데이터센터 연결: 차세대 AI 알파가 숨어 있는 곳

Datacenter Connectivity: Where the Next AI Alpha Hides THE DAILY --- H.E.A.T. Torching Wall Street’s Obsolete Playbook. Delivered straight to your inbox 5 days a week. Subscribe Home Posts Datacenter Connectivity: Where the Next AI Alpha Hides Datacenter Connectivity: Where the Next AI Alpha Hides Oct 2, 2025 Wall Street’s 60/40 formula was born in 1952 — the same year as the first credit card. A lot has changed since. That’s why we created a new approach — The --- H.E.A.T. Formula — to empower investors to spot opportunities, think independently, make smarter (often contrarian) moves, and build real wealth. — # (# ) Table of Contents 🔥 Here’s What’s Happening Now 🧠 Datacenter Connectivity: Where the Next AI Alph … What’s structurally new (and investable) The basket (barbelled and pair-able) Core longs (structural) Selective caution / pairs How this plugs into your JUCE/power thesis Trade setups we’d actually put on KPIs & tells (what we watch weekly) Risk map (what breaks the thesis) 📈 Stock Corner 📬 --- In Case You Missed It 🤝 Before You Go Some Ways I Can Help 🔥 Here’s What’s Happening Now We removed ARGT from our ETF model portfolio and replaced it with KOID. Talked about this yesterday, we can go back into ARGT if it recovers but not holding something going down in a straight line. They tried to sell this market off yesterday on the government shut down. I told you yesterday that I don’t believe this should impact the market’s intermediate term and that any weakness was a buying opportunity. Obviously, everyone on Wall Street reads my newsletter as they all bought the dip. Crypto went parabolic yesterday, Bitcoin looks like it now wants to re test the highs…. Ethereum needs to break above the 50 day… Yesterday’s newsletter asked the question of whether AI is in a bubble. The answer is we won’t know for sure until after the fact. I asked Ron Insana his thoughts on that, his interview is posted below. Short answer, he doesn’t think so. Came across this chart, which also helps put things in some perspective.. And this….. So maybe today’s valuations are the new normal? We will see. For now, trade based on what you see, and what you see is a market in a clear uptrend with dips that get bought pretty much immediately. Always have hedges in place just in case. 🧠 Datacenter Connectivity: Where the Next AI Alpha Hides TD Cowen recently came out with a report on data center connectivity, today we take a deep dive…. The networking silicon TAM is set to > $75B by 2030 with >20% CAGR from 2025–2030, driven by switching and new “scale-up” fabrics. NVIDIA is still positioned to capture outsized value across acceleration and networking. (See the growth bars and category split on p.11.) Active copper (AEC/ACC) gets a real runway inside and across racks; linear pluggables (LPO/LRO) keep CPO at bay near-term; Ethernet likely wins scale-up hardware vs PCIe/UALink—creating both winners and losers. (p.5, p.13– • 14) Stock stance from the report: Top pick = $CRDO ; $MTSI = quality SMID growth; $SMTC = highest risk/reward; $ALAB (Astera) = Hold given competitive threat if Ethernet takes scale-up share. (p.4–5, p. • 15) What’s structurally new (and investable) Three networks now matter “Front-end” (classic CPU/networking), back-end (GPU-to-GPU for training), and scale-up (rack-scale, shared memory for inference). The step-function growth comes from back-end + scale-up . (The evolution timeline on p.3 lays this out.) TAM math that actually moves stocks Layer-1 (cables/optics/retimers) and Layer-2/3 (Ethernet/InfiniBand/scale-up switching) are both growing, but switching is the real dollar engine , with scale-up switch ASICs going from near-zero to a meaningful slice by 2030 (p.11, p.16 tables). The debates that decide winners/losers Copper lasts longer than consensus expects thanks to AEC/ACC signal conditioning (bullish content for AEC/ACC chipsets). Pluggables (LPO/LRO) extend life and keep a multi-vendor ecosystem; CPO adoption slower than hype near-term. Ethernet likely prevails in open scale-up hardware , a looming threat to $ALAB if Amazon and others pivot off PCIe, while $AVGO (merchant Ethernet switching king) stands to benefit. (p.5, p.13– • 14) The basket (barbelled and pair-able) Core longs (structural) $CRDO – “Best SMID idea” per report; AEC leadership + LRO optical DSP share-gain optionality. If AEC SAM hits ~$2B by 2030 and LRO bites into incumbents, multiple support holds. (p.4– • 5) $MTSI – Analog inside optics + CW lasers > SiPho wave; diversified (Defense/Telco/Datacenter). Underappreciated AI-optics lever with quality mgmt. (p.4, p.14– • 15) $SMTC – Re-focused, de-levered; ACC content a greenfield tailwind. “Highest risk/highest reward” but real torque if ACC ramps and legacy margins normalize. (p.4–5, p. • 15) $AVGO – Merchant Ethernet switching scale and SiPho content potential; if Ethernet wins scale-up, Broadcom’s incumbency is the fulcrum. (p. • 14) $NVDA – Remains “top dog” in accelerators and takes a slice of networking economics; the model shows NVDA still capturing a big share of system value (p.11). Use as core plus sell calls on rips. Selective caution / pairs $ALAB – Hold in the report; key risk is Ethernet eating scale-up share at hyperscalers (Amazon pivot cited). Pair long $AVGO / $CRDO vs. $ALAB into scale-up newsflow. (p.5, p.14– • 15) InfiniBand proxies – The model shows Ethernet + scale-up growing while InfiniBand normalizes/declines later in forecast; fade IB-only narratives vs. Ethernet beneficiaries (see the Layer-2/3 split on p.11). How this plugs into your JUCE/power thesis The report reinforces: bottlenecks migrate from compute to network to power—and back again. As copper/optics/switching scale, time-to-power stays a gating KPI. Keep owning grid enablers while you surf connectivity winners. (Use our prior JUCE framework; this doc covers the networking leg of the stool.) Trade setups we’d actually put on Long $CRDO / Short $ALAB (equal-risk) Catalyst path: Amazon/others lean Ethernet in scale-up; LRO wins sockets; AEC TAM revisions up. Use call spreads on $CRDO to cap premium outlay. (p.5, p. • 14) Long $MTSI + calls into SiPho/CW-laser attach updates; sell OTM puts to finance (quality SMID angle). (p.14– • 15) Core $AVGO with covered calls; overlay event-driven 0DTE into major DC-capex prints (NVDA GTC, hyperscaler earnings) for asymmetry. (p. • 14) Tactical $SMTC on weakness**:** buy the dips; exit on ACC confirmation beats. (p.5, p. • 15) KPIs & tells (what we watch weekly) Port mix by speed (800G → 1.6T shipments), in-rack vs. rack-to-rack copper penetration, AEC/ACC unit growth. (p.5, p.13, p. • 16) Pluggable (LPO/LRO) share vs. CPO announcements and real deployments (not just demos). (p.5, p. • 13) Scale-up hardware choices at AMZN/MSFT/GOOGL/META—any PCIe → Ethernet shifts; early UALink transports over Ethernet PHY . (p.5, p. • 14) Vendor breadth at hyperscalers (single-vendor risk down = better for pluggables and merchant silicon economics). (p.13– • 14) NVIDIA networking revenue vs. accelerator growth (does NVDA keep taking the “tax” on the network?). (p. • 11) Risk map (what breaks the thesis) Faster-than-expected CPO adoption compresses the pluggable window (bearish LPO/LRO suppliers). (p. • 13) Scale-up stagnates or ASIC consolidation favors captive stacks, starving merchants. (p.11, p. • 14) Copper physics wall sooner than modeled; optical pulls forward in-rack. (p.5, p. • 13) AI’s Hidden Bottleneck: The Network. TD Cowen pegs networking silicon > $75B by 2030 with switching + new scale-up fabrics doing the heavy lifting. The near-term surprise: active copper and linear pluggables keep optics “modular,” delaying CPO—and Ethernet looks set to win scale-up hardware. That mix favors $CRDO, $MTSI, $AVGO (and selectively $SMTC ) while we stay cautious on $ALAB if hyperscalers pivot off PCIe. Translation: own the pipes that make AI usable, not just the GPUs. (Source: TD Cowen, see charts pp.5, 11, 13–15.) Barron's 📈 Stock Corner Today’s stock is Microsoft (MSFT)….. Not only does this chart pattern look like it’s setting up for a breakout, this news looks real interesting….. Microsoft Folds AI Service Into Office in Bid to Take on ChatGPT Microsoft Corp. is folding its artificial intelligence subscription service for consumers into Office, betting that the productivity suite’s ubiquity will help the company better compete against OpenAI’s ChatGPT. www.bloomberg.com/news/articles/2025-10-01/microsoft-folds-ai-service-into-office-in-bid-to-take-on-chatgpt?srnd=homepage-americas One of the exercises we go through in trying to figure out the big winners from AI is who has the ability to integrate it into something that everyone uses and therefore profit from it the most. That’s why for a while I was talking about AAPL potentially being a big winner because I imagined the impact of AI fully integrated in to my iPhone. AAPL has had some issues executing. AI fully tied into Office looks real interesting, I’d pay more for that. 📬 --- In Case You Missed It Ron Insana joins the show today to talk NVDA, AI, DATs, and much more…. 🤝 Before You Go Some Ways I Can Help ETFs : The Antidote to Wall Street Inside HEAT: Our Monthly Live Call on What Wall Street Doesn’t Want You To Know Financial HEAT Podcast https://www.youtube.com/@TuttleCap Freedom from the Wall Street Hypocrisy Tuttle Wealth Management: Your Wealth Unleashed Advanced HEAT Insights: Matt’s Inner Circle, Your Financial Edge All rights reserved. Keep Reading No posts found View more caret-right 0