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.

From X

@StableBread

Fajasy on X

The bitcoin miner turned AI data center developer, $DGXX, is at $2.13/share, down 46.3% since my August 16 update. Since then, it got a $200M debt commitment, syndicated by $GS, to finish its Alabama AI data center, and said Phase 1 (December) and Phase 2 (March 2027) remain on target. That’s good news if the loan closes (expected within 90 days), since I said in August it would decide whether $DGXX can fund the rest of the build with debt or has to keep selling stock. CEO Michel Amar told Frank Curzio of Curzio Research in a November 2 interview that debt financing “seems to be a little bit scarce these days” and has slowed $DGXX down. Regardless, he said all the equipment for both phases is already bought, with the chillers installed. With ~$135M of cash as of October 13 and no debt at June 30, $DGXX’s EV is ~$81M. Amar also said he wants $DGXX at $500M+ of yearly revenue within 36 months, with its leased data center business earning ~60% EBITDA margins and ~25% cash flow margins after debt service. As of August, only ~$140M of the $250-300M of ARR it targets by Q3 2027 was contracted. $DGXX has used $174.5M of its $175M ATM program, and its share count has been flat at 101.4M since August 14. So if the loan falls through, selling more stock would take a new filing. The October 13 release also says $DGXX sold no ATM shares in Q3, though its Q2 10-Q shows it sold 2.85M after June 30 at an average of $3.95, well below the stock’s $5.45 close on June 30. So I’d track dilution through the share count in the Q3 10-Q, not management’s comments. Its November 2 proxy also asks shareholders to expand its stock award pools to 20.3M shares from 16.3M (20% of shares outstanding), with 11.9M free for new grants, after Amar and his son, President Alec Amar, were awarded a combined $19.5M in 2025 (~9% of today’s market cap), mostly in stock and options. That’s more dilution than I’d like after the share count rose 46% this year, to 101.4M from 69.4M at the end of 2025, and the proxy would also let the board raise the RSU pool 10%/year from 2028 without another shareholder vote. Both Amars have bought $166K of stock in the open market since August, most recently 20,000 shares at $2.76 by Michel Amar on October 14, but that’s small next to the $19.5M they were awarded in 2025. Q3 results come out November 13.
@MaiusPartners

Maius on X

Compute and infrastructure expenses are expected to reach ~$13.0bn in Q4 as the $SPCX contract ramps. Meanwhile, revenue is expected to reach $18.6–20.3bn in Q4, up ~10–20% QoQ, with the competitive environment and holiday seasonality as potential headwinds. That puts revenue per compute dollar at ~1.5x. Meanwhile, roughly $80/$120bn of payments are against disclosed commitments in 2027/28 that help underwrite the entire global AI trade. The harder question is how quickly and profitably can API usage grow next year? Coding has been a clear winner. But sustaining growth requires either deeper usage within the industry and/or broader deployment across other high knowledge-intensive industries (finance, healthcare, legal). For the former, a ceiling is emerging, especially when sophisticated clients choose to train their own models to avoid the closed-source tax (e.g., Palantir). For the latter, those opportunities are large but much harder to capture, as turning them into reliable production systems requires data integration, domain-specific knowledge, corporate and regulatory approvals, and sometimes an entire organizational workflow redesign. Anthropic’s own $100m program to train 10,000 deployment engineers illustrates the work still required. However, wherever you look, the most sophisticated industry players choose to train their own models, e.g., Harvey AI using Kimi. Given this dynamic, it's not hard to paint a scenario where the“revenue per dollar of compute” metric continues to decline further in 2027 as industry capacity ramps, and hence one can see why the market is becoming extremely skeptical of the entire AI trade, given the labs are essentially underwriting the entire semiconductor industry at this stage. Where am I underestimating the upside: usage intensity (AGI? Humanoid Robots?), deployment speed in legacy industries? Or the entire cost curve?

Stevanato down 30%. Yikes.

Stevanato fell 30% after forecasting low-single-digit 2027 growth as lower Nexa syringe volumes from a GLP-1 customer offset solid Q3…

BSP · short

Bending Spoons Spa

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

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

266 firms

Sectors

Ordered by how often our editor reaches for them.