocloud pays up. $NBIS has a GPU loan at a floating benchmark rate plus 2.5%, against CoreWeave's plus 5.5%. Our view: the debate on X is whether this proves AI demand or shows too much leverage. We think the bigger point is cost of capital. When chips are scarce, the buyer that borrows most cheaply can pay for the most of them, and neoclouds paying junk-level rates get squeezed out.

@negligible_cap Oct 7

Negligible Capital on X

What's wrong with $BSP ? https://t.co/XXaB7KnW0R

@convequity Oct 7

Convequity on X

We agree with Ampersand: an agent is vaporware without deep access to the customer's systems of record. But the result is headless SaaS, where the agent becomes the interface, and headless SaaS does not keep today's economics. Convequity's SaaS Survival Scorecard splits a vendor into three layers, the interface, the workflow and the data, and scores how many of the three it keeps once agents arrive. 1. Deep integration into the system of record is the headless path: the user works through Claude, ChatGPT or another third-party agent, and the vendor keeps workflow and data underneath. Our scorecard calls this Route 2: roughly half the SaaS value is replaced, and the vendor is survivable but diminished unless outcome pricing rescues it. It is the most likely near-term route, and a starting point rather than a stable endpoint. 2. Model companies will expand into the workflow layer to keep growing, and a vendor whose workflows mainly connect other SaaS APIs is highly absorbable. A vendor left with only its data layer is plumbing: it still earns on agent traffic, but it gives up a lot of the value. A vendor that keeps hard-to-replace workflows retains more of it. 3. More agents does not mean more SaaS revenue in every domain. Back-office cost centers like HR, accounting and IT service management have a natural ceiling, because making the work cheaper does not create more of it. Agents cannot create demand for an extra tax filing. Revenue-generating functions like sales and marketing can grow with agents, because extra work there produces extra revenue. 4. Keeping all three layers is rare. $PLTR is the full Route 3 case in our scoring so far, because its ontology works as a quasi system of record and it has model-layer leverage. $PANW approaches Route 3 only within security operations, where sensors and inline enforcement are hard for an agent to absorb. Nearly every other vendor will give up the interface. Ampersand's integration layer and our scorecard describe the same future: the agent is the interface, and the vendor keeps workflow and data. We differ on the economics: a vendor keeps more revenue where the work is unbounded and it holds its workflows, and less where the work is bounded or it is left with only its data.

@StableBread Oct 7

Fajasy on X

Read the deep dive on $BWXT here: https://t.co/ogaS6FmtMk I cover how BWXT became the Navy’s reactor supplier, where its growth comes from, what military base power and data centers are worth to BWXT today, and the management, insider selling, and financials behind the plan. Then I conclude with what $145.70/share already assumes and whether I’m buying.

@StableBread Oct 7

Fajasy on X

The Pentagon wants every major U.S. military installation to generate its own power. In August 2026, the Army chose five companies for Janus, a program expected to put more than 20 commercial microreactors on military bases. One of the five already builds the reactors for the Navy’s submarines and aircraft carriers, and its stock is down almost 40%. That company is BWX Technologies $BWXT, and even after a 7.6% jump on October 6, the stock is down 38.9% from its $238.42/share peak on April 15. $BWXT didn’t fall alone: → The VanEck Uranium and Nuclear ETF $NLR is down 33.4% from its January high, even after rising 5.5% on October 6. → Curtiss-Wright $CW, which supplies pumps and valves for the same submarines and carriers, is down 30.2% from its July high. So $BWXT got caught in a broader nuclear and defense selloff, but it fell further than both $NLR and $CW after guiding to slower growth in 2027. Meanwhile, $BWXT’s business got stronger: → Guidance up twice: 2026 adjusted EBITDA guidance rose from $645-660M in February to $662-672M in August. → An $8.4B backlog, more than double the $4.0B $BWXT had at the end of 2023. → CEO Rex Geveden calls $BWXT “the sole provider for naval reactors.” The Navy’s 30-year shipbuilding plan, released in May, calls for a new carrier every four years instead of every five. → The Army chose $BWXT’s 20-megawatt reactor for Janus at Fort Campbell, Kentucky, on August 26. On October 5, Prodigy Clean Energy, a Montreal-based developer of factory-built nuclear plants, chose the same reactor for a transportable power plant in New Brunswick. → 2030 targets: $BWXT targeted $5.5-6.0B of revenue and $1.1-1.2B of adjusted EBITDA in 2030. It’s on track to meet or beat all three targets it set in February 2024. Still, here’s what could hold the stock back: → $BWXT expects growth to slow in 2027. Its preliminary framework calls for high-single-digit growth in both revenue and adjusted EBITDA. → Even after the 38.9% drop, $BWXT isn’t cheap in absolute terms. It trades at 20.9x NTM EV/EBITDA and 29.3x earnings. → By CFO Mike Fitzgerald’s count, only a little over half of the revenue growth to 2030 comes from what $BWXT calls its “predictable core,” the programs backed by backlog and long-term contracts. The rest depends on new orders and acquisitions. → $BWXT’s Janus release discloses no dollar figure. Owning reactors is a new business model for $BWXT, with capital commitments it hasn’t disclosed. → No $BWXT insider has bought stock in the open market since at least January 2025, including during the drop from $238 to $135. → The government is running on a continuing resolution through December 11. At $145.70/share, BWXT has a market cap of $13.35B and an EV of $15.0B ($13.35B + $2.05B of debt - $0.40B of cash). It trades at 22.5x 2026 adjusted EBITDA guidance and 30.7x non-GAAP EPS guidance. In return, you get an $8.4B backlog, more than double the $4.0B BWXT had at the end of 2023. And on NTM estimates, its EV/EBITDA has dropped to 20.9x from ~35x in March, back to its 2024 levels. So is the Navy’s reactor supplier on sale, or does the market see a slowdown the numbers don’t show yet? That's what I answer in my new ~6,800 word deep dive on $BWXT! Credit to @GrumpierBTDay for the idea.

@StableBread Oct 7

Fajasy on X

Read skeptically, Nubeva Technologies $NBVAF (TSXV: $NBVA) is a shrinking software company living off its cash, with an AI decryptor builder that keeps slipping and a license customer that may not have renewed. Read another way, Nubeva is being prepared for a buyer. Here’s what supports the second read: → In March 2024, Netskope $NTSK, now a ~$7.6B company, bought Nubeva’s TLS (web-traffic decryption) technology and five patents for $1M. Greig Bannister, the CTO who co-invented the key-capture patents, followed the technology to Netskope. → In an August 2024 video, CEO Randy Chou said “partnering and M&A groups” from companies worth $20-30B had reached out after the July 2024 CrowdStrike $CRWD outage. → Chou wants all four AI patents granted before any partnership or sale, and two are still pending. → The FY2026 MD&A says Nubeva is “selectively engaging with lighthouse accounts,” meaning “large, high-profile organizations that serve as validation points for the technology,” to make its offering “more attractive for either future expansion or strategic partnership/M&A.” In a June 2025 video, Chou said lighthouse accounts “matter 100%” in a partnership or M&A talk. → Sales and marketing spend was $14,951 for all of FY2026, with $2.49M of cash on hand. In the same June 2025 video, Chou said Nubeva is “really not marketing” until “the patents and everything else comes in.” → The most recent investor deck is from July 2023, and since then Nubeva has communicated mostly through filings, press releases, and the occasional YouTube interview. In other words, it isn’t pitching investors either, which fits a company waiting on a buyer. A longtime investor I spoke with described Chou as reluctant to spend money, PR included, and called a sale to a larger security company the clear end goal. I’d add that Chou, who makes the call, also owns a quarter of the company.

@stockgutter Oct 7

SG on X

need more $GLD $GC https://t.co/sC4KNVanQw

@convequity Oct 7

Convequity on X

$SITM supplies timing devices used in optical modules in NVIDIA-based AI data centers. In an AI data center, a clock exists to keep parts synchronized. Training and inference run across thousands of chips in parallel, and each step depends on data arriving at the right moment. When timing slips, GPUs wait, networks stall, and expensive compute goes unused. The move from 800G to 1.6T optical modules shows how this becomes demand for timing parts. Bandwidth does not simply double. The link is split into more lanes running at higher baud rates. Each lane has a tighter jitter budget, and it often adds clock domains for the DSPs, SerDes groups and laser drivers. So each module carries more timing components, and each component has to meet a higher performance requirement. That pushes average selling prices up alongside unit volumes. The same pattern appears in the switches inside that cluster. A high-radix switch multiplies SerDes lanes as bandwidth scales. When the switch increases radix and speed at the same time, it drives more clock domains across the fabric, the ports, and the control logic. Timing content in the switch therefore multiplies rather than rising in step with the size of the cluster. This is why we see continued hypergrowth for SiTime. In NVIDIA-based data centers, SiTime's parts are discrete devices, separate from the GPU and switch ASICs, allowing timing performance to improve independently as system requirements change, and demand for them rises in several places in the rack at once.